Nixtla

nixtla

AI has radically reshaped predictive analytics, making forecasting more precise than ever before. Businesses in all industries are taking advantage, capitalizing on AI’s wide-ranging capabilities to anticipate future trends and prepare for what’s coming around the corner. The accuracy of any forecast hinges on a number of factors: the sophistication of , the robustness of predictive models, and the quality and quantity of the data they’re trained on, to name a few. AI algorithms is a generative pre-trained model specifically designed for predicting time-series data. It’s a powerful asset that will allow you to forecast future outcomes with greater precision, giving you the foresight to capitalize on upcoming opportunities and dodge possible pitfalls. Nixtla’s TimeGPT Below, we’ll put TimeGPT under the microscope, analyze it from all angles, and show how you can integrate it with your database to maximize the full potential of predictive analytics. What is TimeGPT? TimeGPT is a pre-trained generative model for time-series data forecasting. Its core functionality is to predict future values in a time series based on historical data. The application processes data sequentially, using a windowing technique to analyze historical data points and recognize patterns. This enables it to predict future values by extrapolating these identified patterns. An overview of TimeGPT : Nixtla's TimeGPT applies pre-trained models to time-series data, using the data collected to predict new datasets without specific training, a departure from traditional models. Innovative Pre-training : TimeGPT excels in identifying complex patterns in time series data, surpassing many traditional models in trend, seasonality, and anomaly detection. Advanced Pattern Recognition : By using advanced neural networks, TimeGPT offers efficient forecasting, reducing the time and resources typically needed for time series analysis by eliminating dataset-specific retraining. Efficient Forecasting : Unlike static traditional models, TimeGPT has the potential for ongoing learning and adaptation, maintaining accuracy as new data emerges. Continuous Learning and Adaptation TimeGPT’s features and benefits As one of the leading predictive analytics models, TimeGPT offers advanced capabilities that provide a variety of unique features, making it a powerful asset for managing complex forecasting scenarios. These include: Anomaly Detection Leveraging predictive modeling techniques, TimeGPT excels in identifying unusual patterns or outliers in time series data, which are often indicative of significant but unexpected events. Feature: This feature is crucial for businesses and analysts in quickly detecting and responding to unusual occurrences, ensuring that forecasting models remain accurate and reliable, even in the presence of aberrations. Benefit: Exogenous Variables Handling TimeGPT can incorporate external factors or exogenous variables (like weather conditions for retail sales forecasting) into its analysis. Feature: This ability allows for more nuanced and accurate forecasts by considering the impact of external elements, which is particularly useful in scenarios where external factors significantly influence the data trends. Benefit: Fine-tuning While TimeGPT is a pre-trained model, it offers the capability for fine-tuning to adapt to specific datasets or unique forecasting needs. Feature: This feature enables users to tailor the model more closely to their specific requirements, enhancing the accuracy and relevance of the forecasts in specialized contexts. Benefit: Historical Forecasting TimeGPT can analyze past data to generate historical forecasts, allowing users to understand how the model would have predicted past events. Feature: This is particularly beneficial for validating the model’s accuracy and reliability, giving users confidence in its future predictions. Benefit: Managing complex forecasting scenarios TimeGPT’s features are especially beneficial when handling complex scenarios such as holidays, special dates, and irregular timestamps. These events can significantly impact trends in time series data. TimeGPT’s ability to detect anomalies and incorporate exogenous variables helps in accurately forecasting around these dates, considering their unique impact. Holidays and Special Dates: Time series data can sometimes be irregular or have missing points. TimeGPT’s sophisticated algorithms can handle such irregularities, ensuring consistent and accurate forecasting even with non-uniform data. Irregular Timestamps: Deploying TimeGPT models in MindsDB MindsDB is the platform for building custom AI from enterprise data. It works by connecting any source of data with any AI/ML model or framework and automating how real-time data flows between them. MindsDB bridges data sources with advanced AI/ML models like TimeGPT. This allows you to easily create, train, and manage these models, making it particularly beneficial to those who have minimal experience in machine learning. Integrating TimeGPT into MindsDB begins with creating a TimeGPT ML engine using the below statement and providing the TimeGPT API key. CREATE ML_ENGINE timegpt_engine
FROM timegpt
USING timegpt_api_key = ‘timegpt-api-key’; Next, users can employ the statement to create, train, and deploy the TimeGPT model. CREATE MODEL CREATE MODEL model_table
FROM data_source
  (SELECT * FROM data_table)
PREDICT column_to_be_predicted
ORDER BY date_column
GROUP BY column_name, column_name, ...
HORIZON 3 -- model forecasts the next 3 rows
USING ENGINE = 'timegpt_engine'; Now you can query the model to get forecasts: SELECT m.date_column, m.column_to_be_predicted
FROM data_table AS d
JOIN model_table AS m
WHERE d.date_column > LATEST; With MindsDB, you can automate forecasts and generate them constantly for any downstream usage. It is especially useful when you have dynamic data that gets updated regularly. Then, the model should be fine-tuned with the latest data to provide as accurate forecasts as possible. Here, we create a job that fine-tunes the model with the latest data and saves daily forecasts in a database table. CREATE JOB automate_forecasts (

