are typically your employees that . Often delivering high profile, last minute stuff that is perceived as hugely valuable, they end up as the go to person for the higher ups. The downside is that when High Performers go bad and turn into Rogue High Performers, the effects on your business and your staff (Lesser Mortals in the eyes of the Rogue High Performer) can be dramatic. High Performers get stuff done The “Nobody Else is Good Enough” Bottleneck A far as Rogue High Performers are concerned, they are the rockstars and have never needed support from anyone. In their minds their entire career has been one success after another. Any problems encountered along the way have been down to the Lesser Mortals not understanding the vision and making mistakes. Eventually this translates to an attitude that nobody else is worthy to work with them, so everything must be done by them. Suddenly they are on the critical path for every high profile project, and all of those projects start to suffer. There is One True Way Rogue High Performers know the way to do things. This may have been learned many years ago using tools and languages that have long been superseded, but none of that matters. Their way is the right way and anything else is wrong. They don’t need to look at the latest industry trends or keep up to date with new approaches, as they got it right first time and their way is future proofed by virtue of it’s shining rightness. right Lesser Mortals have a hard time when the One True Way has been adopted, as it is never communicated in advance. Instead, completed tasks are rejected because they don’t follow the righteous path. A Rogue High Performer will never take the time to figure out if the different approach is better, or even valid. It is wrong because it is not the right way and must be done again. Multiple times if necessary. They Are Untouchable A Rogue High Performer will often be immune from the consequences of their actions, leaving Lesser Mortals to take the blame. Sometimes this will be because of their stellar reputation, which leads higher ups to believe that any mistakes must be the fault of Lesser Mortals (compounded by the fact that Nobody Else is Good Enough). Sometimes it is because the higher ups only talk to the High Performers and therefore only ever hear that it’s down to the Lesser Mortals regardless of the facts. Even when it is recognised that the High Performer may have had a hand in the problems, this will be rationalised and minimised: “They are under a lot of pressure” (translation: because Nobody Else is Good Enough they have taken on too much and are now doing many, many tasks badly) “None of this could be foreseen” (translation: none of this is predicted by the One True Way, so how could it have been anticipated?) “They worked night and day to save the situation — they are a hero” (translation: as they overcame the problem through sheer weight of hours, we’ll ignore the fact that they created it in the first place!) They can Break the Rules Rogue High Performers have exacting standards that must be met — e.g. no code goes out without 100% unit test coverage and no design can be signed off until they have been through it with a fine-tooth comb. But these standards turn out to be surprisingly flexible when applied to the Rogue High Performer — they understand the impact of breaking the rules so can do so as they see fit. Their code can have minimal unit test coverage because it’s been written by them and . Their designs can go out without review because they get things right first time, and who among the Lesser Mortals is qualified to review it anyway? If there are problems, they’ll apply the sheer weight of hours and save the company once again, justifying their rock star status. just works They are Difficult to Work With Rogue High Performers struggle to connect with Lesser Mortals, which makes working relationships difficult. As Nobody Else is Good Enough, they tend to talk down to Lesser Mortals, who resent being patronised. As there is One True Way, nobody else’s ideas are evaluated or even welcome, and lots of rework is required as Lesser Mortals don’t understand the One True Way. Rogue High Performers will often take a dislike to a Lesser Mortal if they have the temerity to question anything, and when you are , you can behave how you like, because you get . No need to treat people courteously or even professionally. this good stuff done Lesser Mortals cannot manage Rogue High Performers. The Rogue High Performer knows that what they are doing is right, so they don’t need to take direction or orders from someone who has no idea what they are doing. This will be seen as a management failing, the Lesser Mortal should somehow have asserted control over a Rogue High Performer, even though the higher ups refused to back any of their attempts to do so, for fear of upsetting the status quo. As Nobody Else is Good Enough, Rogue High Performers are often late looking at the work done by Lesser Mortals, which causes go-live dates to be shifted at short notice to accommodate the rework. This isn’t great for customer, but reinforces the status of the Rogue High Performer — another scenario where if they don’t do it, it isn’t done right, and luckily they were able to save it at the last moment. An easy way to spot a Rogue High Performer is to canvas opinion on those that have recently worked with them — if the prevailing attitude is to avoid working with them in the future, congratulations — you landed one! They Cost you Money Not just in terms of rework and delay, although this has a very real impact on your bottom line. Rogue High Performers cause churn — especially where there is intense competition for talent, such as in Salesforce world. Lesser Mortals decide things will be better elsewhere and move on. This is typically rationalised away by the higher ups — the company is better off without them (which begs the question why did you hire them in the first place, and put them on key projects with Rogue High Performers?). Churn costs — think about how long it took to identify, interview and employ that Lesser Mortal. How much time they spent getting themselves up to speed and how much time other Lesser Mortals spent supporting them (the Rogue High Performer typically can’t spare time to help here — they are too busy on critical tasks that are holding up multiple projects). When you sacrifice Lesser Mortals to preserve the ego of the Rogue High Performer you have to do it all again, every time. They don’t Change High Performers don’t go Rogue overnight. It tends to be a gradual process. They insert themselves into the critical paths of projects one at a time, and their behaviour slowly gets worse. If this isn’t nipped in the bud then by the time everyone realises what has been happening, it has been normalised. A Rogue High Performer won’t moderate their behaviour or change their approach going forward, as they are . If a few Lesser Mortals are the cost of this, then so be it. Higher ups will often buy into this — there are a few bumps in the road, but look at how much stuff has got done. Not rocking the boat becomes the default option which allows the situation to escalate. getting important stuff done It Defines your Culture Suddenly you have a toxic —Rogue High Performers are now running the show and morale is plummeting. The higher ups don’t know what to do — they’ve tried nothing and are out of ideas. Promises are made to regain control, but the genie is out of the bottle. The Rogue High Performer sees nothing wrong with how they work, so will always revert to that modus operandi. culture Eventually Rogue High Performers stop even paying lip service to the higher ups, as they know there will be no action taken. They are that they company can’t possibly run without them. The higher ups will often believe this too, or at least prefer believing it to taking difficult decisions. getting so much stuff done Stop it Before it Picks up Momentum Constant vigilance is needed to deal with Rogue High Performers. Listen to Lesser Mortals when they report issues and don’t assume they are the problem because Spot the signs early and shut them down — in the long term you’ll be glad you did. stuff is getting done. I’m better known in the Salesforce community as Bob Buzzard — Umpteen Certifications, including Technical Architect, 5 x MVP and CTO of BrightGen, a Platinum Cloud Alliance Partner in the United Kingdom who are hiring (Rogue High Performers need not apply!). _At BrightGen, we do more than manage Salesforce implementation projects for our customers. We play a pivotal, strategic…_www.brightgen.com BrightGen Careers You can find my (usually) more technical thoughts at the Bob Buzzard Blog is how hackers start their afternoons. We’re a part of the family. We are now and happy to opportunities. Hacker Noon @AMI accepting submissions discuss advertising & sponsorship To learn more, , , or simply, read our about page like/message us on Facebook tweet/DM @HackerNoon. If you enjoyed this story, we recommend reading our and . Until next time, don’t take the realities of the world for granted! latest tech stories trending tech stories

Stellar

Facebook

Salesforce

How Rogue High Performers Hurt Your Culture

The current speed of technological progress and a degree of connectivity promote the cooperation paradigm as the best strategy. What does it mean to be an entrepreneur? It means you are going to the place where nobody has been before. It means to create an idea that brings wealth for everyone. The ideas have to create cooperation via win-win situations for funders, investors, employees, partners and clients. It should bring long-term benefits for all. Lets look at these statements from a mathematician point of view. Prisoner’s Dilemma Consider “ ” example from Douglas Hofstadter: closed bag exchange Two people meet and exchange closed bags, with the understanding that one of them contains money, and the other contains a purchase. Either player can choose to honor the deal by putting into his or her bag what he or she agreed, or he or she can defect by handing over an empty bag. They both don’t know and don’t trust each other. And if they are rational players a defect is always the best choice. To make a deal in a cooperative style they have to bear additional costs. That dilemma brings to existence mediators who can solve the problem for 10% of the deal. Imagine a deal over the Internet. It’s obvious that you need Amazon, Ebay, Alibaba, your Bank and other mediators. But after some time such mediator may become monopolistic. And such monopoly, after all, suffers from the — if it wants to cooperate with own customers or for instance raise a rate up to 15% that will increase profit by betraying all the customers. Such monopoly becomes a source of power. And we need another mediator to manage this situation. same dilemma Bitcoins on the stage In November 2008, a paper was posted on the Internet under the name Satoshi Nakamoto titled . That was the beginning of Bitcoin. It happened in a month after Lehman Brothers, the fourth-largest investment bank in the United States that was operational for 158 years from its founding in 1850 filed for bankruptcy at the peak of the crisis of 2008. bitcoin: A Peer-to-Peer Electronic Cash System Bitcoin did exactly opposite to what banks had been doing for hundreds of years. It has showed that we don’t need a hierarchical organization with own will to manage money for us. Blockchain is technology behind Bitcoin The beauty of Blockchain is that monopoly is not an option for this technology and the power is melted among all users. And everyone is equal. Blockchain is the example of a distributed system where there is no place for hierarchy and a control isn’t concentrated in one hand. The idea is absolutely perfect because it brings equal benefits for all in the same fraction as for you. Meanwhile, a competition means creating a monopoly on some wealth resource, what is selfish and destructive in a longer term. The more projects like blockchain we support the bigger cumulative wealth is coming back to each of us.Using Blockchain-like approach there is no place for monopolies. Some of you follow this idea on everyday basis, some don’t. But there is one more step ahead — you may teach others to use this knowledge. “Ideas create tribes”. And the bigger is a tribe that shares the idea of a cooperation, the better world we will live in. AI Safety There is an to create AI with human-level capabilities and beyond. That is why top AI researchers are talking about the risk of concentration AI in one hand. AI is ultimately too powerful for this option. And we have already got a response for this situation. For instance, opinion that is possible in near future tech giants pledge $1bn for ‘altruistic AI’ venture, OpenAI. is a non-profit artificial intelligence research company. Our mission is to build safe AI, and . OpenAI ensure AI’s benefits are as widely and evenly distributed as possible So the problem is real. Conclusion What is the best strategy to grow your company — help your client to solve problems. What is a great strategy to live with — help people around you. Think cooperative. Help others. This idea is also the basis for all religions. Treat others as you want to be treated by them. You may agree with it or may not. We live in a huge non-zero sum cooperative game, that’s why cooperation is a key for success. What I want is to emphasize that it is not just a good advice but a proved mechanism by the theory of games. If you want to be incrementally better: Be competitive. If you want to be exponentially better: Be cooperative. — Author Unknown My explanation for the quote is that the progress is fast enough nowadays and if you make your project open source today, tomorrow someone can create an amazing technology using it, and the day after tomorrow you’ll wake up in the world where everything is better for you. Elon Musk gave away all the patents I suppose he had the same ideas. When Have my words made you more cooperative? Call to Action: Pr_ess the heart_ or share the article if you found this article interesting Follow me on Medium or Twitter so you see my upcoming articles in your feed. Answer the question: What is your interest in AI?

Amazon

eBay

Tesla

Twitter

YouTube

If You Want to Be Exponentially Better: Be Cooperative.

Thinking about establishing a local community around a specific topic you care about? This 9 step guide gets you going. Meetups are a great way to take your online network to a more tangible real world scenario and gather this likeminded community in order to create great things. Organizing a meetup normally goes hand in hand with some form of planning via group chat or email — both equally painful. Which is we are going to outline how a bunch of our users work with to set up their meetups in a very efficient way. JayPads Planning in a group chat (yes, any group chat) is the gateway to chat hell and you know it! 1 First things first: What do you want to accomplish? Discuss what your goals are — be as specific as you can be. Brainstorm. This is a crucial step in order to target the right people. In a JayPad you’d use the group chat in order to get ideas flowing and once you have settled on a topic you write a note in the “Pad” (the permanent content area of JayPads, described in more detail ) and store this information easily accessible, think of it as a Post-It for important stuff worked out in the chat. If you choose a format of “keynote plus Q&A”: What is your keynote going to be? Who is delivering it? Who manages the Q&A? For a small meetup we’d suggest: The first keynote comes directly from you and management of questions is likely not going to be an issue — just go with the flow. here What do you want to accomplish? How are you going to go about it? 2 Research local meetups. Browse already established communities and see if your goals complement each other (browse at least both and Facebook). You will later use these profiles as inspiration for your own: These guys have already pulled off what you are about to try, so you better study them. Learn how they approach their members, how they advertise their meetups and where. This might carry the necessity to re-calibrate what you have done at step 1. Meetup Make a quick research of different meetups and try to find best practices. 3 Find a date. This is what JayPads were originally built for and this is one of its great strengths: Get people together at a certain time and location. Just poll the availabilities of the team. What we discovered is that business-related meetups work best on Wednesdays or Thursdays. Poll the date which works best for you. 4 Setup a meetup at — they are very great at offering your event to people which might be interested in coming. The community is very active and responsive and people browse suggestions regularly. This is your go to move to get this thing rolling. Meetup Setting up your meetup on Meetup is a thing of minutes. Inputs required in the first step are: . Which is pretty clear, since you want to build a local community. Location . Picking the right keywords is important for Meetup to know which people you are trying to gather. Have a look at your competition note: How do those guys tag their meetups. Meetup itself offers good suggestions and gives indicators how many groups are targeting this topic. Tags for your meetup . This should be easy — but a look at your competition will help you get a better feeling for “good” vs. “no-so-good” names. Name Describe your goals and why someone should join (50 characters minimum). Again, be specific. A more lengthy description. Brainstorm point 1. to 4. using the group chat and store the results in another note — this is essential. Just do it. Discuss in the group chat and store the results in a note. If you are working on step 2. and 3. with more than just two people, consider polling alternatives and come to a democratic result. This will save tons of time and JayPads offer a simple feature for this step. 5Make a Facebook event. Yes, Meetup is an awesome tool — but Facebook offers great ways to build your community as well. This is why you will need to setup this thing as a Facebook event additionally. There is no way around it. Its even easier than setting up your Meetup profile and information required has already been established at step 4. Just paste everything together and you are golden. Setting up a Facebook event does not take long — if you like: You can build a nice logo for your meetup. 6 Get initial members. You have a network already and this is the reason why you think there should be your superawesome meetup. Now is the time to invite them. Just send them the link to your Meetup profile via Email or invite them through Facebook (this is where the Facebook event comes in very handy). Aim for at least 10 fix commitments. We think this should be the baseline in order to have a productive meeting of people which are not all familiar to each other, thus creating a benefit for those who came and making it more likely for them to a) return and b) bring likeminded people from their network you don’t know yet. Keep track of commitments to your meetup. 7 Find and prepare a venue for your meetup. Once you have a better feel for the size of the crowd you are going to assemble you can look for locations. You should have a room for 10 ready at all times but you should also develop fallback plans in case this thing somehow blows up and gets viral. Be prepared. Set up responsibilities: make signs if the location is not straightforward to find, get and check the required equipment, check Wifi and put out signs with access codes if required. Store info and contacts in another note. Prepare a note with options for your locations, contacts included, and make sure you can call them on short notice. 8Prepare for a small get together afterwards. This is a crucial step. You absolutely need to take care of beer. As silly as this might sound to some: Do not underestimate the power of a few drinks with likeminded people. We are dead serious about this point and you should be too! Reach out to potential sponsors, advertising through “this fun time is brought to you by…” is a powerful marketing channel you can offer. And for the love of god, bring business cards — they still use them. Yes. This is vital for your meetup — just do it. 9 Keep the gang together. Follow each other on Twitter, LinkedIn and other resources — become Facebook friends. If you want to keep the brainstorming going and foster community and engagement, we have made great experiences with specific meetup JayPads (even if it is just so everybody has a central directory to share the links to his social accounts in). They are platform agnostic and you can merge users from Facebook with those acquired through Meetup or other channels. Working on ideas, storing and sharing links to interesting articles as well as files and pictures (file upload is a JayPad feature which is to be released very soon, we are currently testing it in Alpha) is convenient and we are not aware of a tool requiring less effort for these types of tasks. Keep the gang together. We built a tool we’d like to call — which is misleading since JayPads are a much more general communication framework. But what we discovered and validated to great lengths is: JayPads are extremely strong when a smaller group of people wants to get something off the ground. And this strength can be utilized much longer than the common Slack-enthusiast might think. If you are interested in further elaboration of this point: and might be for you. “the smallest side project management tool in the world” this this JayPads are extremely strong when a smaller group of people wants to get something off the ground. And this strength can be utilized much longer than the common Slack-enthusiast might think. Just give it a go and set up your next meetup, get together with friends, side project etc. via your very first . JayPad If you liked this article — please make sure to give us a follow here on Medium as well as Twitter and Facebook . We will publish articles on other use case scenarios as well as our general take on collaboration software and the journey of a young startup.

