How to use mlforecast with gradient boosting models like LightGBM?

Simply pass LightGBM's regressor (e.g., LGBMRegressor) to the models list in MLForecast, as shown in the Quick Start example. Ensure the model follows scikit-learn API for seamless integration and training across time series.

mlforecast vs Prophet: which is better for time series forecasting?

mlforecast excels at scalable, ML-based forecasting with support for distributed computing and exogenous variables, while Prophet is better for interpretable, univariate forecasting with built-in seasonality. Choose mlforecast for production-scale ML workflows and Prophet for simpler, explainable models.

Can mlforecast handle real-time streaming predictions?

No, it's optimized for batch forecasting with recursive feature updates, making it less suitable for real-time latency. For streaming, you'd need to implement custom pipelines or use lighter libraries, as mlforecast focuses on scalable, offline training and prediction.

What are the best practices for cross-validation in mlforecast?

Follow the cross-validation guide in the documentation, which includes techniques for robust performance evaluation and cached CV splits. This helps in hyperparameter tuning and avoiding data leakage, especially for multiple time series.

How to scale mlforecast to millions of time series using Dask?

Set up a Dask cluster and use mlforecast's Dask backend support, as outlined in the distributed training guide. This allows parallel processing of features and model training, efficiently handling large-scale datasets without memory issues.

Does mlforecast support multivariate time series with exogenous variables?

Yes, it supports exogenous variables and static covariates to enhance model accuracy, as mentioned in the Key Features. You can include additional columns in your dataframe for training, and mlforecast will incorporate them during feature engineering.

Open-Awesome

mlforecast

Apache-2.0Pythonv1.0.31

A Python framework for scalable time series forecasting using machine learning models, designed for production environments.

Visit Website GitHub

1.2k stars125 forks0 contributors

What is mlforecast?

mlforecast is a Python framework for time series forecasting using machine learning models. It provides fast feature engineering and scalable training to handle massive datasets, addressing the speed and accuracy limitations of existing tools. The framework is designed for production environments, supporting distributed computing backends like Dask, Ray, and Spark.

Target Audience

Data scientists and ML engineers working on time series forecasting projects that require scalability, performance, and integration with standard ML libraries like scikit-learn.

Value Proposition

Developers choose mlforecast for its exceptional speed in feature engineering, seamless scalability to clusters, and familiar scikit-learn API, making it ideal for production forecasting tasks across large numbers of time series.

Overview

Scalable machine 🤖 learning for time series forecasting.

Use Cases

Best For

Forecasting millions of time series efficiently in production
Integrating machine learning models into time series workflows
Scaling forecasting tasks using distributed computing backends like Dask or Ray
Adding probabilistic prediction intervals to forecasts
Leveraging exogenous variables and static features in time series models
Performing cross-validation and hyperparameter tuning for forecasting models

Not Ideal For

Projects requiring classical statistical forecasting models like ARIMA or ETS
Real-time applications needing sub-second prediction latency
Teams with limited infrastructure for distributed computing setups
Forecasting tasks where model interpretability is more critical than predictive accuracy

Pros & Cons

Pros

Blazing Fast Feature Engineering

mlforecast boasts the fastest implementations of lag, rolling, and expanding window features in Python, as highlighted in the Features section, enabling efficient training on large datasets.

Seamless Multi-Backend Scalability

It offers out-of-the-box compatibility with pandas, polars, Spark, Dask, and Ray, allowing easy scaling from single machines to distributed clusters, per the Key Features and documentation.

Familiar Scikit-learn API

The framework uses a standard .fit and .predict interface, making integration with existing ML workflows straightforward, as emphasized in the Quick Start and Features list.

Probabilistic Forecasting Support

Includes Conformal Prediction for prediction intervals, providing uncertainty quantification in forecasts, detailed in the examples and guides for robust model evaluation.

Cons

ML-Only Model Limitation

It lacks built-in support for traditional time series models like ARIMA, forcing users to rely on external libraries or sister projects like statsforecast for classical methods.

Complex Distributed Setup

Leveraging backends like Spark or Dask requires additional infrastructure and expertise, adding operational overhead that may be prohibitive for small teams or simple deployments.

Feature Configuration Overhead

Configuring lag transforms, target transforms, and date features can be intricate, and the documentation assumes prior knowledge of time series feature engineering, potentially steepening the learning curve.

Frequently Asked Questions

Related Projects

Prophet

Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.

Stars20,158

Forks4,638

Last commit25 days ago

FinRL

FinRL®: Financial Reinforcement Learning. 🔥

Stars15,042

Forks3,319

Last commit28 days ago

sktime

A unified framework for machine learning with time series

A python library for user-friendly forecasting and anomaly detection on time series.

Stars9,356

Forks1,001

Last commit2 days ago

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub