Flow Forecast: A deep learning for time series forecasting, classification, and anomaly detection framework built in PyTorch

Abstract

Flow Forecast (FF) is a multi-purpose deep learning for time series forecasting, classification, and anomaly detection framework that contains state of the art time series models. Flow Forecast utilizes modular code design, unit/integration tests, model/prediction visualizations, and native cloud provider integration in order to allow researchers to rapidly experiment with new model architectures, benchmark their results on popular datasets and reproduce their results. Simultaneously it aids industry data scientists to deploy models to production, periodically retrain models, and explain model decisions to stakeholders.

Tasks

Reproductions