   FINETUNE model_table
   FROM data_source
      (SELECT * FROM data_table);

   CREATE TABLE daily_forecasts_{{START_DATETIME}} (
      SELECT m.date_column, m.column_to_be_predicted
      FROM data_table AS d
      JOIN model_table AS m
      WHERE d.date_column > LATEST
   )
)
EVERY 1 day; To see a complete example, visit our . You can find more . docs page here forecasting use cases here How to use TimeGPT in predictive analytics Companies must be agile enough to react to minute-by-minute market changes. Using predictive analytics techniques, such as those enabled by TimeGPT, will help keep you a step ahead and even give you a competitive advantage. Businesses can utilize TimeGPT in several key areas, including: Financial Forecasting TimeGPT’s proficiency in processing and interpreting time-based data leads to more precise forecasts in the fluctuating realm of finance. Its ability to identify current trends and subtle nuances within financial datasets helps to create more informed investment strategies. Demand Prediction In industries like retail and manufacturing, TimeGPT aids in predicting product demand. This is crucial for inventory management, supply chain optimization, and strategic planning. By analyzing historical sales data and market trends, TimeGPT can forecast future demand patterns, helping businesses to prepare and adapt effectively. Anomaly Detection The model's approach to sequential learning proves advantageous in detecting anomalies, like spotting unusual patterns in network traffic for cybersecurity or identifying machine performance irregularities for predictive maintenance. This function is crucial in averting potential problems before they intensify. Predictive analytics use cases Here are just a few ways to use TimeGPT for accurate predictive analytics. TimeGPT performs well compared to traditional methods in stock price prediction, aiding investors and analysts. Stock Market Analysis: : It predicts healthcare trends, which is important for resource allocation and planning. Healthcare Trend Anticipation TimeGPT predicts future demands in logistics, improving efficiency and reducing costs. Supply Chain Optimization: It forecasts retail buying trends, aiding in inventory management and targeted marketing. Retail Customer Behavior Analysis: TimeGPT identifies potential equipment failures, minimizing downtime and maintenance costs. Manufacturing Predictive Maintenance: TimeGPT in the real world Global business is more complex and interconnected than ever. One false move in a single industry can have a ripple effect in several others, making it crucial for companies to leverage predictive analytics to foresee, prepare, and react to future events. With its advanced time-series analysis, deep learning capabilities, adaptive learning, and customizable scalable models, TimeGPT has proven to be an effective tool for more accurate predictive analysis. Integrating TimeGPT with MindsDB creates a powerful combination that enhances the precision and applicability of predictive insights directly within business databases. This gives you the agility to anticipate and react swiftly to a rapidly evolving global market. How would TimeGPT perform on your company’s data? Check out the . tutorial here Also published . here