Flow

BUNCH

Community Is

Slack

Target

9 Steps to Plan and Set Up a Meetup Using JayPads

In the past year, I’ve become convinced that machine is not hype. Strong AI/AGI is no longer a requirement for complex tasks. It doesn’t matter that is out of reach, since we don’t need it in order for automation to take over vast swathes of the job market. I now think that domain specific ML is going to take 10% to 50% of all jobs in the next few decades. The obvious ones are transportation and logistics. But almost every industry is going to be transformed. I have kids that are 1, 3, and 5. And I have no idea what will even exist when they enter the job market in 20+ years. learning AGI My Approach The Stanford Machine Learning Class I’m 2 weeks in to Andrew Ng’s famous on Coursera. It just started last week, so if you hurry you can probably still take it this semester. This class is the one thing I’ve seen everyone involved in ML recommend. I want to learn first principles, so that I can go beyond tutorials and understand how to effectively use tools like and . Machine Learning class Tensor Flow Keras The Applications of AI — Email Newsletter Rob May’s email newsletter has a Light Tech / Business Focus. It comes once a week at 7 AM central on Sunday mornings. He covers the one big idea from the week on topics like “whether or not the DOJ should breakup Google and Facebook over data-monopoly or AI talent-monopoly issues, and whether you will eventually be able to buy insurance against algorithmic bias.” Then he gives summaries of the 5 best articles from the week and gives another 5 that are worth a read. Technically Sentient Podcasts I’m also supplementing with some podcasts. There’s the by OCDevel that gives you an introduction then gives audio commentary on each lesson covered in Andrew Ng’s class. He also gives a number of suggestions on how to break into a job in Machine Learning. Machine Learning Guide I also recommend listening to — This Week In Machine Learning. I listened to a long-winded who is VPE at Quora and ran the ML group at Netflix. And every now and then the great Software Engineering Daily podcast has someone from the world of ML on the show. I enjoyed the “ ” podcast with Zayd Enam from the Stanford AI lab. He talks about how ML is next to impossible to debug. TWiML one with Xavier Amatriain Machine Learning is Hard As for videos and slides, I found some that let me understand how neural networks work but I’m holding back on them until I finish the class. great slides The Class I’ve been working on the class at night, sneaking in a video here or there during my lunch break. I’m going to talk about the class a good bit in the next few weeks and am going to pass along the high level concepts (no linear algebra or advanced math here) of what I’m learning. So here is what I’ve learned in the first two weeks: Brute Force Works We can throw hardware at large data sets and come up with very good predictions. You don’t have to be an expert at math I was really afraid about these requirements (I never took linear algebra), but ended up picking up the high level concepts quickly. The math is intimidating, but once you understand it at a high level, its just importing the data and running a few functions in Matlab. Software is built for scale Gradient Descent is not elegant. These problems can be solved with plain Calculus using the normal equation. However, normal equation doesn’t work on large data sets with 10,000+ features. Get used to Uncertainty Experts largely rely on rules of thumb. They don’t even fully understand how a lot of it works. Debugging is damned near impossible Neural networks are a black box and the length of time to complete an iteration means that you can’t get quick feedback. Quizzes are kinda hard They are strict about requiring 80% (4 out of 5) questions to pass. Save the PDF slides and review them before starting the test. Check and double-check your answers before submitting. ! I hope you found this overview useful. I’ve got to do my first assignment using Octave, but there will be more to come. Adios

Apple

Google

Netflix

How I’m learning “Machine Learning”

A source of employee dissatisfaction is ALWAYS a lack of communication. It’s in the top three complaints your team has. I guarantee it. Your team wants more info than you give. Inevitably you’ve tried a weekly team meeting. You put it on the calendar. The first one was a resounding success. You had so much to say that the meeting ran into overtime. The second meeting wasn’t bad, but the content dropped. By the third meeting, you struggled to scrape together enough to fill the time. You shifted to monthly as a compromise. Within four months the meeting was gone from everyone’s calendar. Or worse yet, it has continued as a painful time waster spent reviewing long to-do lists. Think about this. For a team of 25, a one hour meeting is costing you at least $1k. That’s $50k a year! A shoddy company-wide meeting reflects directly on you. It damages your leadership cred. On the flip side, if you’re not bringing the team together and keeping them up-to-date, you’re wearing that too. Looking for a solution that works? Here’s a PROVEN RECIPE FOR SUCCESS. One of my clients has been using this exact formula religiously, every week, for five years. Their employee satisfaction is off the charts positive. — You can achieve a ton in 25 minutes with a solid schedule. Use the time to maximum benefit Here’s a sample to get you started. Host starts the meeting at the scheduled time Good news stories (4 mins) — An opportunity for anyone in the company to share quick personal or professional good news. A new client. A pregnancy. The launch of a long awaited marketing tool. A promotion. Before the first meeting, prompt a few people to share. It will help get the ball rolling. Industry news (2 mins) — The host shares one or two bits of news from your industry. Something that’s new, exciting and worthy of sharing. Department updates (10 mins) — Each week, on a rotating schedule, one or two departments present an update. This is the time to share successes, learnings and upcoming plans, keeping everyone in the loop. Metrics (2 mins) — Report on the top two or three metrics that your company is tracking. Volume of new orders. Percentage up time. Number of new customers. Open Floor (4 mins) — An opportunity for anyone else to speak if they have valuable information to share. Often this is where other departments, committees, and you provide urgent updates. Recognition (1 min) — The host recognizes someone on the team. If you have company values defined, the recognition gets tied to a value. It needs to be a specific person for a specific action or achievement. A general appreciation for the awesomeness of a department or individual doesn’t cut it. Select the next host (1 min) — The current host chooses someone to be next week’s host. They also announce the departments presenting the following week. Embrace these four philosophies to rock your weekly meetings. Weekly without fail — Pick a time that works and NEVER move it. Steer clear of Mondays and Fridays to avoid public holidays and personal vacation days. The meeting goes, whether you are there or not (You should be there at least 75% of the time). Set the expectation that everyone needs to schedule around this meeting. Keep it tight — Irrespective of the size of your team, this meeting should run less than 25 minutes. If you can’t keep it on schedule, something’s not working. You will start losing the support of your busy team. Depend on rhythm — This meeting works or fails based on rhythm. Create a template, make sure everyone knows it, and stick to it. Keep it fun — If there isn’t a good dose of laughter every week, be concerned. These meetings should be fun. It’s a time for the team to connect. Keep it light. That’s the time tested recipe and here are the critical ingredients. — Someone needs to own this meeting. Often it’s an EA, receptionist or office manager. This person has a few critical responsibilities. Two days before the huddle, they send a reminder email to the host and presenting departments. Before the meeting they collect presentations and compile them into one document. On the day of the meeting, they make sure the space is ready, IT equipment is working, and music is playing. During the meeting they take notes of what’s said. Within 60 minutes they send those notes and the presentation to the entire company. Anyone who missed the meeting is back in the loop within an hour. Define a meeting owner — Set a standard start time for your meeting and be specific. My client starts their meeting at 1:07 SHARP every Wednesday and ends by 1:30. Why 1:07? Meetings scheduled for the top of the hour will start late. Period. Midday is good for spanning time zones. Early morning meetings infringe on your team’s work-life-balance. Late afternoon meetings will be lower in energy. Choose a time that works and stick to it. Start time — About 3 minutes before the start time, pump music through the office. It builds energy and provides a friendly reminder. The meeting host picks the music for the week. Pump the music — Any team member capable must stand. This is a quick huddle not a lingering social event. It’s a standing meeting — There are so many benefits to having someone different host each week. Here are four. The host gets to practice their public speaking and gets some visibility. No two meetings are the same. As the team grows, everyone gets to know everyone else. Most important, the meeting can always run even if you are away. Rotate the meeting host — Everything I’ve shared is perfect when your team is in one place. This is often not the case. Technology is the answer. Buy a portable video camera and microphone and broadcast the meeting using [insert your favourite tool here]. Have someone watch the computer so remote team members can type in questions. Remote employees — If your team is small (under 10), find ways to have everyone take part in every meeting. Keep it short. Have everyone share one good news story or their single biggest priority for the next 5 days. For larger teams, impromptu is better. Everyone has an opportunity to speak but doesn’t need to. Large vs small teams — Some companies use short (<10 min) daily meetings rather than longer weekly ones. Everything here still applies with two changes. Use the same host for an entire week and break the larger agenda down into smaller, shorter pieces. Daily vs weekly meetings Give it a try. I’d love to hear your success stories. If you hit some bumps, contact me and I’ll share more tips to get things on track. Thanks for reading :-) If you found this info helpful, please hit that ❤️ button below. It would mean a lot to me and helps other people discover the story. I blog to help Founder CEOs on their journey from kitchen table to 200 employees. In addition to my writing on Medium, you will find helpful articles and tools . Be the first to receive new articles by signing up for my newsletter . here here

Ball

Zones

Founders: How to rock your company huddle without it becoming an awkward time waster

