Nixtla

🚀 Introducing v2 of our TimeGPT API: Faster, Smarter, and More Powerful! 🚀 We’re thrilled to unveil the latest release of our API—v2, packed with incredible improvements driven by the community’s feedback. The nixtla API lets you seamlessly connect with TimeGPT. Just pip install the newest version of our SDK (v0.6.0), nixtla, and you will have access to these new features. With v2, we’ve focused on what matters most: speed, scalability, flexibility, and precision. Whether you’re working on anomaly detection, forecasting, or cross-validating TimeGPT, these enhancements will enable you to achieve better results, faster. ⚡ Unmatched Speed Improvements One of the standout upgrades in v2 is the dramatic increase in computational performance. We’ve fine-tuned our algorithms and optimized our infrastructure, delivering staggering results: we can detect anomalies 8.9x faster, forecast with exogenous variables 10x, and cross-validation 6x faster than the v1 of our API. These speedups aren’t just numbers—they represent a huge leap in efficiency, allowing you to run complex analyses in a fraction of the time. This is especially crucial in production environments where time-to-insight is key. ⏱️ 🌐 1 Billion Time Series in 6 Hours But that’s not all. With v2, we’ve shattered previous limits. In our latest experiment, we successfully forecasted 1 billion time series in just 6 hours. This unprecedented capability sets a new standard for scalability in time series forecasting, empowering organizations to handle massive datasets with unparalleled speed. 🚀 📊 Advanced Handling of Exogenous Variables We’ve also introduced a highly requested feature: the ability to distinguish between future and historical exogenous variables. You can now leverage past data or future data to fine-tune forecasts even further, boosting the accuracy of your models for a predictive edge. 🔍 🔍 Enhanced Model Explainability with SHAP Values In v2, we’ve also integrated SHAP values to enhance model interpretability. SHAP values allow you to understand the impact of each feature on TimeGPT’s predictions, providing deeper insights into the decision-making process. This is particularly valuable for model explainability and trust, especially in critical applications. 🧠 🛠️ New Integration with Polars In addition to these improvements, we’ve added support for Polars —a lightning-fast DataFrame library. With Polars, you can process large datasets more efficiently, making it easier to manage and manipulate your time series data. This perfectly complements our existing integrations with Dask, Ray, Spark, and Pandas. What This Means for You We’ve listened to your feedback and made the changes you need to push the boundaries of what’s possible with time series forecasting. We’re eager to hear your thoughts and continue improving. 💙 Happy forecasting! #TimeSeries #Forecasting #MachineLearning #BigData #Nixtla #Polars #API #DataScience #AI #Scalability

Forecasting ~1,800 Store-Product Combinations using Polars, Nixtla and Catboost 🐱. In my most recent project I take on the Ecuador Store Sales dataset which involves forecasting 33 different product categories across 55 different stores. This project features lots of data cleaning and manipulation, timeseries analysis/EDA, feature engineering, a 2-step feature selection pipeline using feature engine and tuning catboost using optuna. I was pleased with the end result, outperforming statistical baselines by 35-40% and achieving a final MAE% of 15%, with a bias of -3% across all series. You can find this project on my github here: https://lnkd.in/g6XBPHf9 I am definitely buying the Catboost hype for forecasting... very powerful! #timeseriesforeacsting #datascientist #supplychain #catboost

More on foundation models for time series forecasting in Part 2 of this IIF podcast! In this continuation of the conversation, Azul Garza Ramírez and Mononito Goswami share about: ⛅ Challenges and opportunities presented by foundation models in time series forecasting 📊 The need for massive, diverse datasets for these models to best reach their potential 📈 Their views for the future of time series forecasting "This episode underscores the rapid advancements in time series forecasting and the growing importance of foundation models in pushing the boundaries of what’s possible in this field." 🙏🏽 Thank you for the great conversation Mariana Menchero and Faranak Golestaneh, PhD! We so enjoyed and appreciated the opportunity to join you and Mononito for this thoughtful discussion.