How TimeGPT Transforms Predictive Analytics with AI

টাইমজিপিটি কীভাবে AI এর সাথে ভবিষ্যদ্বাণীমূলক বিশ্লেষণকে রূপান্তরিত করে

Wie TimeGPT Predictive Analytics mit KI transformiert

Cómo TimeGPT transforma el análisis predictivo con IA

Comment TimeGPT transforme l'analyse prédictive avec l'IA

TimeGPT कैसे AI के साथ पूर्वानुमानात्मक विश्लेषण को बदलता है

TimeGPT が AI を使用して予測分析をどのように変革するか

TimeGPT가 AI를 통해 예측 분석을 혁신하는 방법

Como o TimeGPT transforma a análise preditiva com IA

Как TimeGPT преобразует прогнозную аналитику с помощью ИИ

TimeGPT, Yapay Zeka ile Tahmine Dayalı Analitiği Nasıl Dönüştürüyor?

Cách TimeGPT biến đổi phân tích dự đoán bằng AI

TimeGPT 如何利用 AI 改变预测分析

Title heavily inspired by great talk; Photo by from Disclaimer: this Thomas Bormans Unsplash As the name implies, today we want to consider almost trivially simple models. Although the current trend points towards complex models, even for time-series models, I am still a big believer in simplicity. In particular, when your dataset is small, the subsequent ideas might be useful. To be fair, this article will probably be most valuable for people who are just starting out with time-series analysis. Anyone else should check the table of contents first and decide for themselves if they want to continue. Personally, I am still quite intrigued by how far you can push even the most simplistic time-series models. The upcoming paragraphs show some ideas and thoughts that I have been gathering on the topic over time. Models with pure i.i.d. noise We start with the simplest (probabilistic) way to model a (univariate) time-series. Namely, we want to look at plain ndependently, dentically, istributed randomness: i i d This implies that all our observations follow the same distribution at any point in time ( distributed). Even more importantly, we presume no interrelation between observations at all ( distributed). Obviously, this precludes any autoregressive terms as well. identically independently Probably your first question is if such models aren't too simplistic to be useful for real-world problems. Certainly, most time-series are unlikely to have no statistical relationship with their own past. While those concerns are true by all means, we can nevertheless deduce the following: Any time-series model that is more complex than a pure-noise model should also produce better forecasts than a pure-noise model. In short, we can at least use random noise as a benchmark model. There is arguably no simpler approach to create baseline benchmarks than this one. Even smoothing techniques will likely require more parameters to be fitted. Besides this rather obvious use-case, there is another potential application for i.i.d. noise. Due to their simplicity, noise models cand be useful for very small datasets. Consider this: If big, complex models require large datasets to prevent overfitting, then simple models require only a handful of data. Of course, it is debatable what dataset size can be seen as 'small'. Integrated i.i.d. noise Now, things are becoming more interesting. While raw i.i.d. noise cannot account for auto-correlation between observations, integrated noise can. Before we do a demonstration, let us introduce the : differencing operator If you haven't heard about differencing for time-series problems yet - great! If you have, then you can hopefully still learn something new. Definition of an integrated time-series With the difference operator in our toolbox, we can now define an : integrated time-series There are several ideas in this definition that we should clarify further: First, you probably noticed the concept of exponentiating the difference operator. You can simply think of this as performing the differentiation several times. For the squared difference operator, this would look as follows: As we will see, multiple difference operators allow us to handle different time-series patterns at once. Third, it is common convention to simply write We will happily adopt this convention here. Also, we call such time-series without referencing its order or seasonality. simply integrated Obviously, we also need to re-transform a difference representation back to its original domain. In our notation, this means we invert the difference transformation, i.e. must hold for arbitrary difference transformations. If