In this post I look at some of the basic concepts behind agent-based computing, a research area that provides useful ways to , both in isolation and as part of a distributed solution. model intelligent programs Artificial Intelligence is all over the news. There are almost daily announcements of AI-powered products while mainstream media is full of stories of robots taking over , , and even and . AI is going through a golden period and the result will be a change of how we do things across the board. factory floors replacing truck drivers office workers stock traders The question we are going to try and answer here is how can we reason about AI as we are thinking of building our own AI-powered applications. What are the conceptual tools we need to be effective? We are going to provide a simple framework, light on buzzwords, to achieve this. Then we are going use that framework to determine where some specific AI techniques or technologies, such as natural language processing (NLP) fit in the picture. The origins of the term Artificial Intelligence First, though, allow me a small history detour. It is generally accepted that the term Artificial Intelligence came into its own around 1956, although it was used in various forms before that. This was the year a group of scientists decided to get together to specifically study AI. The actual proposal is fascinating, especially when viewed with the benefit of hindsight. McCarthy, Minsky, Rochester and Shannon wrote: “ ” — We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. (..). We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer. A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence Back in 1956, 10 very smart people planned to focus quite hard and make a “significant advance” over a summer. Undoubtedly optimistic and it is easy to smile at it now. However, they showed incredible foresight in realising what the potential was. Over that summer a whole new field of study was created. One with deep roots in computer science that spans to cover a range of fields from philosophy to genetics and psychology. As that initial proposal suggests the conjecture is that every aspect of learning or any other feature of intelligence can be described in a way that a machine can simulate it. Which leads on to the next question. What is a “feature of intelligence”? Let’s quickly start dispelling some myths by saying that it is not about human intelligence. We are also not going to complicate it by using the vague phrase “sufficiently hard problems” or “complex behaviour”. As evolutionary AI teaches us very simple behaviour can lead to some very extraordinary results. Also, AI does not just mean machine learning (ML), although ML is certainly the current driver of renewed enthusiasm in AI as a whole. Nevertheless, while entertaining, it is not particularly useful to agonize over the intricacies of neural networks and deep learning architectures for practical applications the same way it is not useful to worry about the details on the low-level networking communication protocols when building websites. In order to understand AI in a practical way that can help us better understand the properties of the systems we are using, designing and building we will use a different paradigm. We will draw from a field of computer science called Agent-Based Software Engineering or Agent-Based Computing. Agent-Based Computing A simple way to describe agent-based computing is that it is a combination of software engineering and artificial intelligence. It concerns itself with how intelligent programs (agents) can be structured and provides ways to model and reason about your system and the interactions between agents much in the same way object-oriented modelling does. We will understand AI by understanding what an agent is. Reactive Agents One of the classic textbooks for AI ( by Russell & Norvig) describes agents as “anything than can be viewed as its environment through and acting on the environment through ”. While this is not necessarily a complete understanding it is good enough for our purposes. Artificial Intelligence: A Modern Approach perceiving sensors effectors An agent is something that perceives its environment (for example a chatbot waiting for a message from a user) and based on the input it receives (“Hello bot, how are you”) it will respond (“I am fine and how are you?”). An agent can also be a thermostat that perceives the temperature of a room and switches off the heating when the desired temperature is reached. “Hold on there”, I hear you say. “A thermostat is an AI — I thought AI is about hard problems!?”. That is a fair statement but we are trying to build a framework that will help us deal with all sorts of situations without any vague assumptions of “complexity”. Do not get hung up on “hard problems” or “complicated situations”. Everything lies on a continuum and we will get there in time. The examples we gave here are of what would be called . Input A comes in and Output B is offered. Every time the same thing will happen. If no inputs are received our agent will not do anything. Not very exciting but it is a start. reactive agents Proactive Agents Now, imagine your bot is a restless one. It is not happy to just sit there and wait for someone to speak to it. Instead it has its own agenda, its own goals. This restless bot you to interact with it. So it is going to act proactively to achieve this. Say it is a news bot — its objective is to get you to read news. Rather than waiting for you to input something it is going to write to you with “You won’t believe what happened today! Click here for the breaking news”. wants Proactive agents can be described as having desirable environmental goals. They perceive the world (Ron hasn’t read the news yet), can act to change it (“Hey Ron! Read some news!”) and there is some state they want the world to be in (Success! Ron has read the news.). Proactive agents are definitely a bit more fun. Agents that Learn Now, let’s assume our agent is really very very keen about getting us to go read the news. In order to achieve its goal it knows it has to communicate with us and convince us to perform an action. A simple proactive agent may have a set list of actions it can perform to achieve its goals. A learning proactive agent will also attempt to determine the efficacy of those actions (perhaps by applying some utility function on the results) and it will adapt its actions. For example, if “You won’t believe what happened today!” is not an effective hook then it might try “Click here for the latest news” or “The world is changing! Find out how!”. There can be any number of layers of complexity hidden behind this learning proactive agent. The ways it assigns a score to different reactions can get extremely sophisticated and it may be able to draw not just from the reaction of a single person but thousands or millions of users with which it interacts. If the data grows and the hooks to test are multiple it will need special tools to make sense of them. Big Data analysts will be hired to identify the patterns. Hundreds of machines will be crunching numbers. Nevertheless, the goal “Get user to read news” remains a simple one — irrespective of the amount of work that goes on in the background to figure out how to achieve it. It is also a single goal. It may perform different actions (i.e. utter different phrases) but they all want to lead to the same world state. One where we have read the news. Thinking of our learning proactive agent in this way allows us to box of the underlying complexity and understand the immediate issue at hand. Namely, we want to proactively engage with the user and get them to achieve a specific goal. Autonomous Agents Now, let’s make it a bit more interesting. We will assume our agent has a rather more sophisticated objective. Instead of being driven by a goal — i.e. by a simple need to change the world from State A to State B it has motivations. These higher-level objectives such as “Get user to engage with politics more” could be satisfied by a number of goals. One might be “get user to visit site and read policies” while others are: “get user to like Facebook page”, “get user to reply to 5 survey questions”, etc. Each goal will have a number of plans or actions associated with it and our agent will have to decide which to pick based on some mechanisms. At this point we have an autonomous agent. We are not necessarily completely sure what it is going to do next and how the world is going to change because of that action (within a limited sense of course). It can decide how to change the world in order to satisfy its motivations. You can probably imagine how complex things can become if an agent has multiple motivations, each leading to multiple goals with the motivations having different weightings. In short, things can get really interesting. Agent Communities For the sake of completeness let us also briefly consider multiple interacting agents. Assume that an agent community has a common motivation of “get user to engage with political action”. Each individual agent, however, has differing capabilities and motivations. One agent is the Facebook Messenger agent with a particular focus on the Facebook environment and social media engagement, while a website-based agent is more interested in getting a user to offer money to the campaign. We might decide that they need to negotiate amongst themselves about who gets to influence the user in what way. A lot of work in agent-based computing is actually around negotiation mechanisms with agents even holding auctions as they bids for a chance to get to influence the world. The value an agent is willing to assign to an action is a useful mechanism to maximize utility across the system. Once more, you can quickly see how the problem grows as we are essentially attempting to model a whole community of interacting agents with different motivations, different ways of perceiving the world, etc. Building AI-powered Applications But you didn’t mention machine learning! What about Natural Language Processing? Vision? Autonomous driving? Isn’t that the “real” AI? Yes, we haven’t mentioned machine learning in quite the way it is usually talked about. Our objective is to create a simple framework that we can use to understand some of the AI that we hear about in the news and, more importantly, how to think of our own applications. Cloud-based APIs powered by machine learning can make our agents more intelligent than ever. That is why AI is suddenly relevant once more. While we could reason in terms of goals, motivations and actions that was not necessarily too exciting if our agent could not understand what a user is telling them, or cannot accurately predict what a user will like. This means that our agents can behave in ways that are more sophisticated. We, however, need to be able to reason efficiently about the behaviour of the agents we are building. That starts with basic characteristics such as reactivity, proactivity and autonomy. Everything else can build on top of that. Now, let us see look at a couple of the typical technologies or application domain areas that are mentioned as prime examples of AI. Natural Language Processing If our chatbot is plugged into an NLP API it could be modelled as just another sensor our agent uses or it can be modelled as simple reactive agent that our agent is communicating with. In either case we would expect that given the same phrase we would always get the same result. If, however, the NLP agent has a goal of always giving a better response (based on whatever learning mechanisms are in the background) we are better off modelling it as a proactive learning agent and the rest of our application needs to ensure it is happy with the response potentially changing over time. The agent (or agents) that actually learn how to do NLP before we get to use it through an API are not our concern. This is precisely the key. It is easy to get caught up in all the layers of complexity — ultimately however we need to divide that in neat little boxes and focus on what is important for our AI-powered application. Autonomous Driving Autonomous driving is a great example of how we can use agents to reason about a problem. Agent-based computing would suggest we consider a car not as a single agent but as a community of communicating agents. Each one is making thousands of decisions about the placement of other cars, the position of pedestrians, the management of the car. Each one is plugging into complex algorithms to figure this out. Decisions are shared and ultimately competing goals or motivations are satisfied such as “Take me back home”, “Don’t kill anyone while trying to get me there”, “Save fuel”, etc. Next Steps When building applications that need to interact with people or other applications in interesting ways agent-based concepts can greatly help to categorise and reason over what we are trying to do and what AI is. It allows us to have a single vocabulary through which to describe what our system is doing. In this post we merely scratched the surface in order to set the scene. Agent-based computing has a lot further to offer as it describes the internal architecture of individual agents as well as architectures of communities of agents. As machine learning solutions become more widely available and sophisticated programs that proactively and autonomously attempt to solve problems become a necessity agent-based computing offers the framework of both how to think about them and actually build them. In follow-up posts we are going to dive into the specifics of modelling a using agent concepts and how it interacts with the world. chatbot Part 2 of the guide is now published: “Understanding what the user says”. Interested in why chatbots are actually here to stay? Read our post on the drivers behind conversational interface technology or find out more about our chatbot build services. This post was originally published on deeson.co.uk

Making Artificial Intelligence work for you: Part 1 — What is AI?

When I was a kid I read everything I could get my hands on about Nikola . I absolutely idolized the guy. He was pretty cool. He was gifted/cursed with the ability to run experiments in his head, visualizing them as clearly as if he was actually seeing them run in real life. Tesla He made some astonishing breakthroughs in the world of electricity, radio, and robotics. However he died penniless in a run down hotel in New York. Part of loving Tesla as kid was learning to despise Edison. According to Tesla, Thomas Edison was dull, inefficient and “ .” Tesla was the genius, Edison the imposter. money hungry They approached their work differently: very Tesla’s Process Come up with some futuristic, idealistic idea Spend time experimenting and building it out (mostly in his own mind) Get a patent Try and find someone to finance and “market it for you” Edison’s Process Pick a market that you understand. Find a better solution to an apparent need. Test test and test solutions out until it worked in real life. Get a patent Marshal resources until it reached adoption in the real world. How following Tesla’s Process screwed me over. I kept coming up with really cool ideas, ideas that I thought would change the world in amazing ways. I would go out, code them up, and then try and market them to the world. That usually meant a launch to crickets, $0 in revenue, and 0% change for good in the world. You see, the problem is that Tesla’s Process is great for building things that work out well in your mind but not so great if you want to build things that work in reality. “You know what’s better than building things up in your imagination? Building things up in real life.” -Ryan Holiday Eventually I got fed up with this cycle and have decided to move over to Edison’s camp. I really don’t want to die penniless in some random hotel. Edison’s way “Edison blended invention with economics. His electric lighting system was designed to be an economic competitor with gas lighting.” -Wikipedia You can’t make lasting change in the world or hope to have any level of success if you ignore the realities of economics and human nature. If you think that having a “great idea” is what matters most you are screwed. actually In other words, you can only make lasting change and be successful if you get good at business. That’s how Edison did it back then and that’s how people like , , and are doing it today. Elon Musk Jason Fried Mark Zuckerberg So what about you? Are you going to succeed like Edison, or are you going to be like Tesla and eventually die without a dime in some hotel? I’m starting a free slack community for people learning to be better at business together. Join us.

Tesla vs. Edison

Introduction or Why You Should Try a Bot (you may skip it if you already know what to do with your bot) Note: the code samples may be displayed improperly because of markdown. I recommend to to make sure all the examples are displayed properly. continue reading the original article on our blog are everywhere. It seems that only yesterday we did not even know about their existence; now we can barely imagine our life without them. They’ve become widely popular among numerous active users of messengers since they have various scopes of use — from entertaining content including step-by-step games and collecting bonus points in restaurants, to keeping a diet plan, tracking deliveries, and even making payments for different services. Bots Why are they so popular? What’s their secret? I think the more relevant question is why they are more convenient than an app. And there are a few reasons. Well, the bot simply cannot have colorful design. But beyond any doubts, in comparison to numerous apps with different design when you have to remember where and what to tap, bot is more universal and easy; it offers a simple communication via texts. 1) Minimalistic and simple design. You do not have to install hundreds of apps for each service if you can receive all necessary assistance from a bot. This is particularly useful for restaurants and shops. Clients are rarely eager to install apps from a bunch of places they’ve visited. Due to this, business owners miss clients’ feedback and lose communication with them. If each of these places had their own bot available in different messengers, it would be more convenient and friendly to users. Nobody likes to fill up the storage of their phones with unnecessary apps that will be used once or twice. However, clients need to engage with the service owners and they will appreciate doing this through their favorite messenger. 2) Bot has minimum of advertising and focuses on the users’ needs. Using a bot, user passes authorization only once when the bot is added to the chat. Client can use the bot as much as it is necessary, and when there is no need in it anymore, user just blocks the bot. That’s all, easy! No password resets anymore. 3) No need for signing up, authorization, and constant relogin No need to remember passwords or logins used. Adding bot to the site or app increases the number of the user audience as it makes communication with clients and providing them with the assistance much easier and user-friendly. So, we described main advantages of the bots and now you must be willing to create your own bot. Let’s move on to the practice. But first, we’ll take a look at issues that must be addressed in the pre-development stage. Nuances of Telegram Bot Development When we’ve already determined reasons for creating the , now it’s time to think on how we plan to organize the development process and what tools we will need. Further, we will demonstrate in practice how to create your first bot and how to teach it to turn our message inside out. bot In this part, we are planning how to and what development tools to use. Further, we’ll show how to build your first Telegram bot and will teach it to turn our message backwards.Since it is the manual for beginners, we will run the server with a single endpoint that will receive our telegram messages and will make an answer. build the application For that, we will use the following tools: — for our server; a simple and lightweight WSGI micro web-framework bottle — for sending requests to telegram. lib does not need to be overrepresented. It is universally used throughout the world in a variety of projects.Note: you have to install these tools on your computer. We will need them later. For that, open your bash console and install it via pip requests request pip install bottle requests – this is an app which provides us with public URLs for our interaction with Telegram WebHook throughout the development phase (look for info on WebHook below). It is useful, as Telegram will not be able to make a connection with our local server because we cannot specify our local address in the Telegram API configuration.You should download from the and put the installed app in the folder with the project. ngrok ngrok official site How About to Create Your First Bot? If you are keen on this, explore this part where we will provide a detailed Telegram bot development tutorial. First things first. You need to register on Telegram (obviously). I recommend to use for testing the basic concepts. Telegram web client Open Telegram app, search for and start the chat. Send command and follow the instructions. After completing the initial steps, you’ll get — @BotFather /newbot your TOKEN telegram api URL — https://api.telegram.org/bot link on documentation Wells, that’s actually it. At the moment the bot is 100% passive. You need to initialize a conversation with your . Open search and type in the name of your bot. Start a conversation by clicking on button. Type in something like “Hello”. This message is important as it’s the first update your bot is going to receive. bot /start If it’s your first experience with building APIs, you can easily get the idea using your web browser. Open a new tab in your browser and use the Telegram api URL – https://api.telegram.org/bot<token>/getUpdates When you open this URL in your web browser, you make a request to Telegram server, that responds back with JSON. Response resembles Python dictionary. You should see something like this: {"ok":true,"result":[{"update_id":523349956,"message":{"message_id":51,"from":{"id":303262877,"first_name":"YourName"},"chat":{"id":303262877,"first_name":"YourName","type":"private"},"date":1486829360,"text":"Hello"}}]} If you open bots documentation and check method you’ll notice that this method requires 2 additional parameters and . In a browser search bar you can chain parameters using **?**for the first one and for all the consequent. Send message would look like this – /sendMessage section chat_id text & /sendMessage?chat_id=303262877&text=test Try to get the reply from your bot by substituting with one you get by calling . In my case it is 303262877. Text parameter is up to you. The request should look like this chat_id /getUpdates https://api.telegram.org/bot<your-token>/sendMessage?chat_id=&text=<your-text> WebHook (You can skip this part if you’re familiar with WebHook) To put it short, a WebHook is an API concept that grows in popularity. The concept of a WebHook is simple. A WebHook is an HTTP callback: an HTTP POST that occurs when something happens; a simple event-notification via HTTP POST. To explain a bit more, sometimes an interaction between apps online requires immediate response to the event, while solutions for constant and continuous connections are mostly cumbersome, exacting and hard to support. In this case, the best and the easiest solution is immediate callback via HTTP (most often POST). In other words, this solution ensures response to any event inside the one app by means of sending HTTP POST request to other connected app in order to inform it or to make it respond. This exact concept is called WebHook. It is widely used for: receiving data in real time receiving data and passing it on processing data and giving something in return Seems that it is the best solution for interaction of the Telegram client (Telegram app) with our project. Coding Part At last, we start the most practical part where you will be able to develop a Telegram bot. Main task: teach our bot to turn our message backwards Firstly, create a folder for our bot project. Secondly, create bot.py file to make a bottle server. Next, we develop bot.py from bottle import run, post