Thanks Rami Krispin! Love the newsletter, and so excited that statsforecast⚡was featured as your first Library of the Week. In the newsletter Rami Krispin shares a great example of using statsforecast to forecast the demand for electricity in California. Code included. 💙

Last month, we hosted the inaugural Open Source Nights🌛 at GitHub's HQ in San Francisco. We had a full house and the talks delivered were outstanding. Here's what happened 👇 📈 Azul Garza Ramírez from Nixtla delivered a live demo on how to forecast 1 billion time series using Time-GPT1, the first foundational model for time-series forecasting. Forecasting 1 billion time series with AutoArima would take 4,983 days, but Time-GPT-1 can do it in just 8 hours! 👨🔬 Jürgen Cito from Vienna University of Technology presented the latest on HackingBuddy GPT, an LLM agent framework for penetration testing based on academic research. It helps security professionals use AI for hacking, all in 50 lines of code or less. Impressively, in benchmarking experiments, HackingBuddy GPT achieves an 83% success rate, outperforming humans in some cases! 🔭 Denzell Ford from Trieve (YC W24), spoke about why Trieve is "source available" instead of open source. He shared how their source-available approach is helping them offer transparent and streamlined customer support and why more companies should follow their lead. 🔮 Finally, our CEO Rodrigo Mendoza-Smith explained how Quira uses Machine Learning & data-driven approaches to help open-source organisations grow their communities on auto-pilot. He also addressed some of the problems related to the future-of-work and education that Quira is solving through Quests. The full talks will be made available soon! Also, more Open Source Nights are on the way and may soon come to a city near you 👀. Follow us on LinkedIn or sign up for our newsletter to stay tuned.

🧠 Unleash the magic of NeuralForecast? Yes please! ✨ Marcel Boersma, PhD's Unleash the Magic of NeuralForecast: A Practical Guide to Time Series Transformation and Model Building is a great tutorial that starts with data preparation and goes through to building and running a model. "After this tutorial, you should be ready to build your models on top of the NeuralForecast library, letting you focus on building the best layers in your model instead of writing tons of code to prepare the data and track experimentation. Let’s get started with the basics and end this tutorial with a blast!" 💙 He made his code available in Colab too. https://lnkd.in/gga23KRt

I'm thrilled by the overwhelming response from the Forecasting Data Science community to my article "𝐂𝐚𝐧 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐎𝐮𝐭𝐩𝐞𝐫𝐟𝐨𝐫𝐦 𝐒𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐌𝐨𝐝𝐞𝐥𝐬 𝐟𝐨𝐫 𝐓𝐢𝐦𝐞 𝐒𝐞𝐫𝐢𝐞𝐬 𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠?" on Towards AI. With over 15k views, it was even featured by the Medium Curation team, sparking a surge of interest in Nixtla's MLForecast library. To address this interest, I've written a comprehensive guide titled "𝐇𝐨𝐰 𝐭𝐨 𝐌𝐋𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭 𝐲𝐨𝐮𝐫 𝐓𝐢𝐦𝐞 𝐒𝐞𝐫𝐢𝐞𝐬 𝐃𝐚𝐭𝐚!"—complete with a practical case study. You can check it out on Level Up Coding (Links are in Comments). And as always, it's free to access—because knowledge should be free for all! 𝐇𝐚𝐩𝐩𝐲 𝐅𝐨𝐫𝐞𝐜𝐚𝐬𝐭𝐢𝐧𝐠! #foundationalmodels #airesearch #deeplearning #transformers #genai #aiinnovation #generativeai #aigenerated #creativeai #datascience #bigdata #machinelearning #ai #dataanalytics #datavisualization #gpt #timeseriesforecasting #forecasting #demandforecasting #timeseriesanalysis #businessanalysis

Next week Cristian Challu, Ph.D. will share about TimeGPT-1's origin and technology, and some of the main challenges and opportunities of developing foundation models for time series. 👋🏽 It would be great to say hello to anyone at Ai4 - Artificial Intelligence Conferences!

🎉 🧠 New NeuralForecast release, with new models, new features and better documentation! 🧠 🎉   NeuralForecast offers a large collection of neural forecasting models focusing on their performance, usability, and robustness. Included in the 1.7.4 release: 🔥 the KAN model, a univariate forecasting model that uses Kolmogorov-Arnold Networks (KANs); ⏳ the TimeMixer model, a multivariate forecasting model that uses MLPs and fourier techniques 🐘 Support for datasets that can't fit in memory 🌱 A restructuring and update of our documentation 🙏🏽 💙 Thanks to our community members Fabian Bergermann and Jasmine Rienecker for helping fix bugs and add features in this release! NeuralForecast: https://lnkd.in/g4ww5kQB Release notes: https://lnkd.in/g6ThGxfP