we expand this formula, we get These simplifications follow from the fact the difference operator is a linear operator (we won't cover the details here). Technically, the last equation merely says that the next observation is a sum of this observation plus a delta. In a forecasting problem, we will typically have a prediction for the change Let's denote this prediction as to stress that it is not the actual change, but a predicted one. Thus, the forecast for the integrated time-series is Afterwards, we apply this logic recursively as far into the future as our forecast should go: Integrated noise for seemingly complex patterns By now, you can probably imagine what is meant by an integrated noise model. In fact, we can come up with countless variants of an integrated noise model by just chaining some difference operators with random noise. Linear trends from integrated time-series One possibility would be a simply integrated time-series, i.e. It is an interesting exercise to simulate data from such a model using a plain standard normal distribution. As it turns out, samples from this time-series appear to exhibit linear trends with potential change points. However, it is clear that these trends and change points occur completely at random. This implies that simply fitting piece-wise linear functions to forecast such trends can be a dangerous approach. After all, if the changes are occurring at random, then all linear trend lines are mere artifacts of the random data-generating process. As an important disclaimer, though, ' ' means unpredictable from the time-series itself. An external feature might still be able to accurately forecast potential change points. Here, however, we presume that the time-series is our solely available source of information. unpredictable Below, you can see an example of the described phenomenon. While there appears to be a trend change at around t=50, this change is purely random. The upward trend after t=50 also stalls at around t=60. Imagine how your model would have performed if you extrapolated the upward trend after t=60. https://gist.github.com/SaremS/748ed817b1ac395fd8f23f8243213e3f?embedable=true#file-integrated_ts-py Of course, the saying goes 'never say never', even in those settings. However, you should really know what you are doing if you apply such models. Seasonal patterns Similarly to how a simple integration produceds trends, we can also create seasonal patterns: Formally, we now need the s-th difference of our seasonal  process to be a stationary process, e.g. The inverse operation - transforming the i.i.d. process back to the seasonally integrated - works similarly to the one before: You can think of the inverse operation of seasonal differencing as a operation over periods. Since I am not aware of a respective, native Python function, I decided to do to get the desired outcome. Below is an example with : cumsum s reshape->cumsum->reshape s=4 https://gist.github.com/SaremS/fb28ab0782c1e3aa072340e63a54523c?embedable=true#file-seasonal_integration-py As you can see, the generated time-series looks reasonably realistic. We could easily sell this as quarterly sales numbers of some product to an unsuspecting Data Scientist. We could even combine both types of integration to generate a seasonal time-series with trending behavior: https://gist.github.com/SaremS/2f9293fb023ab49e8c164628ca404a97?embedable=true#file-seasonal_and_trend-py At this point, you will probably realize that the title of this article was a little click-baity. Integrated time-series are, in fact, purely linear models. However, I believe that most people wouldn't consider a model with, more-or-less, zero parameters a typical linear model. Memory effects through integration Another interesting property of integrated time-series is the ability to model memory effects. This effect can be seen particularly well when there are larger shocks or outliers in our data. Consider the below example, which shows seasonal integration of order `s=12` over i.i.d. draws from a standard : Cauchy distribution https://gist.github.com/SaremS/32b7825a4cb730f5f9812a0e77f751d9?embedable=true#file-cauchy_memory-py The first large shock in the i.i.d. Cauchy series at around t=20 is sustained over the whole integrated series on the right. Over time, more shocks occur, which are also