Optimity

optimity

Gradient Descent. Image taken from - https://community.deeplearning.ai/t/difference-between-rmsprop-and-adam/310187 What Is Gradient Descent? Gradient descent is like hiking downhill with your eyes closed, following the slope until you hit the bottom (or at least a nice flat spot to rest). Technically, it is a method to minimize an objective function F(θ), parameterized by a model’s parameters θ∈Rn, by updating them in the opposite direction of the gradient ∇F(θ). The size of each step is controlled by the learning rate α. Think of α as the cautious guide who ensures you don’t tumble down too fast. If this sounds like Greek, feel free to check out my previous article, Quick Glance At Gradient Descent In Machine Learning — I promise it’s friendlier than it sounds! 🙂 Circle back: An objective function is the mathematical formula or function that your model aims to minimize (or maximize, depending on the goal) during training. It represents a measure of how far off your model’s predictions are from the actual outcomes or desired values.
For example:In machine learning, it could be a loss function like Mean Squared Error (MSE) for regression tasks or Cross-Entropy Loss for classification tasks.The objective function maps the model’s predictions to a numerical value, where smaller values indicate better performance.In simpler terms:It’s like a “fitness tracker” for your model — it tells you how good or bad your model’s predictions are. During optimization, gradient descent helps adjust the model’s parameters θ to reduce this value step by step, moving closer to an ideal solution. Got it? 😄 Gradient Descent Variants Gradient descent isn’t a one-size-fits-all deal. It comes in three variants, each like a different hiker — some take the scenic route, others sprint downhill, and a few prefer shortcuts (like me 😅). These variants balance accuracy and speed, depending on how much data they use to calculate the gradient. 1. Batch Gradient Descent Batch gradient descent, the “all-or-nothing” hiker, uses the entire dataset to compute the gradient of the cost function: Imagine stopping to look 👀 at every rock 🪨, tree 🌴, and bird 🕊 before deciding where to place your foot 🦶🏽next. It’s thorough, but not ideal if your dataset is as big as, say, the Amazon rainforest 😩. It’s also not great if you need to learn on the fly — like updating your hiking route after spotting a bear 🐻‍❄️. Code Example: for i in range(nb_epochs):
    params_grad = evaluate_gradient(loss_function, data, params)
    params = params - learning_rate * params_grad Batch gradient descent shines when you have all the time in the world and a dataset that fits neatly into memory. It’s guaranteed to find the global minimum for convex surfaces (smooth hills) or a local minimum for non-convex surfaces (rugged mountains). 2. Stochastic Gradient Descent (SGD) SGD is the “impulsive” hiker who takes one step at a time based on the current terrain: It’s faster because it doesn’t bother calculating gradients for the entire landscape. Instead, it uses one training example at a time. While this saves time, the frequent updates can make SGD look like it’s zigzagging downhill, which can be both exciting and a little chaotic. 😅 Imagine updating your grocery list after each aisle — you get to the essentials faster, but your cart might look wild in the process. However, with a learning rate that slows down over time, SGD can eventually reach the bottom (or the best local minimum). Code Example: for i in range(nb_epochs):
    np.random.shuffle(data)
    for example in data:
        params_grad = evaluate_gradient(loss_function, example, params)
        params = params - learning_rate * params_grad 3. The Hero of the Day: Adam 🙌🏽 Now let’s talk about Adam — the “hiking guru” who combines the wisdom of Momentum and RMSprop. Adaptive Moment Estimation (Adam) is like having a smart guide who tracks the terrain and adjusts your steps based on past experiences and current conditions. It’s the go-to optimizer when you want to train neural networks and still have time for coffee. Why You’ll Love Adam Low Memory Requirements: It’s like carrying a lightweight backpack — efficient but still packed with essentials.


Minimal Hyperparameter Tuning: Adam works well out of the box, so you won’t need to fiddle with too many knobs (just keep an eye on the learning rate).


Practical Use: From improving product recommendations to recognizing images of cats on the internet, Adam powers machine learning systems that make your everyday tech smarter. How Adam Works Adam maintains moving averages of gradients (mt​) and squared gradients (vt​) to adapt learning rates for each parameter: Here’s the cool part: Adam corrects biases to ensure accuracy, and updates parameters using this formula: where m^t and v^t are bias-corrected estimates, and epsilon ϵ is a small number to prevent division by zero. Conclusion Adam is the Swiss Army knife of optimizers — versatile, efficient, and reliable. Whether you’re training neural networks to detect fraud or create next-gen chatbots, Adam helps you get there faster and with fewer headaches. So, embrace Adam, take confident steps, and enjoy the view from the summit of machine learning success! References: https://www.ceremade.dauphine.fr/\~waldspurger/tds/22_23_s1/advanced_gradient_descent.pdf https://www.geeksforgeeks.org/rmsprop-optimizer-in-deep-learning/ Gradient Descent. Image taken from - https://community.deeplearning.ai/t/difference-between-rmsprop-and-adam/310187 Gradient Descent. Image taken from - https://community.deeplearning.ai/t/difference-between-rmsprop-and-adam/310187 What Is Gradient Descent? Gradient descent is like hiking downhill with your eyes closed, following the slope until you hit the bottom (or at least a nice flat spot to rest). Technically, it is a method to minimize an objective function F(θ ), parameterized by a model’s parameters θ∈Rn, by updating them in the opposite direction of the gradient ∇F(θ). objective function F(θ The size of each step is controlled by the learning rate α. Think of α as the cautious guide who ensures you don’t tumble down too fast. If this sounds like Greek, feel free to check out my previous article, Quick Glance At Gradient Descent In Machine Learning — I promise it’s friendlier than it sounds! 🙂 Circle back: An objective function is the mathematical formula or function that your model aims to minimize (or maximize, depending on the goal) during training. It represents a measure of how far off your model’s predictions are from the actual outcomes or desired values. For example: In machine learning, it could be a loss function like Mean Squared Error (MSE) for regression tasks or Cross-Entropy Loss for classification tasks. The objective function maps the model’s predictions to a numerical value, where smaller values indicate better performance. In simpler terms: It’s like a “fitness tracker” for your model — it tells you how good or bad your model’s predictions are. During optimization, gradient descent helps adjust the model’s parameters θ to reduce this value step by step, moving closer to an ideal solution. Got it? 😄 Circle back: An objective function is the mathematical formula or function that your model aims to minimize (or maximize, depending on the goal) during training. It represents a measure of how far off your model’s predictions are from the actual outcomes or desired values. Circle back: objective function For example: In machine learning , it could be a loss function like Mean Squared Error (MSE) for regression tasks or Cross-Entropy Loss for classification tasks. machine learning loss function The objective function maps the model’s predictions to a numerical value , where smaller values indicate better performance. numerical value In simpler terms: It’s like a “fitness tracker” for your model — it tells you how good or bad your model’s predictions are. During optimization, gradient descent helps adjust the model’s parameters θ to reduce this value step by step, moving closer to an ideal solution. Got it? 😄 Gradient Descent Variants Gradient descent isn’t a one-size-fits-all deal. It comes in three variants, each like a different hiker — some take the scenic route, others sprint downhill, and a few prefer shortcuts (like me 😅). These variants balance accuracy and speed, depending on how much data they use to calculate the gradient. 1. Batch Gradient Descent Batch gradient descent, the “all-or-nothing” hiker, uses the entire dataset to compute the gradient of the cost function: Imagine stopping to look 👀 at every rock 🪨, tree 🌴, and bird 🕊 before deciding where to place your foot 🦶🏽next. It’s thorough, but not ideal if your dataset is as big as, say, the Amazon rainforest 😩. It’s also not great if you need to learn on the fly — like updating your hiking route after spotting a bear 🐻‍❄️. Code Example: Code Example: for i in range(nb_epochs):
    params_grad = evaluate_gradient(loss_function, data, params)
    params = params - learning_rate * params_grad for i in range(nb_epochs):
    params_grad = evaluate_gradient(loss_function, data, params)
    params = params - learning_rate * params_grad Batch gradient descent shines when you have all the time in the world and a dataset that fits neatly into memory. It’s guaranteed to find the global minimum for convex surfaces (smooth hills) or a local minimum for non-convex surfaces (rugged mountains). 2. Stochastic Gradient Descent (SGD) SGD is the “impulsive” hiker who takes one step at a time based on the current terrain: It’s faster because it doesn’t bother calculating gradients for the entire landscape. Instead, it uses one training example at a time. While this saves time, the frequent updates can make SGD look like it’s zigzagging downhill, which can be both exciting and a little chaotic. 😅 Imagine updating your grocery list after each aisle — you get to the essentials faster, but your cart might look wild in the process. However, with a learning rate that slows down over time, SGD can eventually reach the bottom (or the best local minimum). Code Example: Code Example: for i in range(nb_epochs):
    np.random.shuffle(data)
    for example in data:
        params_grad = evaluate_gradient(loss_function, example, params)
        params = params - learning_rate * params_grad for i in range(nb_epochs):
    np.random.shuffle(data)
    for example in data:
        params_grad = evaluate_gradient(loss_function, example, params)
        params = params - learning_rate * params_grad 3. The Hero of the Day: Adam 🙌🏽 Now let’s talk about Adam — the “hiking guru” who combines the wisdom of Momentum and RMSprop . Adaptive Moment Estimation (Adam) is like having a smart guide who tracks the terrain and adjusts your steps based on past experiences and current conditions. It’s the go-to optimizer when you want to train neural networks and still have time for coffee. Momentum RMSprop Why You’ll Love Adam Low Memory Requirements: It’s like carrying a lightweight backpack — efficient but still packed with essentials. Minimal Hyperparameter Tuning: Adam works well out of the box, so you won’t need to fiddle with too many knobs (just keep an eye on the learning rate). Practical Use: From improving product recommendations to recognizing images of cats on the internet, Adam powers machine learning systems that make your everyday tech smarter. Low Memory Requirements: It’s like carrying a lightweight backpack — efficient but still packed with essentials. Low Memory Requirements: It’s like carrying a lightweight backpack — efficient but still packed with essentials. Low Memory Requirements: Minimal Hyperparameter Tuning: Adam works well out of the box, so you won’t need to fiddle with too many knobs (just keep an eye on the learning rate). Minimal Hyperparameter Tuning: Adam works well out of the box, so you won’t need to fiddle with too many knobs (just keep an eye on the learning rate). Minimal Hyperparameter Tuning: Practical Use: From improving product recommendations to recognizing images of cats on the internet, Adam powers machine learning systems that make your everyday tech smarter. Practical Use: From improving product recommendations to recognizing images of cats on the internet, Adam powers machine learning systems that make your everyday tech smarter. Practical Use: How Adam Works Adam maintains moving averages of gradients (mt​) and squared gradients (vt​) to adapt learning rates for each parameter: Here’s the cool part: Adam corrects biases to ensure accuracy, and updates parameters using this formula: where m^t and v^t are bias-corrected estimates, and epsilon ϵ is a small number to prevent division by zero. Conclusion Adam is the Swiss Army knife of optimizers — versatile, efficient, and reliable. Whether you’re training neural networks to detect fraud or create next-gen chatbots, Adam helps you get there faster and with fewer headaches. So, embrace Adam, take confident steps, and enjoy the view from the summit of machine learning success! References: https://www.ceremade.dauphine.fr/\~waldspurger/tds/22_23_s1/advanced_gradient_descent.pdf https://www.ceremade.dauphine.fr/\~waldspurger/tds/22_23_s1/advanced_gradient_descent.pdf https://www.geeksforgeeks.org/rmsprop-optimizer-in-deep-learning/ https://www.geeksforgeeks.org/rmsprop-optimizer-in-deep-learning/