@post('/')  # our python function based endpointdef main():      return

if __name__ == '__main__':      run(host='localhost', port=8080, debug=True) Let’s try to start our server. For this, open bash in your bot folder. python bot.py In the result, you should see something like this: Then, open a new tab. In the next tab, we will start ngrok ./ngrok http <our_server_port> ./ngrok http 8080 After, you’ll see something like this: Now, let’s set WebHook. https://api.telegram.org/bot<your_token>/setWebHook?url= https://<your_ngrok_url.ngrok.io/ Note: to find ngrok URL, you have to launch ngrok. Then, on the screen similar to the one below, you’ll find URL (it is highlighted on our screenshot). This URL you use in the link for setting WebHook. Response to following the link should be like: {"ok":true,"result":true,"description":"Webhook was set"} Let’s check if you succeeded in setting up WebHook. Follow this link using your token: https://api.telegram.org/bot<your_token>/getWebhookInfo If everything is right, you’ll see the same ngrok address value in front of URL key that you specified when configuring. Congrats, it’s alive! Now, we need to implement a message request/response mechanism.Basically, our endpoint gets data in json format. So, normally, you’ll see data message. from bottle import run, post, request as bottle_request  # <--- we add bottle request

@post('/')def main():      data = bottle_request.json  # <--- extract all request data    print(data)

return

if __name__ == '__main__':      run(host='localhost', port=8080, debug=True) It should be something like that in your console tab where the server is launched. {'update_id': <integer>, 'message': {'message_id': <integer>, 'from': {'id': <your telegram id>, 'is_bot': False, 'first_name': '<your telegram name>', 'last_name': '<...>', 'username': '<...>', 'language_code': 'en-En'}, 'chat': {'id': <integer chat id>, 'first_name': '<...>', 'last_name': '<...>', 'username': '<...>', 'type': 'private'}, 'date': 1535022558, 'text': '1'}} More detailed information on the parameters you may find in the . official documentation of Telegram Now, we have to extract and in order to turn our message backwards and send the answer. chat_id text from bottle import (      run, post, response, request as bottle_request)

def get_chat_id(data):      """    Method to extract chat id from telegram request.    """    chat_id = data['message']['chat']['id']

return chat_id

def get_message(data):      """    Method to extract message id from telegram request.    """    message_text = data['message']['text']    return message_text

def change_text_message(text):      """    To turn our message backwards.    """    return text[::-1]

@post('/')def main():      data = bottle_request.json    answer_data = prepare_data_for_answer(data)

return response  # status 200 OK by default Now, we’ve already prepared the answer. Let’s send it to Telegram bot. import requests  from bottle import (      run, post, response, request as bottle_request)

BOT_URL = 'https://api.telegram.org/bot<YOUR_TOKEN>/' # <--- add your telegram token here; it should be like https://api.telegram.org/bot12345678:SOMErAn2dom/

def get_chat_id(data):      """    Method to extract chat id from telegram request.    """    chat_id = data['message']['chat']['id']

return chat_id

def get_message(data):      """    Method to extract message id from telegram request.    """    message_text = data['message']['text']

return message_text

def send_message(prepared_data):      """    Prepared data should be json which includes at least `chat_id` and `text`    """     message_url = BOT_URL + 'sendMessage'    requests.post(message_url, json=prepared_data)  # don't forget to make import requests lib

def change_text_message(text):      """    To enable turning our message inside out    """    return text[::-1]

def prepare_data_for_answer(data):      answer = change_text_message(get_message(data))

json_data = {        "chat_id": get_chat_id(data),        "text": answer,    }

return json_data

@post('/')def main():      data = bottle_request.json

answer_data = prepare_data_for_answer(data)    send_message(answer_data)  # <--- function for sending answer

return response  # status 200 OK by default

if __name__ == '__main__':      run(host='localhost', port=8080, debug=True) After all preparations and coding, if everything works, let’s try to chat with our bot. Now, let’s make our code more readable and implement OOP structure. import requests  from bottle import Bottle, response, request as bottle_request

class BotHandlerMixin:      BOT_URL = None

def get_chat_id(self, data):        """        Method to extract chat id from telegram request.        """        chat_id = data['message']['chat']['id']

return chat_id

def get_message(self, data):        """        Method to extract message id from telegram request.        """        message_text = data['message']['text']

return message_text

def send_message(self, prepared_data):        """        Prepared data should be json which includes at least `chat_id` and `text`        """               message_url = self.BOT_URL + 'sendMessage'        requests.post(message_url, json=prepared_data)

class TelegramBot(BotHandlerMixin, Bottle):      BOT_URL = 'https://api.telegram.org/bot000000000:aaaaaaaaaaaaaaaaaaaaaaaaaa/'

def __init__(self, *args, **kwargs):        super(TelegramBot, self).__init__()        self.route('/', callback=self.post_handler, method="POST")

def change_text_message(self, text):        return text[::-1]

def prepare_data_for_answer(self, data):        message = self.get_message(data)        answer = self.change_text_message(message)        chat_id = self.get_chat_id(data)        json_data = {            "chat_id": chat_id,            "text": answer,        }

return json_data

def post_handler(self):        data = bottle_request.json        answer_data = self.prepare_data_for_answer(data)        self.send_message(answer_data)

return response

if __name__ == '__main__':      app = TelegramBot()    app.run(host='localhost', port=8080) That’s pretty much it. Now you have a working Telegram bot that can even spell “racecar” backwards. Congrats!You can find error codes . here With a free account you’ll face certain limitations. However here you have a fully functional bot. The article is written by — Backend Engeneer at . This article about is originally posted on . Arez Khalimi Django Stars telegram Django Stars blog Specially shared with readers. Hackernoon You are always welcome to ask questions and share topics you want to read about! If you find this post useful, please tap 👏 button below :)

Chain

Django Stars

Heroku

Super

Read My Stories

How to Create and Deploy a Telegram Bot?

My daily work usually starts by opening an SSH connection to a server, running a docker image (with RStudio Server or Jupyter on it), and analyzing data or programming directly on the browser. It was always convenient like that until I got sudden disconnection last month. Suddenly anything stops working, and I wasted several hours hopelessly trying to fix it. When I went home, I figured out that the electricity in my apartment was very unstable due to a small construction upstairs. Of course, a simple solution is getting a UPS (Uninterruptible Power Supply), but I was fascinated by the idea that maybe I can turn on my server over the internet. That is like having the server’s power switch with me all the time. It would be so cool, especially when I’m away for an extended period and don’t want to waste money on the energy bills. This story is about how I’ve done it. You can have a look at my final network setup first. : Satisfy hardware requirements Step 1 In most motherboards, there is a function called “Wake-on-Lan” (WOL). WoL works by sending a packet of data called a Magic Packet™ to a target machine. When the packet is received, the target machine’s network interface wakes-up the rest of the computer. However, I recognized that I cannot wake-up my server through its wireless interface (I’m using a USB wireless adapter). It turned out that the Magic Packet must be sent to the motherboard’s built-in Ethernet port since it’s the only thing awake when the whole machine off. I fixed the situation by using a , and connect it to the server through an Ethernet cable. Wifi Repeater You can skip this step if your machine is using a wired connection. : Make Wake-On-Lan work on Lan. Step 2 As the name suggested, WOL is designed to wake-up a machine through a local network. So you have to be in the same network with your target device to make it work. In this step, we will try to do just that.First, you have to turn on the WOL feature on your motherboard. Enter the BIOS setup and look for “ Wake up on PCI event” or “ Wake up on LAN” and enable it.Second, on Ubuntu, you have to know your default interface, check if it supports WOL, and turn on the feature. The whole process can be done as following: $ route #Get name of the default interface, for example: $ sudo ethtool eth0#Look for "Supports Wake-on: g" line, which means WOL supported$ sudo ethtool -s eth0 wol g  #Enable WOL eth0 To make the setting sustains after rebooting, you need to add the last command above to the interface’s configuration in . You can read more in . etc/network/interfaces here Now, let’s test it. You can use a mobile app called “Wake-on-Lan” in Android (there are similar apps for iOS, but I haven’t tested it). It can automatically scan your local network and list all of the connected devices. You just search for your machine, add it, and then you can send the magic packet to it whenever you want. Try to turn off your target machine and wake it up by your phone. You can add the device manually if you know its mac address and broadcast address. They are 9E:65:F9:0E:29:FB and 192.168.0.255 in my case. You can use the command to find out that information. Note that you will need your mac address to wake up your machine through the internet. ifconfig : Make Wake-On-Lan work through the internet. Step 3 Note that your router must have a static IP (i.e., ) to complete this step. There are many solutions to send the magic packet through the internet. I’ve listed them below from the easiest to the hardest (sometimes the only) solution: 82.164.3.27 : If your router supports forwarding the magic packet (Port 7 or 9 UDP), you are lucky. You can just forward any packet from port 7 or 9 UDP to your target device. Some routers even allow forwarding broadcast packet. You can use to wake up your server. However, as far as I know, most of the popular commercial routers do not support this (my case). Remember to make a DHCP reservation for your server, so that its IP will not change after reconnecting. First solution this website : You can check if you can install a better firmware for your router. For marketing purpose, many router producers hide functionalities on their products. You can install a third-party firmware to unlock all of these. I highly recommend the . Check if your router is supported by OpenWrt . Second solution OpenWrt here : If your router is not that good, you will need an insider. It’s a device that is always on and connects to your local network. Then you can SSH to your insider over the internet and let it send the magic package for you. In my case, I use a cheap Raspberry Pi 3 (which I already use for controlling other things). I’m using OpenWrt on it, and run the etherwake command as following: Third solution $ etherwake 9E:65:F9:0E:29:FB eth0 The first argument is the mac address of the target machine, while the second argument is the network interface of the insider that will send the package. It should be its default interface. You can check for more detail. Note that your insider can be any device, running any OS. As long as you can SSH to it through the internet, it should work. Remember to configure port forwarding for the insider’s SSH port, so that you can SSH to it from anywhere. this link So, that’s it. I hope you enjoy the story. Leave a comment if you have any trouble following it.