sustained. This memory property can be very useful in practice. For example, the economic shocks from the pandemic have caused persistent changes in many time-series. Benchmarking against NBEATS and NHITS Let us now use the dataset from Nixtla's for a quick evaluation of the above ideas. If you are regularly reading my articles, you might remember the general procedure from . AirPassengers neuralforecast this one First, we split the data into a train and test period, with the latter consisting of 36 months of data: https://gist.github.com/SaremS/510303feb5c1eef012ba7075930f4fba?embedable=true#file-airpass_data-py In order to obtain a stationary, i.i.d. series we perform the following transformation: First, the square-root stabilizes the increasing variance. The two differencing operators then remove seasonality and trend. For the respective re-transformation, check the code further down below. https://gist.github.com/SaremS/1e82a22be22984efe248a4cb837e233a?embedable=true#file-stationary_ts-py We can also check a histogram and density plot of the stabilized time-series: https://gist.github.com/SaremS/2c214bba7c55cf60b5e2d7610698c6bc?embedable=true#file-density_and_histogram-py Our stationary series looks also somewhat normally distributed, which is always a nice property. Now, let us create the forecast for the test period. Presuming that we don't know the exact distribution of our i.i.d. series, we simply draw from the empirical distribution via the training data. Hence, we simulate future values by reintegrating random samples from the empirical data: https://gist.github.com/SaremS/283e26aab0bc9d688b57793a1e33be98?embedable=true#file-simple_forecast-py This looks very good - the mean forecast is very close to the test data. In addition, our simulation allows us to empirically sample the whole forecast distribution. Therefore, we can also easily add confidence intervals. Finally, let us see how our approach compares against rather complex time-series models. To do so, I went with Nixtla's implementation of and : NBEATS NHITS https://gist.github.com/SaremS/3d801020cd1083224acc97748a6c289d?embedable=true#file-neuralforecast_benchmark-py Below are the respective for the test set: RMSEs 25.5021 Simple Model: 42.6277 NBEATS: 62.6822 NHITS: As we can see, our almost trivial model has beaten two sophisticated time-series models by a fair margin. Of course, we need to emphasize that this doesn't allow to draw any general conclusions. Rather, I'd expect the neural models to outperform our simple approach for larger datasets. Nevertheless, as a benchmark, those trivial models are always a worthwhile consideration. Takeaways - What do we make of this? As stated multiple times throughout this article: A seemingly complex time-series could still follow a fairly simple data-generating process. In the end, you might spend hours trying to fit an overly complex model even though the underlying problem is almost trivial. At some point, somebody could come along, fit a simple ARIMA(1,0,0), and still outperform your sophisticated neural model. To avoid the above worst-case scenario, consider the following idea: When starting out with a new time-series problem, always start with the simplest possible model and use it as a benchmark for all other models. Although this is common knowledge in the Data Science community, I feel like it deserves particular emphasis in this context. Especially due to nowadays' (to some extent justified) hype around Deep Learning, it can be tempting to directly start with something fancy. For many problems, this might just be the right way to go. Nobody today would consider a Hidden Markov Model for NLP today when LLM embeddings are available almost for free now. Once your time-series becomes large, however, modern Machine Learning will likely be better. In particular, are very popular for such large-scale problems. Gradient Boosted Trees A more controversial approach would be, you guessed it, Deep Learning for time-series. While some people believe that these models don't work as well here, their popularity at probably speaks for itself. tech firms like Amazon References Hamilton, James Douglas. . Princeton university press, 2020. [1] Time series analysis Hyndman, Rob J., & Athanasopoulos, George. . OTexts, 2018. [2] Forecasting: principles and practice