Extending Stochastic Gradient Optimization with ADAM

Over-optimized loops hurt the eyes TL;DR: Don't optimize loops without a clear need and concrete real-world evidence Problems Premature Optimization
Reduced readability
Increased complexity
Difficult to maintain
Slower debugging Solutions Keep it simple
Prioritize clarity
Avoid premature tweaks
Refactor when needed Context You might think optimizing every loop will improve performance, but this approach backfires when you sacrifice clarity for unproven gains. Writing complex code to avoid hypothetical slowdowns often makes it hard for others (and your future self) to understand or debug your code. It would be best if you prioritized readability. Keep loops simple and only optimize when you know a bottleneck exists in real usage scenarios. Sample Code Wrong # Over-optimized and less readable
result = [item.process() for item in items if item.is_valid()] Right # Clearer and easier to understand
result = []
for item in items:
    if item.is_valid():
        result.append(item.process()) Detection [x]Semi-Automatic Look for list comprehensions or complex loop structures that optimize performance without real performance benchmark evidence. Exceptions Concrete evidence on mission-critical algorithms Tags Premature Optimization Level [x]Intermediate AI Generation AI tools often prioritize functional correctness so that they might produce clean, simple loops. if you prompt AI for performance at all costs, it could create over-optimized code even for straightforward tasks. AI Detection With proper instructions to stress readability and maintainability, AI can detect and fix this smell by simplifying loops and choosing clarity over premature optimization. Try Them! Remember: AI Assistants make lots of mistakes Without Proper Instructions

With Specific Instructions



ChatGPT

ChatGPT



Claude

Claude



Perplexity

Perplexity



Copilot

Copilot



Gemini

Gemini Conclusion Don’t sacrifice readability by optimizing too early. You can optimize later if a loop becomes a proven bottleneck. Until then, clear and simple code will save time, reduce bugs, and make it more maintainable. Related Reading https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-iv-7sc3w8n https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-xxvi https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-ii-o96s3wl4 Disclaimer: Code Smells are my opinion. Credit(s): Photo by Tine Ivanič on Unsplash More computing sins are committed in the name of efficiency without necessarily achieving it than for any other single reason. W. A. Wulf https://hackernoon.com/400-thought-provoking-software-engineering-quotes?embedable=true This article is part of the CodeSmell Series: How to Find the Stinky Parts of your Code Over-optimized loops hurt the eyes Over-optimized loops hurt the eyes TL;DR: Don't optimize loops without a clear need and concrete real-world evidence TL;DR: Don't optimize loops without a clear need and concrete real-world evidence Problems Premature Optimization Reduced readability Increased complexity Difficult to maintain Slower debugging Premature Optimization Premature Optimization Reduced readability Increased complexity Difficult to maintain Slower debugging Solutions Keep it simple Prioritize clarity Avoid premature tweaks Refactor when needed Keep it simple Prioritize clarity Avoid premature tweaks Refactor when needed Context You might think optimizing every loop will improve performance, but this approach backfires when you sacrifice clarity for unproven gains. Writing complex code to avoid hypothetical slowdowns often makes it hard for others (and your future self) to understand or debug your code. It would be best if you prioritized readability. Keep loops simple and only optimize when you know a bottleneck exists in real usage scenarios. hypothetical real usage Sample Code Wrong # Over-optimized and less readable
result = [item.process() for item in items if item.is_valid()] # Over-optimized and less readable
result = [item.process() for item in items if item.is_valid()] Right # Clearer and easier to understand
result = []
for item in items:
    if item.is_valid():
        result.append(item.process()) # Clearer and easier to understand
result = []
for item in items:
    if item.is_valid():
        result.append(item.process()) Detection [x]Semi-Automatic [x] Semi-Automatic [x] Look for list comprehensions or complex loop structures that optimize performance without real performance benchmark evidence. Exceptions Concrete evidence on mission-critical algorithms Concrete evidence on mission-critical algorithms Tags Premature Optimization Premature Optimization Level [x]Intermediate [x] Intermediate [x] AI Generation AI tools often prioritize functional correctness so that they might produce clean, simple loops. if you prompt AI for performance at all costs, it could create over-optimized code even for straightforward tasks. AI Detection With proper instructions to stress readability and maintainability, AI can detect and fix this smell by simplifying loops and choosing clarity over premature optimization. Try Them! Try Them! Remember: AI Assistants make lots of mistakes Remember: AI Assistants make lots of mistakes Without Proper Instructions