Wake-On-Lan through the internet

After 6 years at startups, I've made the transition into venture capital and joined . It has just been over 3 months since I began spending my days screening startups and talking to interesting founders from all around the world. Thus far, my transition from a founder's role into an investing and advisory one has been a breakneck, intense education. Hummingbird Some parts of the job are as I expected but others have been somewhat surprising. Before it all becomes a blur, I thought I’d share some of the most interesting and unexpected observations I've made so far; #1 Being a VC is a very UnStRuCtUrED job I believe there are two underlying reasons for that:(a) While VCs invest in some of the most ground breaking technologies out there, most of the industry relies on the most elementary tools ( Email, Calendar and Excel); (b) the fear of missing out on a deal combined with the amount of available company data — makes it almost impossible to keep laser focus on one deal or one vertical, which results in a rather chaotic workflow. think 'ECE' = and body[data-twttr-rendered="true"] {background-color: transparent;}.twitter-tweet {margin: auto !important;} great entrepreneurs make the most of the cards they're dealt, while great investors fold until they get a good hand. optimism vs. discipline — @scottbelsky function notifyResize(height) {height = height ? height : document.documentElement.offsetHeight; var resized = false; if (window.donkey && donkey.resize) {donkey.resize(height); resized = true;}if (parent && parent._resizeIframe) {var obj = {iframe: window.frameElement, height: height}; parent._resizeIframe(obj); resized = true;}if (window.webkit && window.webkit.messageHandlers && window.webkit.messageHandlers.resize) {window.webkit.messageHandlers.resize.postMessage(height); resized = true;}return resized;}twttr.events.bind('rendered', function (event) {notifyResize();}); twttr.events.bind('resize', function (event) {notifyResize();});if (parent && parent._resizeIframe) {var maxWidth = parseInt(window.frameElement.getAttribute("width")); if ( 500 < maxWidth) {window.frameElement.setAttribute("width", "500");}} In contrary, being a good VC requires a lot of discipline and patience. Many people think the best VCs make a ton of deals, however some of the most successful VCs don't really make a lot of deals. #2 Meaningful Exits and Fierce Competition Source: by Meaningful VC Exits Samuel Gil, CFA There has been a lot written on this topic already, so I will keep it brief. For a normal sized €50M European fund a meaningful exit is around €80M, and a home run that will return the entire fund, is a €250M exit. For founders this obviously means that their startup must have the potential to generate significant revenues in order to raise interest from VCs. It's good to note that there are also many startups that don't want or need investor money. Now, to put things into perspective— for example, in there were around 1800 exits. Out of which 45% were for less than $50M, 72% were for under $200M, and just 1% were in the billion dollar club. 2013 Source: TechCrunch there were 3260 exits. Out of which 44% were acquired after Seed or Series A (61% including Series B and Convertible Notes). Hence, we can see a general trend of companies getting acquired sooner ~ at a lower valuation. If we look only at Europe, there were 658 exits. So my ; there were around 240 meaningful exits, and 60 home runs. Last year educated guess According to , only there are over investors, of which more than are VCs. I think realistically the number of active investors is significantly lower; so I'm guessing there are around 900 active VCs in Europe. But still, quite a competition, right? CrunchBase in Europe 10,000 1700 #3 Figuring out how not to drown in the ocean of data On any given day, there are between 40 and 50 VC deals taking place (that's over 1200 deals on a monthly basis). ( ) Next to that, startups raise funds from angels, incubators and crowd-funding platforms. In addition, there is a ton of conferences, demo-days, startup competitions, and industry curated lists. And not to forget the endless stream of articles, tweets, newsletters, and press releases. As a VC, I need to constantly keep up with four types of data; (a) Company Related, (b) Investor Related, (c) Founder Related, and (d) Industry Related. Disclaimer: those are just publicly announced deals Another big topic is company data. There are tens of databases out there — like CrunchBase, , , , , , , DealRoom, or LinkedIn. And some of them already cover millions of companies. So in theory you shouldn't really need another source of data, right? Well, the truth is, you kind of do. Again, there are a couple underlying reasons for that;(a) Accurate data is scattered across many sources and it tends to get outdated every 45–60 day(b) Many times you still end up missing valuable data points (c) Do you really want to rely on the same data as everyone else? Mattermark AngelList CB Insights PitchBook DataFox Index.co Product Hunt and To put things into perspective again; only last week I screened around 21,000 companies of which 78 passed my screening (0.37%). And out of those, I might take a deeper look at 4 or 5 companies (0.02%). That’s roughly . And it’s good to note the list I was screening was already somewhat curated. 1 out of 5000 #4 Literally, everything becomes relative For every success story, there is a better one. For every funding round or exit, there is a bigger one. And for every metric, there is a company that achieved it faster. And at some point, I think you need to learn to accept that, and stop with the constant comparing and benchmarking. Otherwise, your are risking that you will start feeling quite miserable. I've seen year old founders who are already outstanding CEOs. Nowadays there are even year old VCs, and 13–14 year old teenagers starting deep-tech companies. The impact of technology on entrepreneurship is truly amazing, especially in emerging economies. 18 19 I remember one of my close friends always used to complain about the size of European funding rounds compared to the ones in US. I think nowadays the difference isn't as big anymore. However, if you think some funding rounds and valuations are crazy, I suggest you to have a look at some of the rounds that are taking place in . China #5 The Age of Reverse Entrepreneurship If you are reading this blog, I guess you are already familiar with the concept of . For the past few weeks, I've been working on an essay in which I argue that entrepreneurship is currently going through a very similar phase. I still need to publish it, but to give you a bit of a sneak peek; Reverse Engineering At the beginning of this century, Internet was mainly about aggregation — . A great example is Craigslist a classified website launched in '95, which covers everything from jobs to housing. the collection of related items of content so that they can be displayed or linked to Source: CB Insights In 2010, Andrew Parker from Spark Capital published this now well-known chart which highlights the different startups attacking different parts or of Craigslist (also known as “Unbundling Craigslist”). Now, a couple things to consider — only by 2015 those startups raised over $10BN in funding, and the chart only covers US companies. Try to imagine the same chart including all their global counterparts. Of course, not all of those startups got their inspiration from Craigslist. But what I'm getting at is that from we are moving towards And this is happening through the process of . aggregation hyper specialisation. Reverse Entrepreneurship There are a couple additional trends that support this thesis; ; companies first launching in Europe — then, as a second wave in emerging economies (obvious example is Rocket Internet); with the rise of APIs, we see companies piggybacking on the popularity of others; how many times have you heard/seen a startup saying that they are building the “Uber of XYZ” or “Amazon for ..”; the concept of where companies share their most successful marketing experiments with others. (a) Copycatting (b) API Economy ; (c) Uber for X fad; and (d) Sharing Growth Strategies ; Growth-Hacking This trend becomes especially apparent when you become a VC. Since I joined Hummingbird, I already screened tens of thousands of companies — and honestly, I would have a really hard time putting together a list of 20–30 companies which have a truly original business. #5 “..all I know is that I know nothing” I admire people who confidently put the word 'Expert' in their online bio's. As for me, every day I learn more but know less. On any given week, I speak to ~40 companies ranging from Logistics to Blockchain or Machine Learning. And it's really really difficult to keep up with all the new tech developments, industry shifts and disruptive ideas. (And I'm saying this from a perspective of someone who is digitally savvy, and who has experience with working with clients across 70+ industries.) I think the most important VC skill is the ability to ask smart questions. That is — . And that doesn't come without a lot of learning. challenging questions that help to keep the conversation both engaging and as efficient as possible Written by Dominik Vacikar . and don’t reflect the views of my employer. Views are my own Feel free to contact me at; dominik@hummingbird.vc

Funding

Techcrunch

Uber

My First Months as a VC

Note: This article is derived from a sample chapter from my upcoming book . The book guides you from humble beginnings through deploying an app to production. The book is now in pre sale, and you can grab a free chapter right now! Build A SaaS App in Rails 6 Also, the beta for my new project is has been released. If you’re losing track of pull requests, have old ones lingering around or would just love a dashboard that aggregates these requests over multiple services (Github, Gitlab, and Bitbucket), . Pull Manager check it out Testing, while requiring a little bit more cognitive load up front, and even loathed by some developers, will often pay dividends down the road. It will almost always have you saying “I’m glad I wrote tests for this” at some point. TDD(Test Driven Development) is one approach that has grown on the Rails community. Although, not completely accepted by all in the community. This approach has you writing your test(s) first and then writing code to make your tests pass. Why Test Testing will provide a few main benefits when done correctly. First, it will help you develop the behaviors within your app. You may write a test that your model is validating the presence of an email. Then realizing you needed to include validations for the password or add a “retype password” input box. Another great reason to test is to protect yourself against regressions within your application as your modifying code or adding new features. This will also give you confidence and assurance when your refactoring classes, collapsing code or removing extraneous code. If written tests pass both before and after changes you should be good to go! Another testing method is to write a test after you find an issue or bug in your application. Ensuring the fix works and does not plague your application in the future. Testing Frameworks Like many other parts of Rails, there are default gems and libraries. As well as other optional alternatives, which at times are more used by than the defaults. RSpec vs MiniTest As of the latest version of Rails, the default testing framework that is included is MiniTest. Which in itself is an improvement on the original testing framework for Rails Test::Unit. Personal preference can play a huge part in choosing many aspects of a Rails application. Testing is no exception. The largest alternative to MiniTest is RSpec. Which focuses on readability and composability of tests. It uses words such as “describes” and “it” when naming tests and test groups. For example: describe “the test” do it “is a success” do expect(1).to eq(1) end end Whereas MiniTest will use the asset verbiage found in other programmatic testing frameworks. Personally, I like RSpec, find it easier to reason about, and has a little bit better of a command line interface(CLI). Capybara Feature Tests vs Integration Tests Another noteworthy alternative in the testing work is the addition of Capybara. While you can use built in Integration tests to test workflow. It leaves desiring more for real world browser workflows. This is where Capybara comes in. Allowing you to write feature tests taking you through pages of your choosing within your application. Performing behaviors such as clicking, filling in a form or verifying that everything is behaving as expected. Capybara plugs right into RSpec, allowing you to use the same syntax and phrasing to write tests. Here is an example right from their docs! describe “the signin process”, :type => :feature dobefore :each doUser.make(email: ‘user@example.com’, password: ‘password’)end it “signs me in” dovisit ‘/sessions/new’within(“#session”) dofill_in ‘Email’, with: ‘user@example.com’fill_in ‘Password’, with: ‘password’endclick_button ‘Sign in’expect(page).to have_content ‘Success’endend FactoryGirl The last alternative I’ll mention is the gem FactoryGirl. The default method of test objects within Rails is fixtures. Fixtures files are in written the YAML format. Providing a basic static mock object representation. Fixtures will get tougher as the model objects get more complex. Including their associations proliferating. FactoryGirl describes itself as an API to create test objects. Meaning, you will programmatically use the API to create mock objects for your tests. You can provide a factory definition, in Ruby, then use the API much like normal object to `.build` or `.create`. FactoryGirl also makes it super simple to include associations through the factory definitions. Testing Types There are three main types of tests you will find in a typical Rails application. Model(Unit) test, Controller tests, and View tests. Additionally, you can also use Feature Tests. These tests will function as an end to end walkthrough test of your application. Each type of test has it’s place in the testing world and may be equally important. Model/Unit Tests Model specs are most likely the simplest of the test types you will deal with in Rails. With these tests you are testing the normal CRUD(Create/Read/Update/Delete) operations on the model object. Additionally, they will test validations and/or associations. Model specs will normally run super quick and be fairly self-contained FactoryGirl.define do factory :user do name “Same Name”

email “test@user.com” end end class User < ApplicationRecord validates :email, presence: truevalidates :email, uniqueness: truevalidates :name, presence: true … describe User do it “has a valid factory” do expect(FactoryGirl.build(:user).save).to be\_true end it “is invalid without a name” do expect(FactoryGirl.build(:user, name: nil).save).to be\_false end it “is invalid without a unique name” do user = FactoryGirl.create(:user)

expect(FactoryGirl.build(:user, name: “Same Name”)).to be\_false end it “is invalid without an email” do expect(FactoryGirl.build(:user, email: nil)).to be\_false end end For this example, there is an already valid user factory definition that is setting some default values. This particular scenario is testing that the User model can create a user. Then validates user creation fails on a few key Rails Model validators on this User model. Controller Tests Controller tests will be your bread and butter when it comes to testing all your conditional logic. With tests based on request type, user type, success/failure of Models and so on. Controller specs are a little bit slower than Model specs due to their use of the request/response cycle. Though are still considerably faster that feature specs. This makes it a great place to do the bulk of your testing. describe UsersController do describe “GET index” do it “assigns @users” do

  user = FactoryGirl.create(:user)

  get :index

  expect(assigns(:users)).to eq(\[user\])

end

it “renders the index template” do

  get :index

  expect(response).to render\_template(“index”)

end end describe “GET #show” do it “renders the #show view” do

  get :index, {id:user.id}

  response.should render\_template :show

end end end end In this test file, you will find that we are testing a page that would list the users. First, testing the action assigns the right output. After creating a user, and getting the :index action the controller will have an instance variable that equals an array of the one user from the test. The second test is simply testing that getting the :index action, renders the index template. Obviously, these tests are not as complex and are abbreviated to give you a taste of how easy controller testing can be. In a real world controller file, you would be testing every action. Testing its behavior and rendering. As well as each branch of the action such as success or failure of a database transaction. View Tests View specs are often overlooked by developers who test all aspects of the visual representation in feature tests. Feature tests are meant to give you the high-level behavior testing. However, they definitely should not be used to test every last detail of a server rendered template. Many times you will need to test different markup based on some sort of if statement or logic, that can be done in a view spec. FactoryGirl.define do factory :user doname “Same Name”email “ ”admin falseend test@user.com end <%- if @user.admin? %> <h1>Welcome back admin</h1> <%- else %> <h1>Welcome back pal</h1> <%- end %> describe “rendering homepage” do it “displays admin message” do assign(:user, FactoryGirl.create(:user, admin: true))

render :template => “home/index.html.erb”

expect(rendered).to match /admin/ end it “displays regular user message” do assign(:user, FactoryGirl.create(:user))

render :template => “home/index.html.erb”

expect(rendered).to match /pal/ end end In this example, we have a simple view template that renders a message in the markup. The noun is based on whether the user referenced in the user variable is an admin or not. Here, the view spec renders out the message with the user object created in each test(one admin, one not) and tests for the text with regular expressions. Feature Tests Finally, feature specs(tests) are a great way to test your entire application at once(or large parts of behavior). Using RSpec and Capybara, you will be able to write plain simple tests that will walk through your app as if you were an end user. describe “Signing Up”, :type => :feature do it “allows a user to sign up” dovisit root_path expect(page).to have_content ‘Sign Up’within(“form#user_new”) do fill_in ‘Name’, with: ‘John’fill_in ‘Email’, with: ‘user@example.com’fill_in ‘Password’, with: ‘password’ fill_in ‘Retype Password’, with: ‘password’endclick_button ‘Sign Up’expect(page).to have_content ‘Welcome to Sample App’ expect(page).to have_content ‘John’endend This test simulates a user visiting your application for the first time. Then seeing a sign-up page, filling it out and successfully accessing the application after signup. As you can see, Feature tests provide an amazing way to test at a high level, just as a user would be interacting with your application. However, with that comes the cost of the fact feature tests will run slower than all other types of tests. For that reason, it would be smart to not test every branch or possibility in your application. Using controller or other specs to test edge cases. Other Testing Tools Additionally, there a few extra tools I would like to go over and will be included in your application’s codebase. These tools will help with issues such as test cleanup. Another, being a test coverage completeness tool which includes a local web tool to see analysis file by file. Lastly, is a tool to easily mock external resources, such as third-party APIs. : Database Cleaner is a testing tool, that allows you to use different strategies to wipe the database to a clean state after every test. The importance here is that while Rails attempts to wrap their tests in transactions. Which prevent one tests database changes from affecting another’s tests. Yet, when using javascript drivers, those tests are performed in another thread. Meaning you will need another means to return the database to a pristine state, using truncation. Database Cleaner, in particular, will allow you to set the “cleaning” method based on test type. Database Cleaner : We will use this gem to calculate, display and investigate code coverage. This tool will use built in Ruby coverage information, along with merging the results of different tools’ and test types coverage. Afterward, it will generate the information, save it to a report and automatically create an HTML site for you to visit to look through the code coverage analysis. Simplecov : VCR allows you to “record” HTTP transactions and save their responses locally. This allows you to “playback” the response from a file, making it easy and fast to run tests that interact with external resources. Along with the speed increases, it allows for offline testing and not worrying about a third party change affecting your tests. However, I recommend you still perform some sort of testing on the actual external resource’s response. VCR CI and You The last thing to discuss in this chapter is Continuous Integration. Which means you will be continuously merging changes into a master branch, where various pipelines make work on the newly merged code. In most cases, this is a way to merge changes into one central branch. Then run the tests on an external system and using the test results to chose another action such as deploying or requesting a fix to the codebase. There are a few different tools one could use for CI. The first that may be recommended in . The great thing about Jenkins is that it is an open-source project and free. However, there is no hosted solution available through their site. Meaning, you will either have to build and maintain your Jenkins server or seek a service that would do so for you. Jenkins There is good news for those who do not want to maintain their own server for running continuous integration. There are quite a few services that run a CI service. Having integrations wall of the major repository services(GitHub, Bitbucket, etc). This means a simple push to your repository could trigger your test to run automatically on their service. Followed by a report with the results in any response medium you would like. Such as email, chat services, and others. Without any affiliation, the server I usually use for my applications is . I find their ease of use as well as their pricing(free for most of my purposes) to for my needs well. Other choices would be and . I would recommend checking these products out and choosing the one that fits you best. CircleCI TravisCI SemaphoreCI

