Back to Gap Map Infrastructure / Classic ML & computer vision

NeuralForecast

Nixtla
open source / Overall score: 3.0

NeuralForecast is Nixtla's library of neural forecasting models, from recurrent networks, N-BEATS and NHITS to transformer models such as TFT, PatchTST and iTransformer, all trained and run through one fit-and-predict interface. It handles exogenous variables and probabilistic forecasts, and each model has an Auto version that tunes its settings on Ray or Optuna. The interface is shared with Nixtla's StatsForecast and MLForecast.

Openness

5 high confidence
5.0
license
Apache-2.0 (OSI)
source
public (github.com/Nixtla/neuralforecast)
core features withheld
no — Nixtla's paid offering is the separate TimeGPT service

Apache-2.0 and built from the public repository. Nixtla sells TimeGPT, a closed forecasting model reached through an API, and an enterprise plan built on it; its pricing lists NeuralForecast among the open libraries that are free, with nothing withheld.

Adoption

3 high confidence
3.0

Measured on monthly PyPI downloads of the neuralforecast package.

Capability

3 high confidence
3.0

NeuralForecast is a catalog of deep-learning forecasters with the tools to train and compare them, and its Auto variants tune each model's settings. Choosing which architecture to use is left to the user, as in sktime, whose toolkit is broader.

  • https://raw.githubusercontent.com/Nixtla/neuralforecast/HEAD/README.md recorded 2026-11-12

    README: "The models range from classic networks like RNNs to the latest transformers", listing MLP, LSTM, GRU, RNN, TCN, TimesNet, BiTCN, DeepAR, NBEATS, NBEATSx, NHITS, TiDE, DeepNPTS, TSMixer, DLinear, NLinear, TFT, Informer, AutoFormer, FedFormer, PatchTST, iTransformer, StemGNN and TimeLLM; features "Automatic Model Selection with distributed automatic hyperparameter tuning".

Verified 2026-11-12