With Specific Instructions



ChatGPT

ChatGPT



Claude

Claude



Perplexity

Perplexity



Copilot

Copilot



Gemini

Gemini Without Proper Instructions

With Specific Instructions ChatGPT

ChatGPT Claude

Claude Perplexity

Perplexity Copilot

Copilot Gemini

Gemini Without Proper Instructions With Specific Instructions Without Proper Instructions Without Proper Instructions With Specific Instructions With Specific Instructions ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT ChatGPT Claude Claude Claude Claude Claude Claude Claude Claude Perplexity Perplexity Perplexity Perplexity Perplexity Perplexity Perplexity Perplexity Copilot Copilot Copilot Copilot Copilot Copilot Copilot Copilot Gemini Gemini Gemini Gemini Gemini Gemini Gemini Gemini Conclusion Don’t sacrifice readability by optimizing too early. You can optimize later if a loop becomes a proven bottleneck. Until then, clear and simple code will save time, reduce bugs, and make it more maintainable. Related Reading https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-iv-7sc3w8n https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-iv-7sc3w8n https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-xxvi https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-xxvi https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-ii-o96s3wl4 https://hackernoon.com/how-to-find-the-stinky-parts-of-your-code-part-ii-o96s3wl4 Disclaimer: Code Smells are my opinion. Disclaimer: Code Smells are my opinion . opinion Credit(s): Photo by Tine Ivanič on Unsplash Credit(s): Photo by Tine Ivanič on Unsplash Tine Ivanič Unsplash More computing sins are committed in the name of efficiency without necessarily achieving it than for any other single reason. More computing sins are committed in the name of efficiency without necessarily achieving it than for any other single reason. W. A. Wulf W. A. Wulf https://hackernoon.com/400-thought-provoking-software-engineering-quotes?embedable=true https://hackernoon.com/400-thought-provoking-software-engineering-quotes?embedable=true This article is part of the CodeSmell Series: How to Find the Stinky Parts of your Code This article is part of the CodeSmell Series: How to Find the Stinky Parts of your Code How to Find the Stinky Parts of your Code

Code Smell 279 - Loop Premature Optimization

Primer on Large Language Model (LLM) Inference Optimizations: 3. Model Architecture Optimizations

Web Scraping Optimization: Tips for Faster, Smarter Scrapers

Web Scraping Optimization: Daha Sürətli, Daha Ağıllı Scrapers üçün Məsləhətlər

Optimización del web scraping: consejos para que los scrapers sean más rápidos e inteligentes

Web スクレイピングの最適化: より高速でスマートなスクレイパーのためのヒント

Žiniatinklio grandymo optimizavimas: greitesnių ir išmanesnių grandiklių patarimai

Fanamafisana ny fakosoka amin'ny Internet: Toro-hevitra ho an'ny mpikirakira haingana kokoa

वेब स्क्र्यापिंग अप्टिमाइजेसन: छिटो, स्मार्ट स्क्र्यापरहरूको लागि सुझाव

Optymalizacja Web Scraping: Wskazówki dotyczące szybszych i inteligentniejszych Scraperów

Gukuramo Urubuga Optimisiyoneri: Inama zo Kwihuta, Byoroheje

Оптимизација веб-стругања: савети за брже и паметније стругаче

Оптимизатсияи веб скрепинг: Маслиҳатҳо барои скреперҳои тезтар ва оқилтар

Web Scraping Optimization: Mga Tip para sa Mas Mabilis, Mas Matalino na mga Scraper

I-Web Scraping Optimization: Amathiphu Okwenza Ama-Scrapers Asheshayo, Ahlakaniphile