Your Guide to testing in Ruby on Rails 5

The culture of any organization is shaped by the worst behavior the leader is willing to tolerate — Gruenter and Whitacker Leadership can broadly be distilled down to two essential things: and . Entire books have been written on each of those topics. There is depth and plenty of surface area to explore in each. For now, we will simply say that ‘influence’ is the ability to open people to ideas, conclusions, or directions they otherwise might not have reached, and ‘expectations’ is helping people have a standard of performance and building the accountability to achieve it. influence expectations The unfortunate reality of the tech industry is that it lacks quality leadership as the rule, not the exception. Culture, by nature, is harder to define and no simple blog definition will satisfy everyone. Even so, a useful framing for culture in a group (company, division, department, or team) is the collective actions, habits, behaviours, and reactions of everyone working in that group. Leaders are for the culture of their group. Said another way, the cultural brokenness of a group can in almost all cases be traced directly to poor leadership, either past or current. If we explore culture as programmable AI, we know that given some semi-sophisticated and consistent data, a quality ML system will produce good output. Give it garbage input, random bits, with some sloppy rules and you get out about what you would expect in that case. Culture is the output and leadership is the input + AI. One cannot expect good output from poor input or broken rules. accountable Leaders need to set the tone, direction, and boundaries if they care about healthy culture. And this is the crux: you have to care about culture. Good culture doesn’t just happen. Culture is messy. It’s hard to define, it wants to entropy or run off in alternate directions while you aren’t looking. If you don’t care about culture now, at some point you will, for all the wrong reasons. I urge leaders to care about the little details of culture early and do it on their time frame when they want to, not when they are forced to. Call to action Here are some simple actions to impact culture via leadership now. Values — Have, publish, and reinforce your values Expectations — set and be consistent in your expectations Behaviours — Reward positive behaviours, stop negative behaviours, remove behaviours if necessary (and early). Make sure both positive and negative behaviours can be traced back to values Live the example — Be what you want the org to be. Always. Require yourself to be the high bar, not the low bar. Embody your values Take feedback seriously — Listen for themes, correct when needed Listen more than you talk — Leading starts with listening, a lot Learn via listening — Listening leads to learning Stay positive, always — Be solution oriented, look for the good, re-frame and avoid negative spirals. Repeat — This list never ends, it’s a loop Ultimately the culture of a company is defined by the CEO and allowed to continue by the action/in-action of the company’s board (if they have one). It’s reinforced by folks brought into the org or put into place by the CEO and eventually other senior leaders or managers. However there are many leaders in an org and all of them are also accountable for the culture of their group. Yes, it would be hard to fight a toxic corporate culture, though it’s possible to be the lost island of positive culture in a sea of toxicity. It’s hard work and requires great leadership, but it’s possible. It’s also possible to have an amazing culture from the top and a group or two within that is not up to the corporate standard. Both scenarios exist. However it’s much more rare to see good local culture in a bad macro culture org. Culture is contagious, both good and bad. Closing thoughts Leaders have the ability to shape culture in amazing and subtle ways. If you routinely ignore your test suites, it becomes a part of your culture. Is that what you want? If you allow passive-aggressive comments in your pull requests, that becomes part of your culture. Is that what you want? If you allow a co-worker to verbally attack or insult someone else in front of their peers in a meeting, how you handle that influences culture. And if you happen to allow systematic harassment of any kind, it becomes part of your culture. . Fix it. Better yet, stop it before it starts. Be a leader. Care about your culture. It’s literally your job. You don’t want that Postscript — Nuts and Bolts of Leadership and Culture I wanted to keep the above short and consumable. There is much more to be said on the topic and I thought it important to add a slight more in-depth, yet still too short, postscript so as to not get lost in a follow-up post. Even so, this postscript is barely scratches the surface. Culture just is You cannot ignore culture. Culture in your group or company is a thing even if you never give it a second thought. Every place has culture. If it wasn't intentionally created, it will simply grow organically and without direction. With luck your culture could grow into something amazing, or more likely, devolves into something toxic. Recognizing and intentionally doing the important work is an core characteristic of leadership. To intentionally craft culture, start by establishing your organizational values. a Further, define success early and marry it with values to achieve success. Doing so allows you to build a sustainable culture and have success with it. Many times I’ve seen companies with toxic work cultures that are lauded as successful because of financial achievements. While companies that measure success purely in terms of profits or valuation often achieve short term goals quickly, their neglect of culture results in an organization with a shaky foundation that will threaten their long term financial goals. There is no doubt that long term outcomes benefit from culture. I argue that even short term outcomes benefit from culture. Why? Sustainability, even in the short term. Culture as reputation Ours is a small industry. Reputations matter in a variety of ways. At the end of the day, almost all decisions come down to a few simple things: risk, reward, and trust. We all have a list of companies, based on reputation, that we would never work for. We also have a list of companies, based on reputation, that we might consider but it would cost a lot more to actually make the move. And we have some we would take market or maybe even less than market, at a drop of the hat to be part of. What we are doing is weighing the risk/reward with how much we trust the company and what we think we are getting into. It’s quite simple. A toxic corporate culture is going to cost you money, talent, and opportunities. Leadership, if they aren’t asleep at the wheel, know this inside and out and should be working to create a place that attracts the type of talent the company needs and wants. Culture early If you are starting a company or group, keep culture in mind from the jump. Work on your values, write them down. Be intentional and explicit about performance. Be authentic with people. Review your progress, gaps and vulnerabilities on a regular basis. Repeat as you grow. Change where needed. Discard what no longer works. Keep getting better. Turning a Ship — How to change culture Books can be written on this. However, broad strokes culture can be changed with a few key things (not an exhaustive set, obviously) Values — Be explicit in who and what you are Standard of Performance — Have a standard of performance Expectations — Hold people accountable to your values and standard of performance Coaching/mentoring/training — Invest in your people Change — Make small changes (people, process, organization, projects) to get momentum, big changes (people, mission, organization) to swing directions and set tone Time — Changing culture takes time. The larger you are, the longer it is Culture at scale If you didn’t start early, culture at scale is going to be a nightmare. It will be (extremely) slow to change, cause churn, and will be a period of unease for folks making the transition. But it’s also possible to do. And it’s worth it still. At scale, culture is about leading by example, values (always values), standards of performance, accountability, and middle/line management. At scale, middle/line management are the most important cultural leaders in a company. It is literally where culture is made or broken. For this reason it’s critically important to hire, train, educate and mentor your middle/line managers well. It’s also critically important you set, maintain, and enforce a standard of performance with your managers. This means, among the harder parts of leadership, letting people go who do not fit or do not want to fit. Further, who you ask to become managers will signal to others what you implicitly value in leadership. It is a cultural signal. Take this seriously and remember your values. A few bad managers can sink an entire org. Final Thought Leadership is hard. And we have a lack of good leaders (never mind great leaders) which makes it even harder. We have few examples to hold up as true inspirations. For all the leaders and future leaders out there, here’s something I very much want us all to take to heart: We must earn the right every day to continue to do the job we were given. Our job is in service to our employees and our organizations. How we react to this will dictate if we are good or bad leaders. Good leaders take it as their responsibility to care for the people, the product(s) and the organization (department, company etc) at the same time. Achieving this balance is what true leaders do; doing it effectively makes a leader great. Any manager can be revenue driven and work people to death. Any manager can create a culture of comfort and get nothing done. Any manager can create an ‘us v them’ attitude for departments or divisions. Any manager can can foster an entitled or political or conniving or fearful environment. None of this is hard. It’s quite easy. These are sadly what we see to often in our industry. Leaders create cultures that welcome high achievers of all types, set standards of performance, and meaningful missions. And do this in a way that makes people feel valued, safe, and included. Leadership is guiding organizations to this place with intent and purpose. Leadership is focusing on what matters and helping others do the same. All this leads the culture of your organization. Culture starts with leadership.

Keep

Mind

On Leadership and Culture

The autonomous car era is upon us and accelerating at an astonishing rate. Where just several years ago the state of the art in “autonomous” tech was limited to so-called radar-based “active cruise control” systems, today Tesla, Mercedes-Benz, and Volvo all offer vehicles with a level of semi-autonomy that would be have been unthinkable just three or four years ago. Indeed, autonomous systems and technologies have advanced at a staggering rate: both the hardware with which vehicles see the world around them, as well as cloud-based software to help systems learn and improve over time by aggregating drivers’ journeys — or, in Mercedes’ case with the 2017 E-Class, being situationally aware of other 2017 E-Class vehicles — are poised to enable the first fully autonomous cars in customers’ hands by 2020 or so. This is an incredible, fantastic thing; a real feat of engineering prowess and determination, and it’s improving at an astonishing rate. There’s just one small problem though: no amount of clever computer image processing, LiDAR, and radar feedback, nor any infinitely large database of aggregated driving data will ever enable autonomous cars to “see” snow-covered roads; roads without adequate lane markings; or other situations in which even a human would struggle with sure footing. Put another way, technological innovation with respect to autonomous cars’ visual acuity will soon reach an impasse beyond which further optical fidelity will no longer be limited by the onboard hardware and software, but rather by the mundane eventuality of a road which, for all intents and purposes, is not visible: not to cameras, not to radar, not to anything. Fortunately, a new proposal known as AVRIS — Autonomous Vehicle Road Improvement and Standardization — seeks to remedy this looming technological impediment by doing for the 21st century’s autonomous vehicles what paved roads did for the 20th century’s first motor vehicles: it wants to make the roads easier for autonomous cars to negotiate, just as asphalt was an improvement over dirt paths 100 years ago. If you haven’t yet heard of AVRIS, there’s a good reason for it: it doesn’t exist. Not yet, anyway. In fact, it’s just a little something I’ve come up with; and, if I’m honest, I’m not entirely thrilled with the acronym, but it’ll do for now. So this new proposal — proposal, really — is simple: the USDOT needs to roll out new standards for road design, including and especially with respect to markings — or, more accurately, technology-enhanced markings — to ensure that lanes and lane markings are always clearly visible to autonomous cars, no matter the driving conditions. my Two general points come readily to mind: First, there needs to be a far higher standard of maintenance for all roads: lane markings must not be allowed to fade, and must be clearly visible at all times. For roads paved with concrete instead of asphalt, lane markings must be in a contrasting color: white lane markings on gray concrete will not suffice. This higher standard must be applied to any and all types of roads, from local neighborhood streets, to multilane boulevards and avenues, to freeways. For roads in areas that even just occasionally receive snow, or even thick fog, additional technologies must be implemented to render visible an otherwise invisible road surface. For the first point, I think this is a relatively trivial, not to mention obvious, partial solution: picture German freeways, and you’ve got some sort of an idea what I’m on about: impeccable road maintenance, with clearly and explicit lane markings that are refreshed the moment they start to fade. This will be expensive, but it will not be technologically challenging. Indeed, it could usher in an era of entirely new graphics design for roads and signs: still easily legible by humans, but with a priority in mind towards the cars’ benefit. Perhaps more contrasting colors; greater variety in color coding; more defined and better illustrated icons, then. It’s the second point, however, that will prove most daunting, and will be limited not just by time and money and pretty new graphics, but clever engineering. There needs to be a way to make roads and lanes visible even beneath several centimeters of snow. Some ideas come to mind. First, road shoulders could implement an electronic barrier of sorts, not dissimilar to how the glide slope and localizer work for at an airport runway: posts could be installed along shoulders that emit two radio frequencies — say a low and high frequency — and the car knows it’s at the shoulder when the car “sees” both frequencies roughly equal to one another. If the car sees more of, say, the higher frequency, then it’s falling into the shoulder. There’s probably better ways to accomplish this in 2017 than with 1960s aviation radio frequency tech, but at least this should serve as a starting point for discussion and to illustrate my point. instrument landings Lane markings, however, pose another altogether more challenging problem: obviously we can’t have obstructions in the middle of the road; no posts emitting radio waves, then. Lane markings could be wired up to produce either heat and/0r radio emissions of their own, effectively causing them to glow, at least in certain parts of the electromagnetic spectrum. If the emissions are heat, then the cars’ infrared cameras could (potentially) see the heating through the snow and ice; alternatively, if sufficiently strong, it could simply melt the snow and ice entirely. This idea isn’t entirely without . precedent (As an aside, this could be further assisted using solar panel road surfaces, with excess charge stored in giant capacitors and used to heat the road surface, with any remaining excess still sold back to the grid.) If the emissions are indeed some sort of electromagnetic waves, they would have to be sufficient to tunnel through a layer of snow and ice. It’s been a while since I’ve studied electricity and magnetism, but I suspect that microwaves may do the trick: the microwaves would (again) serve as a heating element which would be absorbed by the snow and ice, and thus heat them, and again, render them visible to infrared cameras (or, again, simply melt it altogether). I guess the main issue is whether the heat could be adequately confined to the narrow lane marking graphics and not overly dissipated into a useless blob of warmth. Perhaps thorough melting is, after all, the optimal solution. Maybe terahertz imaging, otherwise known as millimeter wave, is the solution? It is, after all, remarkably adept at visualizing (through) things with high liquid content; this is what the full-body scanners at airports use. In any case, the overarching goal must be to make otherwise invisible road surfaces nevertheless perfectly visible. This can, of course, only be accomplished by making the snow, ice, or other debris obstructing the road surface effectively transparent to whatever imaging technology the cars are using, whether it’s radar, LiDAR, infrared, or otherwise. I realize what I’m proposing is a fairly massive thing, a veritable overhaul of the entire United States road network. But once upon a time, not that long ago, we didn’t have any paved roads at all; today, we don’t have any autonomous-friendly, always-visible roads. If we were once able to overhaul our roads from dirt to asphalt, I don’t see anything (besides money and bureaucracy) that should prevent us going from asphalt to fully digital, as it were. As I’ve written and time , I have no doubt that autonomous cars will start to trickle into customers’ hands by 2020; will be mandated on certain freeways by the 2030s; constitute a majority of vehicular traffic by the 2050s-2060s; and fully replace non-autonomous cars by the 2070s-2080s. (And no, that ridiculous “trolly paradox” simply .) time again isn’t an issue My concern, however — due in no small part to a complete and utter lack of discourse on the matter, until, as far as I know, this essay now — is that such progress will eventually reach a peak and be severely hindered if we continue to focus only on the cars themselves, and not on the roads as well. Instrument landing systems for airplanes would have been impossible but for radio towers emitting their beautiful overlapping bubbles of 90Hz and 150Hz radio signals — no amount of fancy computer vision equipment could land a plane today without making those runways effectively visible even through pea soup fog — so it should be seen as a similarly vain effort to expect cars to drive on roads utterly invisible due to, say, snow and ice, absent some technology to augment the roads’ visibility in the first place. Put another way, giving a blind person vision is of little use if the pages of the book they’re trying to read are covered in dirt. A little known fact: I a California-licensed attorney; any fellow attorneys want to help me draft a bill and get this moving forward? NB. am