A Data Scientist's Guide to Simplistic Time-Series Models

Solving Time Series Forecasting Problems: Principles and Techniques

Welcome! Happy to see you in the seventh part of my Vuejs Amsterdam Conference 2022 summary series, in which I share a summary of all the talks with you. You can read my JSWorld Conference 2022 Summary series (in four parts) , where I summarized all the first day’s talks. You can also find the previous Talks of the Vue Amsterdam conference 2022 in my blog. here (Recurring) Introduction After two and a half years, JSWorld and Vue Amsterdam Conference were back in Theater Amsterdam between 1 and 3 June, and I had the chance to attend this conference for the first time. I learned many things, met many wonderful people, spoke with great developers, and had a great time. On the first day the JSWorld Conference was held, and on the second and third days, the Vue Amsterdam. The conference was full of information with great speakers, each of whom taught me something valuable. They all wanted to share their knowledge and information with other developers. So I thought it would be great if I could continue to share it and help others use it. At first, I tried to share a few notes or slides, but I felt it was not good enough, at least not as good as what the speaker shared with me. So I decided to re-watch each speech, dive deeper into them, search, take notes and combine them with their slides and even their exact words in their speech and then share it with you so that what I share with you is at least at the same level as what I learned from them. A Very Important Point and I have only tried to learn them so that I can turn them into these articles. Even This means if you learn something new, it is because of their efforts. Everything you read during these few articles is the result of the effort and time of the speaker itself, many of the sentences written in these articles are exactly what they said or what they wrote in Slides. Last but not least, I may not dig into every technical detail or live codings in some of the speeches. But if you are interested and need more information, let me know and I’ll try to write a more detailed article separately. Also, don’t forget to check out their Twitter/Linkedin. you can find the program of the conference. Here The Nuxt Web - Co-Founder of Nuxt Sebastien Chopin Our mission at is to provide the best Developer Experience to deliver the best User Experience for your end users. We are doing this with Nuxt for six Years now. @nuxtlabs Nuxt is a web framework for creating of Vue app. That means you can do: any kind Servier Side Rendering (SSR) Static Site Generation (SSG) Client Side Rendering (CSR) Edge Side Rendering (ESR) thanks to Nuxt 3 Since the beginning of Nuxt in October 2016 it has been downloaded about 46M times on npm, has 300k live websites, and 18k GitHub contributors. A comparison between Nuxt 2 and Nuxt 3 For the development and production web server, Nuxt 2 uses to connect, which is the core of Express.js, but for Nuxt 3 they created h3 and it works in any kind of JavaScript environment. Web server: Webpack 4 is used in Nuxt 2, and for Nuxt 3 they started with Webpack 5, but after the release of Vite, it is officially supported as the default Bundler. Bundler: Vue 2 in Nuxt 2 and Vue 3 in Nuxt 3 UI framework: Vue Router 3 in Nuxt 2 and Vue Router 4 in Nuxt 3, which will not be included if no directory is found. Routing library: pages/ Since Nuxt does SSR, it’s important to be able to manage meta tags. Vue Meta is used in Nuxt 2 and VueUse Head in Nuxt 3. Meta management: This is introduced in Nuxt 3, which compacts the nuxt application for production by removing the node_modules so your Nuxt application will be about 1 megabyte. Server(less) packager: This is a hello world bundle size comparison between Nuxt 2 and Nuxt 3: One of the reasons Nuxt 3 is so much smaller is that it's tree shakable, like Vue 3. So if you don't use a feature, it will not be included in the bundle. The Next equivalent bundle size is around 80kB. In that Nuxt 3 JS bundle, 25.3kB are from Vue, and the 8.7kB left includes: App entry with hydration Root component with <Suspense> Universal and lightweight router: , and useRouter() middleware <NuxtLink> Head composable & components: , , , , … useHead() <Title> <Meta> <Script> Universal data fetching: $fetch() Default error pages: and 404 500 and logic Plugins runtimeConfig composable and hooks: , , , … useNuxtApp() app:created app:mounted page:start Edge Side Rendering Edge Side Rendering - also named JavaScript containers by Ryan Dahl - is the capability of running JavaScript at the CDN nodes somewhere around the world close to the end user. The advantages: Runs milliseconds from end users 0ms cold starts No servers to maintain Automatic scaling Affordable Some of the Edge computing providers right now: CloudFlare Workers Vercel Edge Netlify Edge Deno Deploy Lambda Edge StackPath Content V2 Nuxt Content is a that reads Markdown, YAML, CSV, and JSON files in the directory to create a powerful data layer for your application and is out now with more features. Nuxt module content/ Content V2 Nuxt 2.X What’s coming for Nuxt 2.X: Vue 2.7 Nuxt Bridge (beta) Vite Nitro Composition API & Script setup TypeScript nuxi CLI PostCSS 8 Nuxt 3.0 What's coming for Nuxt 3.0: Nuxt Image Hybrid rendering: Server + SWR + Pre-rendering Full static generation Preview mode Server sessions and Auth Service Workers (PWA) SEO helpers and i18n support End of the Eighth Talk I hope you enjoyed this part and it can be as valuable to you as it was to me. Over the next few days, I’ll share the rest of the talks with you. Stay tuned… Also published . here

Vue Amsterdam 2022: Part VIII - The Nuxt Web

Vue Amsterdam 2022: Parte VIII - The Nuxt Web

Vue Amsterdam 2022 : Partie VIII - Le Web Nuxt

Vue एम्स्टर्डम 2022: भाग VIII - द नक्स्ट वेब

Vue アムステルダム 2022: パート VIII - Nuxt Web

Vue Amsterdam 2022: Parte VIII - A Nuxt Web

Vue Amsterdam 2022: Phần VIII - Web Nuxt

Vue Amsterdam 2022：第八部分 - Nuxt Web

Stories, News & Mentions about NIXTLA

Nixtla’s Stories on HackerNoon