We’ve all seen significant changes in user acquisition (UA) advertising in the last few years. But we still see that for channels like Facebook, TikTok, etc. “Creative is king.” Despite this, many teams still face challenges with systematically working processes with creatives, struggling to catch users’ attention and win the auction. Based on my experience in Scentbird, Praktika, and Playrix, the performance of top creatives may 5 times differ from low creatives’ performance, and “good” creatives live longer in rotation and burn out slower. Let's discuss how we can think of creatives based on the DV360 example. Frequency is one of the most critical factors for campaign success. For example, Facebook says ~2 - 2.5 times per week is optimal; higher numbers won’t be adequate and only add extra cost. Others say that 1.8 - 4 frequency is okayish, and a higher frequency is ineffective. But can we use this approach in other channels? Can we use it as a “rule of thumb” before analyzing? Our content and creatives may meet potential users in different mindsets and environments. They may scroll through their Facebook feed in the middle of the working day or on the phone while moving from one place to another. At the same time, our users may watch YouTube on their laptops while cooking, cleaning the house, or before sleep. DV360 provides many more formats, such as CTV,out-of-home, in-game interstitials, small banners, etc., and all of these are genuinely different contexts. Considering this rich inventory, can we even think of creatives for all these placements similarly? The most challenging thing for the marketer here is to see DV360 not line one channel but like a constellation of channels in one interface. What is DV360 DV360, or Display & Video 360, is a comprehensive advertising platform developed by Google. It enables advertisers and marketers to plan, manage, and optimize their display, video, and programmatic campaigns across various online channels. DV360 offers tools for audience targeting, creative management, and real-time bidding to help businesses reach their desired audiences effectively through digital advertising. Some may say that it’s just another way to buy YouTube ads, but it’s much more than that. The main feature of DV360 is the flexible frequency setting on each level of an advertising campaign, and we won’t find it in Google UAC. But besides that, there are a vast number of different inventories, like banners, video, native banners and video, YouTube, CTV, and out-of-door. Let’s discuss one at a time. Banners There are an excessive number of available banner formats. How can we determine which to use and which to skip? First, you can consider the most popular formats. And then run a test flight. After this flight, you can create a report where Google will show you particular formats your users may see depending on where your target audience can be found and which devices they have. Then, you’ll be able to score the formats by potential reach and make an informed decision about which formats you should remake (not resize) your creatives. The more appropriate sizes you’ll cover, the more “quality reach” you’ll have. What do I mean by “quality reach”? This won’t be a surprise if you see the most potential reach for small banners like 728х90. This banner is relatively small and can appear almost on any resource. This is both an advantage and a disadvantage of such formats. The advantage is that using this format is the way to reach your users when there is no opportunity to reach them with any other ads. This is the case, especially when reach is important or a channel KPI. The disadvantage is that if you put it along with the other banners, it’ll take the most impressions (and therefore the budget). Since it has relatively low performance compared to other formats, your campaign may have low performance at first glance. So, how do you have a foot in both camps? The solution is to place small creatives to separate Line Items. Here, Google DV360's remarkable ability will help us. You can set low frequency and low budget, particularly for these types of banners. For example, you set frequency 4 for the whole campaign but limit frequency setting to this particular Line Item to 1. (The lower the setting in the campaign hierarchy always overrides the higher). Thus, if your users were captured in several formats, they’ll see a thin banner 1 time, 3 other times, and other more prominent banners. But if your user only appears in this particular format, you’ll reach them at least once. This workaround works excellently for Google DV360, unlike other DSPs or Google Ads Campaigns, where it’s more challenging or even impossible to separate creatives this way. Also, it may be helpful to make these thin banners animated to capture the attention. Video There are a lot of placemats available for video on Google DV360. It has Video, Native Video, YouTube, YouTube CTV, or even Out of Home. There is a trick here. When you launch a video on Facebook, even if it will run on different placements, it’s still in a single Facebook context, either mobile or desktop. This is not the case for DV360. For example, when you run Native Video in the game, advertising another game, the Native Video takes all available space on the screen, capturing undivided user attention. Also, there is a button that the user may click on if they are interested. Imagine you run the same video on CTV. The user is watching TV and suddenly sees a video from the game: absolutely another context, no device at hand, no button to click. Even if they are interested, do they have to remember the brand somehow, find the phone, and find the app in the store? Or what? No doubt, many teams are struggling to test the CTV because other creatives are failing, and producing different creative concepts takes effort, time, and money. So, how do we think of video creative for Google DV360? Let’s proceed with the game ad example. First of all, we should consider the context. What is the user doing when they see our ads? If you think about the Native Video format, you can group the game publishers in separate Line Items, and you know that the user is playing; you can provide a gameplay video or even a playable ad so that the experience will feel seamless. If you think about the YouTube or CTV format, showcasing the brand and using a game story is better. For example, your game characters do something funny or exciting, or there is a mystery there, you name it. Thus, the user will be impressed with another story and may remember your app. Then, you can separate the users who saw one part of your game story on YouTube and show them another part of it… the whole storytelling. Yep, Google DV360 supports even that. Also, imagine that if the user is impressed after seeing your ads during movie night on CTV, they still have to pause the movie (which is only sometimes nice or possible), find their phone, etc. So? You may want to help them remember your brand better by increasing the frequency! Yes, “normal” frequency 2-3-4 is not the case here. You can show your ad significantly more times! Will it work, and will it be profitable? The answers are different for different businesses and products. In such situations, you can perform a test. You can divide your audience into groups, set different frequencies for each group, and determine the most profitable comparing costs to projected LTV. Yes, it’s quite an expensive study that requires a decent investment, both financial and FTE, from the analytics team. The potential outcome is that you may unlock a great scalable channel. The context is much more important when you test the Out-of-Home format when you should consider not only the user context from his activity perspective but also the surroundings - district, buildings, events, time of the year, and much more. And here, the frequency may be even greater. Final words Google DV360 is a very controversial channel for media buyers. Besides Google, it charges by impressions and not by clicks, so guys who ran Google Ads Campaigns for a while initially feel quite uncomfortable. Media buyers who run Facebook feel uncomfortable with the interface. They call it a DSP but with YouTube and CTV… What is it finally? The entire campaign structure is so unusual that you’ll need at least a month to wrap your head around it. And after all, there are different creative strategies and contexts. If you are in a mature company with a solid product-market fit and resources and looking for your successive big win, maybe it’s an option. At least give it a chance. We’ve all seen significant changes in user acquisition (UA) advertising in the last few years. But we still see that for channels like Facebook, TikTok, etc. “ Creative is king .” Despite this, many teams still face challenges with systematically working processes with creatives, struggling to catch users’ attention and win the auction. Creative is king Based on my experience in Scentbird, Praktika, and Playrix, the performance of top creatives may 5 times differ from low creatives’ performance, and “good” creatives live longer in rotation and burn out slower. Let's discuss how we can think of creatives based on the DV360 example. Frequency is one of the most critical factors for campaign success. For example, Facebook says ~2 - 2.5 times per week is optimal; higher numbers won’t be adequate and only add extra cost. Others say that 1.8 - 4 frequency is okayish, and a higher frequency is ineffective. Facebook Facebook Others say Others say But can we use this approach in other channels? Can we use it as a “rule of thumb” before analyzing? Our content and creatives may meet potential users in different mindsets and environments. They may scroll through their Facebook feed in the middle of the working day or on the phone while moving from one place to another. At the same time, our users may watch YouTube on their laptops while cooking, cleaning the house, or before sleep. DV360 provides many more formats, such as CTV,out-of-home, in-game interstitials, small banners, etc., and all of these are genuinely different contexts. Considering this rich inventory, can we even think of creatives for all these placements similarly? The most challenging thing for the marketer here is to see DV360 not line one channel but like a constellation of channels in one interface. What is DV360 DV360, or Display & Video 360, is a comprehensive advertising platform developed by Google. It enables advertisers and marketers to plan, manage, and optimize their display, video, and programmatic campaigns across various online channels. DV360 offers tools for audience targeting, creative management, and real-time bidding to help businesses reach their desired audiences effectively through digital advertising. Some may say that it’s just another way to buy YouTube ads, but it’s much more than that. The main feature of DV360 is the flexible frequency setting on each level of an advertising campaign, and we won’t find it in Google UAC. But besides that, there are a vast number of different inventories, like banners, video, native banners and video, YouTube, CTV, and out-of-door. Let’s discuss one at a time. Banners There are an excessive number of available banner formats. How can we determine which to use and which to skip? First, you can consider the most popular formats. And then run a test flight. After this flight, you can create a report where Google will show you particular formats your users may see depending on where your target audience can be found and which devices they have. Then, you’ll be able to score the formats by potential reach and make an informed decision about which formats you should remake (not resize) your creatives. The more appropriate sizes you’ll cover, the more “quality reach” you’ll have. What do I mean by “quality reach”? This won’t be a surprise if you see the most potential reach for small banners like 728х90. This banner is relatively small and can appear almost on any resource. This is both an advantage and a disadvantage of such formats. The advantage is that using this format is the way to reach your users when there is no opportunity to reach them with any other ads. This is the case, especially when reach is important or a channel KPI. The disadvantage is that if you put it along with the other banners, it’ll take the most impressions (and therefore the budget). Since it has relatively low performance compared to other formats, your campaign may have low performance at first glance. So, how do you have a foot in both camps? The solution is to place small creatives to separate Line Items. Here, Google DV360's remarkable ability will help us. You can set low frequency and low budget, particularly for these types of banners. For example, you set frequency 4 for the whole campaign but limit frequency setting to this particular Line Item to 1. (The lower the setting in the campaign hierarchy always overrides the higher). Thus, if your users were captured in several formats, they’ll see a thin banner 1 time, 3 other times, and other more prominent banners. But if your user only appears in this particular format, you’ll reach them at least once. This workaround works excellently for Google DV360, unlike other DSPs or Google Ads Campaigns, where it’s more challenging or even impossible to separate creatives this way. Also, it may be helpful to make these thin banners animated to capture the attention. Video There are a lot of placemats available for video on Google DV360. It has Video, Native Video, YouTube, YouTube CTV, or even Out of Home. There is a trick here. When you launch a video on Facebook, even if it will run on different placements, it’s still in a single Facebook context, either mobile or desktop. This is not the case for DV360. For example, when you run Native Video in the game, advertising another game, the Native Video takes all available space on the screen, capturing undivided user attention. Also, there is a button that the user may click on if they are interested. Imagine you run the same video on CTV. The user is watching TV and suddenly sees a video from the game: absolutely another context, no device at hand, no button to click. Even if they are interested, do they have to remember the brand somehow, find the phone, and find the app in the store? Or what? No doubt, many teams are struggling to test the CTV because other creatives are failing, and producing different creative concepts takes effort, time, and money. So, how do we think of video creative for Google DV360? Let’s proceed with the game ad example. First of all, we should consider the context. What is the user doing when they see our ads? If you think about the Native Video format, you can group the game publishers in separate Line Items, and you know that the user is playing; you can provide a gameplay video or even a playable ad so that the experience will feel seamless. If you think about the YouTube or CTV format, showcasing the brand and using a game story is better. For example, your game characters do something funny or exciting, or there is a mystery there, you name it. Thus, the user will be impressed with another story and may remember your app. Then, you can separate the users who saw one part of your game story on YouTube and show them another part of it… the whole storytelling. Yep, Google DV360 supports even that. Also, imagine that if the user is impressed after seeing your ads during movie night on CTV, they still have to pause the movie (which is only sometimes nice or possible), find their phone, etc. So? You may want to help them remember your brand better by increasing the frequency! Yes, “normal” frequency 2-3-4 is not the case here. You can show your ad significantly more times! Will it work, and will it be profitable? The answers are different for different businesses and products. In such situations, you can perform a test. You can divide your audience into groups, set different frequencies for each group, and determine the most profitable comparing costs to projected LTV. Yes, it’s quite an expensive study that requires a decent investment, both financial and FTE, from the analytics team. The potential outcome is that you may unlock a great scalable channel. The context is much more important when you test the Out-of-Home format when you should consider not only the user context from his activity perspective but also the surroundings - district, buildings, events, time of the year, and much more. And here, the frequency may be even greater. Final words Google DV360 is a very controversial channel for media buyers. Besides Google, it charges by impressions and not by clicks, so guys who ran Google Ads Campaigns for a while initially feel quite uncomfortable. Media buyers who run Facebook feel uncomfortable with the interface. They call it a DSP but with YouTube and CTV… What is it finally? The entire campaign structure is so unusual that you’ll need at least a month to wrap your head around it. And after all, there are different creative strategies and contexts. If you are in a mature company with a solid product-market fit and resources and looking for your successive big win, maybe it’s an option. At least give it a chance.