Waves

AVRIS, the USDOT’s bold initiative to upgrade roads for self-driving cars

This article is a part of series: Configuring Android Project Little Things That Matter Version Name & Code Static Code Analyses Tools Continuous Integration Everything we have discussed in current article is available in project. template gitignore When you create a new android project in it already comes up with file, but usually it does not contain all necessary rules. Android Studio gitignore To quickly generate and download files I recommend you to use site. Just type in necessary keywords like — , and click generate button. gitignore gitignore.io Android Intellij Check out file in template project. gitignore tools folder If you have some 3rd party , or which are related to your project don’t just drop them in root directory — it will create a mess. scripts rulesets other files (especially for those who are using Project View, not Android View) Try to create a folder (e.g. tools) and put all those files into this folder. Usually I put there custom , rules for and static code analysis tools like , , gradle scripts files proguard pmd findbugs lint . Check out folder in template project. tools flavors Flavors are used to create builds with different set-up. In most cases I immediately set-up two flavors — and which have different: dev prod applicationId versionCode / versionName server endpoints google services keys … <a href="https://medium.com/media/150ffe9e2eac6e98f04e23425871a4a7/href">https://medium.com/media/150ffe9e2eac6e98f04e23425871a4a7/href</a> Check out in template project. productFlavors keystore A is a binary file that contains one or more private keys used to sign your application. keystore When running or debugging your project from the , automatically signs your with a debug certificate generated by the . IDE Android Studio APK Android SDK tools There are several issues when using local debug : keystore expiration date of 365 days installing application from multiple computers requires uninstalling first google services require keystore SHA-1 fingerprint That’s why I usually generate debug and commit it to version control system. keystore <a href="https://medium.com/media/7d4160e8568fb62440027f97bd777ee0/href">https://medium.com/media/7d4160e8568fb62440027f97bd777ee0/href</a> Check out in template project. signingConfigs proguard On Android is used for three things: proguard shrink unused code — helps you to survive the 64k limit optimize code & APK obfuscate code — makes your APK difficult to reverse engineer The problem is that and significantly increase compilation time and make debugging harder. obfuscation code optimization That’s why it is better to use different rules for release and debug build: proguard rules-proguard.pro rules-proguard-debug.pro <a href="https://medium.com/media/cbe51bd728c582af5c1df256ac72b760/href">https://medium.com/media/cbe51bd728c582af5c1df256ac72b760/href</a> rules for build must have following lines to force ignore warnings, skip code obfuscation and optimization: Proguard debug proguard <a href="https://medium.com/media/c3d6ea14f4008e3de0f5ee2205dd4ae2/href">https://medium.com/media/c3d6ea14f4008e3de0f5ee2205dd4ae2/href</a> For version it will be much harder to setup rules, because almost every library has it’s own specific rules. Fortunately there is open source repository called — which contains rules for all major libraries. release proguard android-proguard-snippets proguard <a href="https://medium.com/media/7a10e36468e5b5690187ce6d02c1fcec/href">https://medium.com/media/7a10e36468e5b5690187ce6d02c1fcec/href</a> Check out and in template project. rules-proguard.pro rules-proguard-debug.pro strict mode Android helps you to detect different kinds of problems: StrictMode closable object is not closed file reading / network requests performed on main thread uri exposed … Whenever such problem is detected it can show you appropriate log or crash your application, depending on your configuration. I suggest you to turn it on only for build and use methods to detect all kind of problems. debug detectAll <a href="https://medium.com/media/e507e64b63df84012de67395421b43f3/href">https://medium.com/media/e507e64b63df84012de67395421b43f3/href</a> Here is example of log when you forgot to close : SQLiteCursor <a href="https://medium.com/media/865507608541eeb989184fb6a8612c92/href">https://medium.com/media/865507608541eeb989184fb6a8612c92/href</a> Check out in template project. StrictMode

Configuring Android Project — Little Things That Matter

PICO-8 This article is the final part in a series describing my real-time lighting engine for the , a cute fantasy console with limited horsepower. It works as a stand-alone article, but you still might want to if you haven’t read the previous entries yet. PICO-8 start from the beginning Last time we parted, we had a cute flickering with most of the features already in place, but one important component was still missing: real-time shadows. To add that missing component, we need to figure a few things out: how to represent shadows, how to calculate where they are and how to draw them efficiently. light We’re in luck, since that’s what this article is all about. Speaking volumes Every lit object casts a behind it. In the real world, this is caused by rays of light from the light source being blocked by the object. shadow Some rendering techniques, such as , create virtual light rays and simulate how they would behave. These simulated rays are blocked just like real ones would, so pretty shadows just happen for free. What doesn’t happen is any kind of reasonable performance — despite all the technological advances since it was invented, raytracing is still woefully slow and mostly inadequate for real-time rendering. raytracing The way GPUs actually do business is by concentrating on the objects themselves. There are no light rays in a modern rendering pipeline. Instead, we start from a pixel on the object’s surface and apply lighting to it based on it’s position relative to the light source. The downside of doing it this way is that effects created by bouncing light rays (like shadows and reflections) become much more difficult to achieve, as we no longer have an actual ray to work with. Two common ways to get around that are and . The latter approach is what people usually use these days, but both have their strengths, and shadow volumes have one important advantage — they adapt very easily to 2D and don’t really require off-screen buffers, which are hard to do on the PICO-8. shadow volumes shadow mapping Each yellow shape is a shadow volume for one of the spheres (image from ) English Wikipedia The idea behind shadow volumes is all in the name. For every object, we can imagine its shadow as a volume of space in which everything is darkened. These volumes are three-dimensional objects themselves, and the basic approach to using them looks like this: :generate a draw this to a special check each pixel against the to see if it's in shadow only if it's not for each lit object shadow volume shadow volume bufferwhen we draw actual objects later: buffer light it Explaining this in detail would be an , but the core idea we can take away from it is this: an object’s shadow can be represented as a shape. article on its own Flat press The two-dimensional equivalent of the shadow volume is actually a — a flat shape representing all the pixels in a given object’s shadow. shadow area A 2D object and the shadow “volume” behind it Calculating that area for a complex object can be troublesome, so we’ll use our favorite cop-out: we’ll break it down into simpler components. In the 3D world, this is usually achieved by dividing surfaces into triangles. It’s much easier to figure out the shadow volume of a triangle than, say, a teapot. The 2D equivalent of the triangle is the line segment. Just like we can break a 3D object’s surface into triangles, we can break any two-dimensional shape into its constituent edges. How the shadow volumes of the three segments add up The shadow volume of each segment can be calculated separately. The volumes of neighboring segments will be connected with each other, adding up to the shadow of the whole thing. Building up the walls Once we know that all shadow caster are going to be represented by line segments, we have to tell our engine where these lines are. The basic idea is like this: every object in the game (wall, pillar, enemy) is allowed to have a list of — lines casting shadows. These lines are represented by and , the endpoints of the segment itself, and a unit vector that points away from the wall and lets us know which side of the wall is the outside (if you’re vector-savvy: the wall’s ). walls s e d normal Walls and their d-vectors in a sample corridor. Once we have our shadow-casting structure, it’s time to make use of it. The original 3D algorithm uses a special off-screen buffer to be able to quickly figure out whether a pixel is inside a shadow volume. That’s not really an option on the PICO-8 and the alternative of doing a point-in-polygon collision check for each potentially-lit pixel doesn’t sound too promising either. Thankfully, we can use the shadow volumes in a simpler manner: we can just draw them in black on top of the rendered scene. For a single light source and pitch-black shadows, the end result will be the same, but rendering a polygon is going to be much more efficient than doing a check for every pixel we draw. After this simplification, the whole process of adding shadows to the engine boils down to a single loop at the very bottom of our render loop. figure out if its shadow is at allif it is, calculate the in solid black for each wall: visible shadow volume draw the shadow volume The first step is crucial, as it cuts down the number of shadows we need to draw significantly. This is shown in the image above, with the walls that we actually need to do anything about highlighted in yellow. There are just two rules: if the light source is behind the wall, no shadow needs to be drawn if both ends of the wall are outside the light’s maximum range, we can ignore that wall as well — the shadow wouldn’t be visible anyway What’s in a shadow Now that we know which walls can be safely ignored, our new mission is figuring out the shadow volumes for those that we do have to handle. Warning: this section may contain trace amounts of vector math. A shadow volume and the light rays bounding it (shown in red) From the illustration above, it’s evident that the shadow area of a single wall is some kind of polygon. The wall casting the shadow becomes one of the polygon’s edges, since it’s part of the shadow’s perimeter. Two more edges can be obtained by connecting the light source with each of the wall’s endpoints and projecting forward —we can actually think of those as light rays bounding our shadow. In theory, these rays go to infinity, but drawing infinite polygons works just as well as counting to infinity does. Fortunately, the light already has a finite range, represented by the handy clipping rectangle that contains all potentially-lit pixels. We can use this rectangle to truncate our unwieldy infinite polygon into a smaller, self-contained equivalent with no delusions of grandeur, using the intersection points of the light rays with the rectangle. p,rng = lgt.pos,lgt.rng s,e = wall.s, wall.e ds,de = s-p,e-p cs, ce=rng/max(abs(ds.x),abs(ds.y)),rng/max(abs(de.x),abs(de.y)) proj_s, proj_e=p+ds*cs, p+de*ce -- light position and range local -- wall endpoints local -- calculate light rays towards s and e local -- extend the rays until they intersect with the-- nearest boundary defined by light range-- (white points) local local This gives us four vertices which we could connect up into a polygon right now and get a reasonable shadow in this one simple case. By this point in the series though, we have already learned that the simple case is never the only one. Two situations in which four points aren’t enough — the regions in red would miss out on all the fun When the light rays intersect different edges of the rectangle, using just the four points isn’t going to cut it. To hit all shadowed pixels within the rectangle, we need to add more vertices —namely, the corners we would encounter by going from one intersection point to the other along the perimeter. We do this by checking which quadrant each intersection point belongs to and iterating clockwise until we get all the missing corners. qs,qe = quadrant(ps), quadrant(pe) (qs<qe) qs += 4 vertices = {s, e, proj_e} q = qe,qs-1 _-- corners are pre-stored in a table as [_±_1, (vertices, p+corners[q]*rng) (pts, proj_s) -- find the quadrants local -- make sure qe->qs is increasing if -- assemble vertices in clockwise order local for do ±_1] vectors-- we just have to translate/scale them to our rectangle add end add Drawn and ordered We have our shadow volume, but actually drawing it is another story. Most rasterizing algorithms deal only with triangles, but what we have can be anything from a quad to hexagon. The go-to solution for more amply-faced polygons is breaking them down into triangles first. This works great in GPU-world, since “rendering a triangle” is accelerated so much there that it’s basically free. On the PICO-8, the overhead of drawing a triangle is pretty significant, and that’s bad news for us and our hexagons, which consist of a total of four annoying triangles. Fortunately, there is a simple algorithm for rendering polygons regardless of the number of faces they have, as long as they are — which our shadow volumes are. convex The first step is going to be our usual fare when dealing with rendering a shape: we’re going to split the polygon into individual lines. This turns the problem of rendering a polygon into simply figuring out its extents line by line. ymin, ymax = ymin, ymax:xls[y] = find the leftmost x still inside the polygon at line yxrs[y] = find the rightmost x still inside the polygon at line y ymin, ymax:fill(xls[y], xrs[y], y) Y coordinates of the top/bottom of the polygonfor each y between for each y between The way polygons work, both extremes have to lie on one of the edges. This means we can take all the edges, go through all the pixels on each (as if we wanted to draw it as a line) and update the or table accordingly. The only problem is that we don’t know which table to update: we have no idea whether any given edge is on the left or right side of the polygon. xls xrs What is stuffily known as the clockwise winding order. The simplest way out of this conundrum is making an assumption about the vertices, namely that they always come in a defined order: either clockwise or counterclockwise. Once we do that, we can easily figure out which side an edge belongs to by looking at the direction it’s going. Our shadow volume code gives us clockwise-ordered points, so any edge going up (ending less than starting ) must be on the left side of the polygon. On the other hand, if an edge is going down, it must be on the right side. y y ngon(pts, ln) xls, xrs, npts = {}, {}, #pts-- update xls and xrs using each edge in turn i=1, npts ngon_edge(pts[i],pts[i%npts+1],xls, xrs) -- use the tables to draw each horizontal line y, xl (xls) xr = xrs[y]ln(xl, xr, y) function local for do end for in pairs do local end end In this code snippet, the “figure out minimum/maximum y” section from the pseudo-code is nowhere to be found. That’s because we get this part for free: the / tables will only contain an entry for the y coordinates our polygon actually hits. Instead of going through all 's in order, we can iterate on one of these tables directly. xls xrs y Inversion diversion Having gone through all that trouble, we’re finally ready to plug everything in. We set up the walls, plug the shadow rendering loop in, and… Well, we’re definitely seeing some shadows, so that’s a plus. The problem is that the shadow of each wall completely obscures the wall itself. This might work for some games, but we didn’t go to all the trouble of drawing a cool ancient temple tileset just to limit our rendering to the floors. We’d like to stop the walls from casting shadow on themselves. There are a few ways to achieve that, and the one we’re going to use is a dirty trick. We’re going to turn our walls inside out by reversing the vector, making all the shadow-casting segments face . Since we check what side of the wall the light is on, this will make only the back faces able to cast a shadow. The engine will behave as if a ray of light was only blocked whenever it a solid object, not when it tries to enter. d inward leaves Back-face shadows, in garish Technicolor This gets us what we wanted. The shadows from the back walls add up to the correct shape, but the interior of the walls is untouched. Hole in the whole Since it works so well for solid walls, let’s apply it to something more ancient-templey. Say, an etched pillar. The base behind the spire casts a shadow the spire. MC Escher approves. on Well, that’s still not it. The shadow cast by the base covers the part that should be high above it, completely ruining the illusion of depth. This is the second time vertical surfaces cause us pain, and also the second time we will get around this problem by using a simple trick. Objects with vertical components will define a special mask in their table, which is just a rectangle labeled “do not touch”. We can use this rectangle to cover the vertical spire and make sure it won’t get shadows drawn on top of it. hole walls We extend the polygon routine to accept such masks and take them into account when drawing. Each line, we will split the segment we’re drawing into a “before the hole” and “after the hole” section, and draw these instead — leaving the interior of the rectangle intact. MC Escher no longer approves. All’s well that ends well At this point, I‘m happy with how the shadows look. The engine can still draw nonsense at times, with an object shadowing something it really shouldn’t, but such situations are pretty rare and hard to notice when you have a whole game to care about. I’m pretty sure the only way to definitively fix these remaining issues would be to actually do it in 3D — and I have to leave something fun for future articles, don’t I? That concludes our four-part journey from a black screen to a workable lighting engine on the PICO-8. I hope you enjoyed it as much as I did. I had to omit of detail to finish this write-up before the next century comes calling, so if you have any questions or feedback, I’ll be happy to take it on . a lot Twitter Writing these has been equal parts great fun and absolute exhaustion, so I will probably take a short break from writing now. Still, do keep an eye out for my next article — we’ll talk about how having strict limitations (like PICO-8’s) can influence game design for the better. Until then, may you always finish what you start. Part 1 | Part 2 | Part 3 | Part 4 | Play the game

