Roadmap

Project Utility

• Provide users an easy to use platform that contains recent research models applicable to any time series task.
• Allow users to easily track experiments and explore best hyper-parameters.
• Enable users to easily combine multiple modalities of data to improve performance.
• Facilitate explainability with easy to use dashboard so both ML users and non-technical stakeholders can gain insights into model predictions.

Current Objectives

Objective	Overview	Priority	Related Issues
Meta-data incorporation	We want to enable users to incorporate static meta-data into forecasts. To do this we need to add modules for auto-encoders, semi-supervised embeddings techniques as well as model synthesis	Medium	Meta-Data Embedder Training Module Meta-Data embedder fusion params
Deployment/extended evaluation interface	We want to allow users to deploy models easily to production environments without having to refactor outputs. We also want to enable continuous evaluation as new temporal data appears.	High	Inference mode Continue Evaluation
Platform Integration	We need to enable integration with other commonly used platforms such as Amazon Web Services, Microsoft Azure and GCP	High	Dataverse and G-Cloud integration
Multi-task learning support	Multi-task learning often enables more robust models that can address multiple business needs at once.	Medium	Add multitask support in decoder
Adding additional base models	There are tons of time series forecasting models available. Currently, our repository only has a fraction.	Medium	Add XC Attention Model Refactor DA-RNN
Adding more loss functions	There are tons of useful loss functions that can increase performance at forecasting/classification	Medium
Documentation	Documentation is essential for any widely used repo. We want to add easy to use docs to enable user of flow to easily use and extend our repo to suit their business and research needs.	Medium
Interpretability	We want to make predictions interpretable to various stake holders (both technical and non-technical)	High
Bug Fixes	Obviously existing bugs that lead to incorrect results or unexpected behavior should be fixed.	Very High	Fix DA-RNN GPU Bug

Long term outlook