Optimizing User Acquisition with Google DV360: Creative Strategies for Success

Authors:
(1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu);
(2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu);
(3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4.2 Determined versus sender-designed experiments Given the model, the sender can manipulate the phase-K interim belief only when designing an experiment at phase K − 1. If an experiment at phase K − 1 is determined, then the phase-K interim belief is a function of interim belief at phase K − 1. Therefore, figuring out how these two types of experiments, determined and sender-designed experiments, will influence every given phase’s interim belief is the key to solving for the optimal signaling strategy. We start by noting that if the posterior belief at a leaf node is given, then the receiver’s action is determined - he will take the action with the highest posterior probability unless there is a tie, in which case he is indifferent and will follow the sender’s recommendation. Therefore, we can use backward iteration and the principle of optimality to determine the optimal signaling. We start by considering the last phase’s experiments when the sender can design them. Experiments in phase N − 1 In the second-last phase, results in Section 3.4 describe a sender-designed experiment’s role in the optimal signaling strategy: pick the strategy on the frontier of all persuasion-ratio curves. However, if the experiment is determined in the second-last phase, an additional constraint on the interim belief between the second-last phase and the last phase is enforced. That is to say, the set of (feasible) strategies will shrink. Fortunately, after enforcing the constraints, the process of searching for the optimal signaling strategy under a determined experiment is the same as the sender-designed experiment, i.e., pick the strategy in the frontier of all persuasion ratio curves. Therefore, at each possible branch of phase N − 1, we can plot an optimal persuasion ratio curve capturing the sender’s optimal signaling strategy at phase N − 1 and phase N. Non-binary Outcome Experiments When the experiments have non-binary outcomes, the same approach derived above works with an increased number of phases (if complexity is not an issue). For general non-binary experiments, see the proof of Lemma 6 in [23] for a detailed construction from non-binary to binary experiments. This paper is available on arxiv under CC 4.0 license. Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Authors: Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Problem Formulation 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4.3 Multi-phase model and classical Bayesian persuasion and References 4.2 Determined versus sender-designed experiments Given the model, the sender can manipulate the phase-K interim belief only when designing an experiment at phase K − 1. If an experiment at phase K − 1 is determined, then the phase-K interim belief is a function of interim belief at phase K − 1. Therefore, figuring out how these two types of experiments, determined and sender-designed experiments, will influence every given phase’s interim belief is the key to solving for the optimal signaling strategy. We start by noting that if the posterior belief at a leaf node is given, then the receiver’s action is determined - he will take the action with the highest posterior probability unless there is a tie, in which case he is indifferent and will follow the sender’s recommendation. Therefore, we can use backward iteration and the principle of optimality to determine the optimal signaling. We start by considering the last phase’s experiments when the sender can design them. Experiments in phase N − 1 In the second-last phase, results in Section 3.4 describe a sender-designed experiment’s role in the optimal signaling strategy: pick the strategy on the frontier of all persuasion-ratio curves. However, if the experiment is determined in the second-last phase, an additional constraint on the interim belief between the second-last phase and the last phase is enforced. That is to say, the set of (feasible) strategies will shrink. Fortunately, after enforcing the constraints, the process of searching for the optimal signaling strategy under a determined experiment is the same as the sender-designed experiment, i.e., pick the strategy in the frontier of all persuasion ratio curves. Therefore, at each possible branch of phase N − 1, we can plot an optimal persuasion ratio curve capturing the sender’s optimal signaling strategy at phase N − 1 and phase N . Experiments in phase N − 1 N N N N Non-binary Outcome Experiments When the experiments have non-binary outcomes, the same approach derived above works with an increased number of phases (if complexity is not an issue). Non-binary Outcome Experiments For general non-binary experiments, see the proof of Lemma 6 in [23] for a detailed construction from non-binary to binary experiments. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Optimizing Signaling Strategies with Sender-Designed Experiments in Multi-phase Trials

Authors:
(1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu);
(2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu);
(3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4 Binary-outcome Experiments in Multi-phase trials This section generalizes the structural results in Section 3 to multi-phase trials. First, we generalize the model in Section 2.1 to multi-phase trials and then propose a dynamic programming algorithm to solve for the optimal signaling strategy. The state for the dynamic program will be the interim belief on the state of the world that results at any node in the extensive-form delineation of the problem. As the belief at each level is determined based on the actions in earlier stages (if any), in the backward iteration procedure, we will determine the optimal choice of experiments by the sender (if there is a choice) for any possible interim belief. In this dynamic programming, there is only a terminal reward that arises from the receiver’s action based on the outcome of the final trial and based on the receiver’s resulting posterior beliefs. 4.1 Model of binary-outcome experiments in multi-phase trials This paper is available on arxiv under CC 4.0 license. Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Authors: Authors: (1) Shih-Tang Su, University of Michigan, Ann Arbor (shihtang@umich.edu); (2) Vijay G. Subramanian, University of Michigan, Ann Arbor and (vgsubram@umich.edu); (3) Grant Schoenebeck, University of Michigan, Ann Arbor (schoeneb@umich.edu). Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Problem Formulation 2. Problem Formulation 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 2.1 Model of Binary-Outcome Experiments in Two-Phase Trials 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3 Binary-outcome Experiments in Two-phase Trials and 3.1 Experiments with screenings 3.2 Assumptions and induced strategies 3.2 Assumptions and induced strategies 3.3 Constraints given by phase-II experiments 3.3 Constraints given by phase-II experiments 3.4 Persuasion ratio and the optimal signaling structure 3.4 Persuasion ratio and the optimal signaling structure 3.5 Comparison with classical Bayesian persuasion strategies 3.5 Comparison with classical Bayesian persuasion strategies 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4 Binary-outcome Experiments in Multi-phase trials and 4.1 Model of binary-outcome experiments in multi-phase trials 4.2 Determined versus sender-designed experiments 4.2 Determined versus sender-designed experiments 4.3 Multi-phase model and classical Bayesian persuasion and References 4.3 Multi-phase model and classical Bayesian persuasion and References 4 Binary-outcome Experiments in Multi-phase trials This section generalizes the structural results in Section 3 to multi-phase trials. First, we generalize the model in Section 2.1 to multi-phase trials and then propose a dynamic programming algorithm to solve for the optimal signaling strategy. The state for the dynamic program will be the interim belief on the state of the world that results at any node in the extensive-form delineation of the problem. As the belief at each level is determined based on the actions in earlier stages (if any), in the backward iteration procedure, we will determine the optimal choice of experiments by the sender (if there is a choice) for any possible interim belief. In this dynamic programming, there is only a terminal reward that arises from the receiver’s action based on the outcome of the final trial and based on the receiver’s resulting posterior beliefs. 4.1 Model of binary-outcome experiments in multi-phase trials This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

A Dynamic Programming Approach to Optimizing Signaling Strategies in Multi-phase Trials:

Primer on Large Language Model (LLM) Inference Optimizations: 2. Introduction to Artificial Intelligence (AI) Accelerators

Primer on Large Language Model (LLM) Inference Optimizations: 2. Úvod do akcelerátorů umělé inteligence (AI)

Introducción a las optimizaciones de inferencia de modelos de lenguaje grandes (LLM): 2. Introducción a los aceleradores de inteligencia artificial (IA)

Manual sobre as optimizacións de inferencia do modelo de linguaxe grande (LLM): 2. Introdución aos aceleradores de intelixencia artificial (IA)

वृहद भाषा मॉडल (एलएलएम) अनुमान अनुकूलन पर प्राइमर: 2. कृत्रिम बुद्धिमत्ता (एआई) त्वरक का परिचय

Introduzione alle ottimizzazioni dell'inferenza del Large Language Model (LLM): 2. Introduzione agli acceleratori dell'intelligenza artificiale (IA)

大規模言語モデル (LLM) 推論最適化入門: 2. 人工知能 (AI) アクセラレータの概要

Primer on Large Language Model (LLM) დასკვნის ოპტიმიზაცია: 2. შესავალი ხელოვნური ინტელექტის (AI) ამაჩქარებლებში

Үлкен тілдік үлгідегі (LLM) қорытындыны оңтайландырудағы праймер: 2. Жасанды интеллект (AI) үдеткіштеріне кіріспе

