This repository contains hands-on Jupyter notebooks that demonstrate the basics of time series forecasting using Nixtla's ecosystem, including StatsForecast, mlforecast, NeuralForecast, and TimeGPT.
-
Introduction_to_Nixtlaverse.ipynb Introduction to classic time series forecasting using StatsForecast. Covers data exploration, decomposition, classic models, and rolling cross-validation.
-
Introduction_to_MLForecast.ipynb Machine learning approaches for time series forecasting using mlforecast. Includes feature engineering, local vs global models, and validation strategies.
-
Introduction_to_NeuralForecast_TimeGPT.ipynb Forecasting with NeuralForecast and TimeGPT, including neural network models and foundation models for time series.
-
utilities.py
Utility functions for time series decomposition and plotting, used by the notebooks. -
retail_sales.parquet
Example dataset: Monthly sales data for various countries. -
retail_sales_product_level.parquet
Example dataset: Monthly sales data at the product level.
This repo uses uv. You can run uv sync to install the environment.