Polygon

Caster

Routine

Trace

Yellow

Lighting by hand #4: into the shadows

So I’ve decided to go ahead and write a about . It’s called and is available through and (more on that later). book JavaScript The JavaScript Way Leanpub GitHub But… Why? It’s true there are tons of learning resources about JS. However, many are outdated and teach obsolete practices. Many others cover advanced topics like design patterns, language internals, or frameworks that are of little interest to newcomers. I’m starting this project because I think there is room for a beginner-centric book teaching modern JavaScript. I hope to make it a useful companion for anyone wishing to (re)discover the many facets of JS. Design and content The book is targeted towards JavaScript newcomers: complete beginners with no programming experience whatsoever or “switchers” coming from another language background. It won’t be a in-depth study of JavaScript (which is , by the way). I’ll try to keep it as beginner-friendly as possible while covering some advanced concepts and good practices. already existing It was pretty obvious from the start that the book would be written entirely using the more future-proof and often easier to explain ES6/ES2015 standard. As I want to keep things simple, many tricky and primarily ES5-related topics like or won’t be covered. hoisting behavior this As a teacher myself, I know that real learning comes from practicing. Each book chapter will be accompanied by a series of exercises to put the newly acquired skills into practice. I also know that students don’t like to lose hours configuring their environment before getting their feet wet. So there won’t be any complex setup: throughout the book, all you’ll need is a JavaScript playground ( and are the recommended ones) for coding along. CodePen Gomix Publication After much afterthought, I’ve decided to self-publish my book through and to make its source code available on . Leanpub GitHub I’d like to reach as many people as possible, regardless of wealth. I also hope to leverage the open source collaborative model to improve and validate the content. The chosen license, Creative Commons , lets anyone share or build upon this work for non-commercial purposes, as long as it is properly attributed and using the same license. BY-NC-SA Let’s roll! Publishing a book is a lot of work, but also a pretty exciting journey. Thanks for reading this, and I hope you’ll soon join me on the JavaScript Way.

CodePen

Mozilla

Walk This (JavaScript) Way

All the major projects in deep-learning and neural networks are seeming to converge on Python as a glue language. As a long time Python engineer, I am utterly confused. I would think that there are better tools for this job. Python is slow and won’t be able to implement custom logic. Python is fairly hard to bind to from C++ or even C. Python is stuck in a gridlocked transition from version 2.7 to 3.0 And yet, here we are programming our GPUs with Tensorflow, Theano, Keras, Lasagne etc.. What in the world could have united so many projects onto this one language? Have the religious wars ended or are data-scientists more enlightened than the rest of us? Maybe someone can point me to some rationale or manifesto, because I am totally lost on this one. — Student

Glue

Why is Python so popular for data science?

rems and ems, and why you probably don’t need them

By Fabien Chraim, Co-founder and VP of R&D at Civil Maps Last week, we hosted some distinguished house guests. By now, our neighborhood is used to a steady stream of automotive execs and R&D types from all over the world, but this particular group was comprised of lawmakers and the mission was purely about learning and discussion. After Federal Secretary for Transport and Digital Infrastructure of Germany Alexander Dobrindt and his entourage landed at SFO, their first stop was our in Albany, CA. Arriving in a Mercedes sprinter van and a large black SUV, the sight of diplomats piling out of vehicles and journalists in our front yard drew a few residents out of their houses to check out what was happening. home-office Secretary Dobrindt’s representatives reached out a few weeks ago, expressing the desire to learn about Civil Maps in order to better understand the future of autonomous driving. Our company works on technology that is critical to this task; we give cognition to cars, enabling them to gain context about their driving environment and crowdsource dynamic 3D maps. While the group behaved mostly like good students, listening intently and asking many questions, they are actually policy with regards to autonomous vehicle legislation. leaders On January 25th, Germany became the first country to providing initial frameworks for autonomous and semi-autonomous vehicle research and driving. Mr. Dobrindt and his team, who drafted the legislation, wanted to reassure German automakers, who collectively , that federal policy would be supportive of their research initiatives. In the future, the Secretary says this will likely entail the use of a “black box” to record when driverless systems are active. pass legislation produced 8.6 percent of the world’s automobiles in 2016 “There are certain prerequisites that need to be met,” said Secretary Dobrindt told reporters who gathered at our house/office for the visit. “For example, the vehicle is equipped with a black box which can tell at all times whether the vehicle or the human was in control over the vehicle or if the computer was functioning well at all times.” After exchanging initial greetings on our driveway, the group piled into the house. Our conference room, formerly the master bedroom for past tenants at 1605 Sonoma Ave, was stuffed with journalists, policy makers, and a few members of our own team. It was my pleasure to lead the discussion covering these topics: What is — Civil Maps gives cars contextual awareness, enabling autonomous vehicles to make better decisions for safer driving and crowdsourcing of 3D map data. Cognition for Cars Making Sense of Sensor Data — From point cloud to localizing in six dimensions within five cm precision using Civil Maps’ Signature-based Localization s — Projecting 3D HD semantic map data (lane markings, signage, and etc.) into the field of view of the car’s sensors in real time. The goal: enabling rider trust, while facilitating the reliable navigation and safety of an autonomous vehicle. Augmented Reality Map Crowdsourcing of maps through connected vehicle swarms Potential Solutions to the High Cost of Autonomous Vehicle R&D — Low compute and cost-effective sensing equipment, (beta) Atlas DevKit How Simulation Engines Accelerate Development Working with automotive OEM Partners Globally After some Q&A, Mr. Dobrindt went outside to get a demo of our Augmented Reality Maps and localization technology. Using low processing power and highly advanced compression of sensor data in real time, the car localizes itself in six dimensions and is presented with a machine readable map. With this information, the car knows where it is and exactly what to focus on in the scene while driving. This clarity reduces the cognitive load on the vehicle. The demo shows what the car sees, which is only the information it needs for navigation and safety. Before the delegation left for its next destination, we gave them an ample supply of Civil Maps schwag and said our quick goodbyes. Their next stop that day would be the Stanford Center for Automotive Research. For their last day of the two-day trip, they were scheduled to visit Otto (Uber). Secretary Dobrindt and his group made a similar learning trip to the U.S. about three years ago. Back then, he visited Facebook and Google.

Autonomous Driving Regulators At Our Doorstep

CPython 3.6 is out, 3.7 branches are all setup and CPython is now on . Exciting times for the Python community. GitHub Python 2.7 was the last release of the 2.x series, so by Sir Bedevere logic, Python 4.0 will be the next major release after 3.7. ‘All wood burns,’ states Sir Bedevere. ‘Therefore,’ he concludes, ‘all that burns is wood.’ OK, so that is totally illogical, but if 4.0 were to come out soon- which features would you want to see? Before we start the list, it’s important to remind ourselves what is in the CPython project. Python is 3 things, a , a and an . Compare that with , which is just a language. It has standard library, and the Google v8 engine does the along with many other options. language standard library execution engine JavaScript no execution, Here’s my laundry list, it consists of mainly changes to the component of CPython, (aka the “runtime”). I believe that the standard library is feature packed and anything not found in the standard library is readily found on PyPi. The language doesn’t need any new operators or yet-another-way-of-formatting-strings, IMHO. execution engine 1. JIT as a first class feature JIT or Just-In-Time compilation, is a method of compiling the byte-codes as they happen, instead of all at once. It can lead to better performance, depending on the scenario. is an implementation of Python with its own JIT, is an implementation of Python using as a JIT compiler. Each of these compromises on functionality (or correctness of the language) over speed. PyPy Pyston LLVM Having a first-class JIT plugin architecture for CPython would bridge the performance and compatibility needs together and efforts spent on PyPy, Numba etc could be directed into the main CPython ecosystem. The good news? This is already accepted as 3, thanks to core developer Brett Cannon and his Microsoft colleague Dino Viehland. The API was added in 3.6, but we’re yet to see an implementation out there of the API. The Pyjion project still needs updates. PEP-52 2. A stable .0 release Python 3 release history Python 3.0 — December 3, 2008 Python 3.1 — June 27, 2009 Python 3.2 — February 20, 2011 Python 3.3 — September 29, 2012 Python 3.4 — March 16, 2014 Python 3.5 — September 13, 2015 Python 3.6 — December 23, 2016 Python 3 has been out for 9 years, but for the first 2 versions (3.0, 3.1) there were a number of serious bugs and security issues. Python 3.3 was the first release of the 3.x series to include fast types, virtualenv as a core function and namespace package support. decimal Today, most major packages on PyPi that support “Python 3”, like support >3.3. The early adopters that tried Python 3 between 2008 and 2012, may have walked away with the perception that it isn’t ready. It’s not possible to set requests first impressions twice. My biggest ask for 4, is that 4.0 is a major functioning release, not an experimental release. 3. Static type hinting Anyone following the Mypy project will have seen the BDFL’s involvement and endorsement, but Mypy is a static type hinter for Python. optional Mypy aims to combine the benefits of dynamic (or “duck”) typing and static typing. def fib(n: int) -> Iterator[int]:a, b = 0, 1while a < n:yield aa, b = b, a+b I would like to see this in Python 4.0, because with more and more data being consumed from REST-ful services in JSON, libraries like can be hugely useful in type-checking inputs. schema 4. A GPU story for multiprocessing Using and the library in 3.6 (and below) you can leverage the power of GPU processing for highly parallel computation. For Python to cement its position as the data science, machine learning King, native support for GPU parallel processing would be awesome. PyOpenCL multiprocessing Check out , which uses a seperate JIT engine for compilation to parallel-native threading but it’s specific to NumPy. Along with the PEP523 (JIT) support, a native compilation module could use CPython to have parallel execution in GPUs, of of the box. NumbaPro 5. More community contributions Ok, I said 4 but this isn’t really a feature. In 4.x I would love to see more and broader ranging support and contributions from the community. With CPython now being on GitHub, that can mean 1 of 2 things, the community argues about features on the issue tracker (see other core language projects) or, it means that more people raise pull-requests, fix bugs and improve Python platform support. Is that it? I think every major/language project has a “ ”, .NET was 1.1 to 2. , . change hump PHP was 5 to 7 Perl was 5 to 6 AngularJS 2? Once they get over this change hump, the community can settle down, accept minor losses in backward compatibility in the name of innovation and plan for change. If the community diverges, it typically tolls the end of the project until something better comes along. Python 4 doesn’t have to have the same huge number of breaking changes that 3 did, but those changes were needed to take Python forward into the future and now we can talk about fun things in 4 instead of still discussing string types. What about? Ternary operators? Like in C# 6.0. Python has as per , not likely to have a specific operator. ?? x if y else z PEP-308 Like my post? Please click the 💚 below. Also, feel free to comment with your own ideas.

HackerNoon Tech Stories Archive February 27th, 2017

@bob_buzzardKeir Bowden

@mikecorpMichael Kropivka

@TalkAboutJackThe JayPad Team

@harpermaddoxharpermaddox

@brentbloweBrent Lowe

@ronald_istosRonald Ashri

@AndrewDBassAndrew Bass, Digital Marketing Expert

@djangostarsDjango Stars

@nikasa1889Dang Ha The Hien

@hackernoon-archivesDominik Vacikar

@rob__raceRob Race

@jasonwarnerJason Warner

@MarcHoagMarc Hoag

@dmytrodanylykDmytro Danylyk

@krajzegJakub Wasilewski

@bpesquetBaptiste Pesquet

@andrew_luckerAndrew Lucker

@david.gilbertsonDavid Gilbertson

@CivilMapsCivil Maps

@anthony.p.shawAnthony Shaw

@bob_buzzard
Keir Bowden

@mikecorp
Michael Kropivka

@TalkAboutJack
The JayPad Team

@harpermaddox
harpermaddox

@brentblowe
Brent Lowe

@ronald_istos
Ronald Ashri

@AndrewDBass
Andrew Bass, Digital Marketing Expert

@djangostars
Django Stars

@nikasa1889
Dang Ha The Hien

@hackernoon-archives
Dominik Vacikar

@rob__race
Rob Race

@jasonwarner
Jason Warner

@MarcHoag
Marc Hoag

@dmytrodanylyk
Dmytro Danylyk

@krajzeg
Jakub Wasilewski

@bpesquet
Baptiste Pesquet

@andrew_lucker
Andrew Lucker

@david.gilbertson
David Gilbertson

@CivilMaps
Civil Maps

@anthony.p.shaw
Anthony Shaw