Том хэлний загвар (LLM) дүгнэлтийг оновчтой болгох праймер: 2. Хиймэл оюун ухаан (AI) хурдасгуурын танилцуулга

Horudhac ku saabsan Qaabka Luqadda Weyn (LLM) Kordhinta Farshaxanka: 2. Horudhaca Sirdoonka Artificial (AI) Dardariyeyaasha

Primer on Large Language Model (LLM) Inference Optimizations: 2. Panimula sa Artificial Intelligence (AI) Accelerators

Sách nhập môn về Tối ưu hóa suy luận mô hình ngôn ngữ lớn (LLM): 2. Giới thiệu về bộ tăng tốc trí tuệ nhân tạo (AI)

This post represents some key SEO strategies for guesthouses and hotels websites, based on what was discussed during the discussion of an SEO blog post. Creating a Website Both sources note that having online visibility is important, especially when it comes to companies operating in the travel market. According to the blog post, WordPress is a very good choice in which to operate within the USA market because it's friendly, easy to customize, and has a very strong support network. Understanding the Travel Market Sources underscore the need to understand the details of the English travel market. Tourists are often seeking something specific - for instance, proximity to some important cultural site or some spectacular natural landmark - so targeting keywords reflecting these needs is important, such as "guest house near Rila Monastery" or "guest house with traditional English breakfast." Listings in the travel directories and guesthouse websites will continue to be influential, especially in tourist capture. Reviews and Social Proof Reviews are cardinal on Google My Business and TripAdvisor for travelers. On the other hand, sources indicate one up-and-coming review platform and remind one that local context is important in SEO strategy. Since word of mouth is a big part of the culture, adding in social proof, such as testimonials and positive mentions on social media, is key. Outsmarting the Booking Giants While valuing the dominance of sites such as Booking.com and Airbnb, all sources strongly advocate not relying exclusively on these reservation platforms. They recommend balancing marketing efforts and making direct contact with guests through a website using SEO. This involves creating unique content related to local experiences, food, and traditions that appeal to a special type of visitor. Keyword Research and Content Creation This blog post, therefore, highlights that researching keywords is one of the best means of persuading owners not to use generic terms but to focus on keywords depicting their experiences and amenities. Instead of "guest house with a pool," think "guest house with natural spring-fed pool" for those special touches. The focus should be on highly qualitative and captivating content worth the visitor's time, such as blog posts covering local attractions, photo galleries, and even videos. Adding traditional folk music or telling stories about them will make them even more immersive. Link Building and Collaboration Sources stress the importance of relationship-building with other businesses and websites in the travel industry. It is possible to reach out to local bloggers, tourism organizations, and even other guesthouses to discuss collaborations. For instance, two guesthouses from different regions could create a joint blog post showcasing a beautiful road trip, which would extend their exposure and build quality backlinks. The Human Touch and Future of SEO The sources stress the human psychological aspect of SEO, where English travelers find authenticity and personal recommendations particularly appealing. They foresee further growth in voice search and the requirement for optimization regarding conversational searches. The sources, however, still believe human judgment will be important, especially in deeper insights and catering to the peculiar needs of each guest, as AI-powered search engines get smarter. It's all about finding that balance in how to automate and remain authentic, using AI as a strategic tool that makes sure that the distinctive character of a guesthouse is maintained. Holistic Marketing Approach The sources go on to point out that the holistic marketing approach is an integration of online-offline activities. Examples are the creation of an e-welcome package for guests with useful information and a reflection of traditional English hospitality within online communication. Similarly, social media should be utilized for contemporary guest interaction, sharing behind-the-scenes content, and local influencers assisting further increasing the scope and effectiveness of this interaction. In other words, effective SEO strategies for guesthouses and hotel websites balance technical prowess with cultural sensitivity and highlight the heart and soul of a guesthouse experience through meaningful connections between guesthouse owners and travelers. This post represents some key SEO strategies for guesthouses and hotels websites, based on what was discussed during the discussion of an SEO blog post. Creating a Website Creating a Website Both sources note that having online visibility is important, especially when it comes to companies operating in the travel market. According to the blog post, WordPress is a very good choice in which to operate within the USA market because it's friendly, easy to customize, and has a very strong support network. Understanding the Travel Market Understanding the Travel Market Sources underscore the need to understand the details of the English travel market. Tourists are often seeking something specific - for instance, proximity to some important cultural site or some spectacular natural landmark - so targeting keywords reflecting these needs is important, such as "guest house near Rila Monastery" or "guest house with traditional English breakfast." Listings in the travel directories and guesthouse websites will continue to be influential, especially in tourist capture. Reviews and Social Proof Reviews and Social Proof Reviews are cardinal on Google My Business and TripAdvisor for travelers. On the other hand, sources indicate one up-and-coming review platform and remind one that local context is important in SEO strategy. Since word of mouth is a big part of the culture, adding in social proof, such as testimonials and positive mentions on social media, is key. Outsmarting the Booking Giants Outsmarting the Booking Giants While valuing the dominance of sites such as Booking.com and Airbnb, all sources strongly advocate not relying exclusively on these reservation platforms. They recommend balancing marketing efforts and making direct contact with guests through a website using SEO. This involves creating unique content related to local experiences, food, and traditions that appeal to a special type of visitor. Booking.com Keyword Research and Content Creation Keyword Research and Content Creation This blog post, therefore, highlights that researching keywords is one of the best means of persuading owners not to use generic terms but to focus on keywords depicting their experiences and amenities. Instead of "guest house with a pool," think "guest house with natural spring-fed pool" for those special touches. The focus should be on highly qualitative and captivating content worth the visitor's time, such as blog posts covering local attractions, photo galleries, and even videos. Adding traditional folk music or telling stories about them will make them even more immersive. Link Building and Collaboration Link Building and Collaboration Sources stress the importance of relationship-building with other businesses and websites in the travel industry. It is possible to reach out to local bloggers, tourism organizations, and even other guesthouses to discuss collaborations. For instance, two guesthouses from different regions could create a joint blog post showcasing a beautiful road trip, which would extend their exposure and build quality backlinks. The Human Touch and Future of SEO The Human Touch and Future of SEO The sources stress the human psychological aspect of SEO, where English travelers find authenticity and personal recommendations particularly appealing. They foresee further growth in voice search and the requirement for optimization regarding conversational searches. The sources, however, still believe human judgment will be important, especially in deeper insights and catering to the peculiar needs of each guest, as AI-powered search engines get smarter. It's all about finding that balance in how to automate and remain authentic, using AI as a strategic tool that makes sure that the distinctive character of a guesthouse is maintained. Holistic Marketing Approach Holistic Marketing Approach The sources go on to point out that the holistic marketing approach is an integration of online-offline activities. Examples are the creation of an e-welcome package for guests with useful information and a reflection of traditional English hospitality within online communication. Similarly, social media should be utilized for contemporary guest interaction, sharing behind-the-scenes content, and local influencers assisting further increasing the scope and effectiveness of this interaction. In other words, effective SEO strategies for guesthouses and hotel websites balance technical prowess with cultural sensitivity and highlight the heart and soul of a guesthouse experience through meaningful connections between guesthouse owners and travelers. effective SEO strategies for guesthouses and hotel websites

