Before we start, I want to mention that this is adapted from an article I wrote on my personal blog. We are re-starting Data Intelligence consulting work in Latin America, and I thought it only proper to begin by talking about companies that inspire us and help us in our mission to bring Data Science and Business Analytics to companies — helping them bridge the gap between Europe, Asia, and the US.
I am a little jealous about a company I just discovered, called Nixtla. You can check their website here. According to their own words, Nixtla "democratizes access to state-of-the-art predictive insights, eliminating the need for a dedicated team of machine learning engineers."
Tall order, and one I am a little doubtful about — because I am so used to people with no or minimal statistical skills basically butchering forecasting. But hey, maybe that's just me having bad luck in the real world.
The thing I am most envious of about Nixtla is their love for open source and their large codebase contribution to Github. For a company that wishes to democratize predictive analytics, they sure put their money where their mouth is. The open way they are giving away code and examples is enough for more than one consultant to open their own forecasting venture on their excellent libraries alone. I count six repositories with a plethora of predictive analytical code:
- StatsForecast — a collection of widely used univariate time series forecasting models, including automatic ARIMA, ETS, CES, and Theta modeling, optimized for high performance using numba. Also includes a large battery of benchmarking models.
- mlforecast — a framework to perform time series forecasting using machine learning models, with the option to scale to massive amounts of data using remote clusters.
- NeuralForecast — a large collection of neural forecasting models focusing on performance, usability, and robustness. The models range from classic networks like RNN to the latest transformers: MLP, LSTM, GRU, TCN, DeepAR, NBEATS, NBEATSx, NHITS, DLinear, NLinear, TFT, Informer, AutoFormer, FedFormer, PatchTST, StemGNN, and TimesNet.
- HierarchicalForecast — a collection of reconciliation methods, including BottomUp, TopDown, MiddleOut, MinTrace, and ERM. Plus probabilistic coherent predictions including Normality, Bootstrap, and PERMBU.
- tsfeatures — calculates various features from time series data. A Python implementation of the R package tsfeatures.
- Nixtla Open Source Time Series Ecosystem — a compendium of all the above, plus additional classes building on them.
And yes, I listed the whole thing — not because I am lazy, but rather because I feel a little overtaken by this treasure chest of forecasting and predictive goodness.
In an age of increasingly proprietary code and tools, such a degree of technical philanthropy is not unheard of, but not common either. As I said, I feel jealous of Nixtla for being a company so far advanced in predictive analytics that they can contribute to the wider world of data science and make a big impact on the rest of the community.
But then, that is certainly something worth being jealous of.