SEO Strategies for Optimizing Hotel Websites

የሆቴል ድር ጣቢያዎችን ለማመቻቸት SEO ስልቶች

SEO Estrategias ukax Hotel ukan sitios web ukanakar sum apnaqañataki

Στρατηγικές SEO για βελτιστοποίηση ιστοσελίδων ξενοδοχείων

Estrategias SEO para optimizar sitios web de hoteles

Stratégies SEO pour optimiser les sites Web des hôtels

ホテルのウェブサイトを最適化するための SEO 戦略

SEO სტრატეგიები სასტუმროს ვებსაიტების ოპტიმიზაციისთვის

ຍຸດທະສາດ SEO ສໍາລັບການເພີ່ມປະສິດທິພາບເວັບໄຊທ໌ໂຮງແຮມ

Зочид буудлын вэбсайтыг оновчтой болгох SEO стратеги

SEO-strategieën voor het optimaliseren van hotelwebsites

SEO Strategies Yekukwidziridza Hotels Webhusaiti

กลยุทธ์ SEO สำหรับการเพิ่มประสิทธิภาพเว็บไซต์โรงแรม

AWS DynamoDB is a fully managed, serverless document store that offers single-digit millisecond performance at any scale. It supports various data models and supports transactions, global tables for multi-region replication. It is crucial to design the schema well in DynamoDB, otherwise it can lead to significant cost implications and degraded performance. In this article we will go through the techniques that can be applied for cost optimizations and scalability for an orchestration application. The pricing structure of DynamoDB can be found here and here. Orchestration Application The application is an orchestration service that stores requests from clients, sends the event to multiple upstreams. Each upstream fetches the request on receipt of event, processes it and replies with it’s status. The caller then asynchronously can find out the status of the request. The status contains each upstream responses. This application experiences high volume surge for several minutes followed by inactivity. On-demand pricing is better choice for these type of applications. We will start with a simple schema solution for this application and then go into detail to optimize the schema for better cost and performance. The cost estimates can be calculated using AWS Pricing Calculator Scenario 1 We store the request as well as upstream responses in the same dynamo document
Everytime the upstream sends the response, we update this document
Inputs

Size of document storing request = 100 KB
Size of responses from upstreams = 100 KB
Total average document size = 200 KB
Requests per day = 1 Million
number of upstream services to handle each request = 10 Pricing Operation

Calculation

Total Cost



Write Operations







1 write request

1 WRU per 200 KB = 200 WRUs

-



10 upstream writes

10 * 200 WRUs = 2,000 WRUs

-



1 million requests

1 million * 2,200 WRUs = $1.25 * 2,200

$2,750











Read Operations (Strongly consistent)







1 read request

1 RRU per 200 KB = 50 RRUs

-



10 upstream reads

10 * 50 RRUs = 500 RRUs

-



1 million requests

1 million * 500 RRUs = $0.25 * 500

$125











Total Cost per Day



$2,875 Scenario 2 The major cost in previous scenario is due to writes with larger item size. In our application, we really don’t need to maintain the request data and responses data in a single document. If we split them into two, we do not need to write whole document each time the upstream sends reply. We update only the responses document which is 100 KB, reducing writes on 100 KB per upstream per request Pricing Operation

Calculation

Total Cost



Write Operations







1 write request

1 WRU per 100 KB = 100 WRUs

-



10 upstream writes

10 * 100 WRUs = 1,000 WRUs

-



1 million requests

1 million * 1,100 WRUs = $1.25 * 1,100

$1,375











Read Operations (Strongly Consistent)







1 read request

1 RRU per 100 KB = 25 RRUs

-



10 upstream reads

10 * 25 RRUs = 250 RRUs

-



1 million requests

1 million * 250 RRUs = $0.25 * 250

$62.5











Total Cost per Day



$1,437.5 This design resulted in 50% reduction in cost from Scenario 1 Scenario 3 Looking deep into the use case of the upstream responses, there is no dependency on replies between each upstream services. We do not need to write each reply in a single document. If we keep them in a single document, it makes easier to read them. Given our use case, the writes are higher compared to reads. So we can try to split the responses document into separate ones for each upstream service. In this scenario, we will have 10 documents with 10 KB size for upstream responses instead of single 100 KB document. Pricing Operation

Calculation

Total Cost



Write Operations







1 write request

1 WRU per 100 KB (request) = 100 WRUs

-



1 write request (upstream)

1 WRU per 10 KB = 10 WRUs

-



10 upstream writes

10 * 10 WRUs = 100 WRUs

-



1 million requests

1 million * 200 WRUs = $1.25 * 200

$250











Read Operations (Strongly consistent)







1 read request

1 RCU per 100 KB = 25 RRUs

-



10 upstream reads

10 * 25 RCUs = 250 RRUs

-



1 million requests

1 million * 250 RRUs = 0.25 * 250

$62.5











Total Cost per Day



$312.5 This design achieves 78.26% reduction in cost compared to Scenario 2 and 89.1% reduction compared to Scenario 1 Conclusion The schema design plays an important role in DynamoDb cost optimizations. The size of each document should be kept as minimal as possible for write heavy applications. To reduce costs, each document’s size should be minimized, as larger items incur higher write costs. AWS DynamoDB is a fully managed, serverless document store that offers single-digit millisecond performance at any scale. It supports various data models and supports transactions, global tables for multi-region replication. It is crucial to design the schema well in DynamoDB, otherwise it can lead to significant cost implications and degraded performance. In this article we will go through the techniques that can be applied for cost optimizations and scalability for an orchestration application. The pricing structure of DynamoDB can be found here and here . here here Orchestration Application The application is an orchestration service that stores requests from clients, sends the event to multiple upstreams. Each upstream fetches the request on receipt of event, processes it and replies with it’s status. The caller then asynchronously can find out the status of the request. The status contains each upstream responses. This application experiences high volume surge for several minutes followed by inactivity. On-demand pricing is better choice for these type of applications. We will start with a simple schema solution for this application and then go into detail to optimize the schema for better cost and performance. The cost estimates can be calculated using AWS Pricing Calculator AWS Pricing Calculator Scenario 1 We store the request as well as upstream responses in the same dynamo document Everytime the upstream sends the response, we update this document Inputs

Size of document storing request = 100 KB
Size of responses from upstreams = 100 KB
Total average document size = 200 KB
Requests per day = 1 Million
number of upstream services to handle each request = 10 We store the request as well as upstream responses in the same dynamo document Everytime the upstream sends the response, we update this document Inputs Size of document storing request = 100 KB
Size of responses from upstreams = 100 KB
Total average document size = 200 KB
Requests per day = 1 Million
number of upstream services to handle each request = 10 Size of document storing request = 100 KB Size of responses from upstreams = 100 KB Total average document size = 200 KB Requests per day = 1 Million number of upstream services to handle each request = 10 Size of document storing request = 100 KB Size of responses from upstreams = 100 KB Total average document size = 200 KB Requests per day = 1 Million number of upstream services to handle each request = 10 Pricing Operation

Calculation

Total Cost



Write Operations







1 write request

1 WRU per 200 KB = 200 WRUs

-



10 upstream writes

10 * 200 WRUs = 2,000 WRUs

-



1 million requests

1 million * 2,200 WRUs = $1.25 * 2,200

$2,750











Read Operations (Strongly consistent)







1 read request

1 RRU per 200 KB = 50 RRUs

-



10 upstream reads

10 * 50 RRUs = 500 RRUs

-



1 million requests

1 million * 500 RRUs = $0.25 * 500

$125











Total Cost per Day



$2,875 Operation

Calculation

Total Cost Write Operations 1 write request

1 WRU per 200 KB = 200 WRUs

- 10 upstream writes

10 * 200 WRUs = 2,000 WRUs

- 1 million requests

1 million * 2,200 WRUs = $1.25 * 2,200

$2,750 Read Operations (Strongly consistent) 1 read request

1 RRU per 200 KB = 50 RRUs

- 10 upstream reads

10 * 50 RRUs = 500 RRUs

- 1 million requests

1 million * 500 RRUs = $0.25 * 500

$125 Total Cost per Day



$2,875 Operation Calculation Total Cost Operation Operation Calculation Calculation Total Cost Total Cost Write Operations Write Operations Write Operations Write Operations 1 write request 1 WRU per 200 KB = 200 WRUs - 1 write request 1 write request 1 WRU per 200 KB = 200 WRUs 1 WRU per 200 KB = 200 WRUs - - 10 upstream writes 10 * 200 WRUs = 2,000 WRUs - 10 upstream writes 10 upstream writes 10 * 200 WRUs = 2,000 WRUs 10 * 200 WRUs = 2,000 WRUs - - 1 million requests 1 million * 2,200 WRUs = $1.25 * 2,200 $2,750 1 million requests 1 million requests 1 million * 2,200 WRUs = $1.25 * 2,200 1 million * 2,200 WRUs = $1.25 * 2,200 $2,750 $2,750 Read Operations (Strongly consistent) Read Operations (Strongly consistent) Read Operations (Strongly consistent) Read Operations (Strongly consistent) 1 read request 1 RRU per 200 KB = 50 RRUs - 1 read request 1 read request 1 RRU per 200 KB = 50 RRUs 1 RRU per 200 KB = 50 RRUs - - 10 upstream reads 10 * 50 RRUs = 500 RRUs - 10 upstream reads 10 upstream reads 10 * 50 RRUs = 500 RRUs 10 * 50 RRUs = 500 RRUs - - 1 million requests 1 million * 500 RRUs = $0.25 * 500 $125 1 million requests 1 million requests 1 million * 500 RRUs = $0.25 * 500 1 million * 500 RRUs = $0.25 * 500 $125 $125 Total Cost per Day $2,875 Total Cost per Day Total Cost per Day Total Cost per Day $2,875 $2,875 $2,875 Scenario 2 The major cost in previous scenario is due to writes with larger item size. In our application, we really don’t need to maintain the request data and responses data in a single document. If we split them into two, we do not need to write whole document each time the upstream sends reply. We update only the responses document which is 100 KB, reducing writes on 100 KB per upstream per request Pricing Operation

Calculation

Total Cost



Write Operations







1 write request

1 WRU per 100 KB = 100 WRUs

-



10 upstream writes

10 * 100 WRUs = 1,000 WRUs

-



1 million requests

1 million * 1,100 WRUs = $1.25 * 1,100

$1,375











Read Operations (Strongly Consistent)







1 read request

1 RRU per 100 KB = 25 RRUs

-



10 upstream reads

10 * 25 RRUs = 250 RRUs

-



1 million requests

1 million * 250 RRUs = $0.25 * 250

$62.5











Total Cost per Day



$1,437.5 Operation

Calculation

Total Cost Write Operations 1 write request

1 WRU per 100 KB = 100 WRUs

- 10 upstream writes

10 * 100 WRUs = 1,000 WRUs

- 1 million requests

1 million * 1,100 WRUs = $1.25 * 1,100

$1,375 Read Operations (Strongly Consistent) 1 read request

1 RRU per 100 KB = 25 RRUs

- 10 upstream reads

10 * 25 RRUs = 250 RRUs

- 1 million requests

1 million * 250 RRUs = $0.25 * 250

$62.5 Total Cost per Day



$1,437.5 Operation Calculation Total Cost Operation Operation Calculation Calculation Total Cost Total Cost Write Operations Write Operations Write Operations Write Operations 1 write request 1 WRU per 100 KB = 100 WRUs - 1 write request 1 write request 1 WRU per 100 KB = 100 WRUs 1 WRU per 100 KB = 100 WRUs - - 10 upstream writes 10 * 100 WRUs = 1,000 WRUs - 10 upstream writes 10 upstream writes 10 * 100 WRUs = 1,000 WRUs 10 * 100 WRUs = 1,000 WRUs - - 1 million requests 1 million * 1,100 WRUs = $1.25 * 1,100 $1,375 1 million requests 1 million requests 1 million * 1,100 WRUs = $1.25 * 1,100 1 million * 1,100 WRUs = $1.25 * 1,100 $1,375 $1,375 Read Operations (Strongly Consistent) Read Operations (Strongly Consistent) Read Operations (Strongly Consistent) Read Operations (Strongly Consistent) 1 read request 1 RRU per 100 KB = 25 RRUs - 1 read request 1 read request 1 RRU per 100 KB = 25 RRUs 1 RRU per 100 KB = 25 RRUs - - 10 upstream reads 10 * 25 RRUs = 250 RRUs - 10 upstream reads 10 upstream reads 10 * 25 RRUs = 250 RRUs 10 * 25 RRUs = 250 RRUs - - 1 million requests 1 million * 250 RRUs = $0.25 * 250 $62.5 1 million requests 1 million requests 1 million * 250 RRUs = $0.25 * 250 1 million * 250 RRUs = $0.25 * 250 $62.5 $62.5 Total Cost per Day $1,437.5 Total Cost per Day Total Cost per Day Total Cost per Day $1,437.5 $1,437.5 $1,437.5 This design resulted in 50% reduction in cost from Scenario 1 50% reduction Scenario 3 Looking deep into the use case of the upstream responses, there is no dependency on replies between each upstream services. We do not need to write each reply in a single document. If we keep them in a single document, it makes easier to read them. Given our use case, the writes are higher compared to reads. So we can try to split the responses document into separate ones for each upstream service. In this scenario, we will have 10 documents with 10 KB size for upstream responses instead of single 100 KB document. Pricing Operation

Calculation

Total Cost



Write Operations







1 write request

1 WRU per 100 KB (request) = 100 WRUs

-



1 write request (upstream)

1 WRU per 10 KB = 10 WRUs

-



10 upstream writes

10 * 10 WRUs = 100 WRUs

-



1 million requests

1 million * 200 WRUs = $1.25 * 200

$250











Read Operations (Strongly consistent)







1 read request

1 RCU per 100 KB = 25 RRUs

-



10 upstream reads

10 * 25 RCUs = 250 RRUs

-



1 million requests

1 million * 250 RRUs = 0.25 * 250

$62.5











Total Cost per Day



$312.5 Operation

Calculation

Total Cost Write Operations 1 write request

1 WRU per 100 KB (request) = 100 WRUs

- 1 write request (upstream)

1 WRU per 10 KB = 10 WRUs

- 10 upstream writes

10 * 10 WRUs = 100 WRUs

- 1 million requests

1 million * 200 WRUs = $1.25 * 200

$250 Read Operations (Strongly consistent) 1 read request

1 RCU per 100 KB = 25 RRUs

- 10 upstream reads

10 * 25 RCUs = 250 RRUs

- 1 million requests

1 million * 250 RRUs = 0.25 * 250

$62.5 Total Cost per Day



$312.5 Operation Calculation Total Cost Operation Operation Calculation Calculation Total Cost Total Cost Write Operations Write Operations Write Operations Write Operations 1 write request 1 WRU per 100 KB (request) = 100 WRUs - 1 write request 1 write request 1 WRU per 100 KB (request) = 100 WRUs 1 WRU per 100 KB (request) = 100 WRUs - - 1 write request (upstream) 1 WRU per 10 KB = 10 WRUs - 1 write request (upstream) 1 write request (upstream) 1 WRU per 10 KB = 10 WRUs 1 WRU per 10 KB = 10 WRUs - - 10 upstream writes 10 * 10 WRUs = 100 WRUs - 10 upstream writes 10 upstream writes 10 * 10 WRUs = 100 WRUs 10 * 10 WRUs = 100 WRUs - - 1 million requests 1 million * 200 WRUs = $1.25 * 200 $250 1 million requests 1 million requests 1 million * 200 WRUs = $1.25 * 200 1 million * 200 WRUs = $1.25 * 200 $250 $250 Read Operations (Strongly consistent) Read Operations (Strongly consistent) Read Operations (Strongly consistent) Read Operations (Strongly consistent) 1 read request 1 RCU per 100 KB = 25 RRUs - 1 read request 1 read request 1 RCU per 100 KB = 25 RRUs 1 RCU per 100 KB = 25 RRUs - - 10 upstream reads 10 * 25 RCUs = 250 RRUs - 10 upstream reads 10 upstream reads 10 * 25 RCUs = 250 RRUs 10 * 25 RCUs = 250 RRUs - - 1 million requests 1 million * 250 RRUs = 0.25 * 250 $62.5 1 million requests 1 million requests 1 million * 250 RRUs = 0.25 * 250 1 million * 250 RRUs = 0.25 * 250 $62.5 $62.5 Total Cost per Day $312.5 Total Cost per Day Total Cost per Day Total Cost per Day $312.5 $312.5 $312.5 This design achieves 78.26% reduction in cost compared to Scenario 2 and 89.1% reduction compared to Scenario 1 78.26% reduction 89.1% reduction Conclusion The schema design plays an important role in DynamoDb cost optimizations. The size of each document should be kept as minimal as possible for write heavy applications. To reduce costs, each document’s size should be minimized, as larger items incur higher write costs.

Cost Optimization Techniques for AWS DynamoDB in Orchestration Applications

Stories, News & Mentions about OPTIMITY

Optimity’s Stories on HackerNoon