Flow-Forecast PyData Global Sprint πŸƒ 2020 (11-10 to 11-18)

** Attention PyData participants: Even though our official sprint is listed as being on Friday. We will be running our sprint all week long 11/10-11-18. We encourage you to participate on whatever days work best for you.

Why participate?

Flow forecast is an open-source deep learning for time series framework licensed under GPL-3.0 that aims to make it really easy to leverage deep learning for time series forecasting and classification. By participating in the sprint you will be

  • Contributing to an open source project that is currently being used to address many AI4Good initiatives such as COVID-19, river/stream flooding, clean energy forecasting, and much more. Additionally, many researchers and small business rely on flow-forecast to give them accurate temporal forecasts πŸ“ˆ.

  • Improving πŸ’ͺ your general machine learning, programming, cloud infrastructure and time series analysis skills.

  • Learning the internal mechanics of flow-forecast so you can utilize at your company πŸ€“πŸŽ‰ when the need arises (trust me at some point your company will need an effective forecast).

  • Building your open source profile (who doesn’t like a green commit history and real PRs they can share with future employees).

  • Earning cool flow-forecast swag πŸ‘•πŸ‘’πŸŽ©

How can I participate?

The sprint officials begins on 11/10 1pm EST. In the meantime you can do the following:

We plan on having plenty of issues suitable for all skill levels; anything from fixing documentation to porting the latest state of the art models. Or if you are interested in seeing how flow-forecast models perform on your own (public) dataset then we would be happy to help you get started. The main goal of this sprint is to add more deep learning time series models, benchmark existing models and improve our cloud provider integration. However, we will also welcome PRs that solve documentation issues or bugs.

Primary Contacts

@Isaac Godfried

@kriti mahajan

Sprint Timeline

Issue Board

Issues

Category

Required Skills

Complexity

Assignee

Issues

Category

Required Skills

Complexity

Assignee

Add Temporal Fusion Transformer

New models

PyTorch, refactoring research code, Pandas

10-20 hours depending on developer skill.

Β 

Add DSA-Net Model

New models

PyTorch, refactoring research code

10-20 hours depending on developer skill.

Β 

Add Deep-AR Model

New models

PyTorch

5-10 hours depending on developer skill

Β 

Add GRU-Bayes Model

New models

PyTorch, Pandas, refactoring skills

15-25 hours

Β 

Integrate Neural ODE models/library with flow

New models

PyTorch, Pandas, ODE knowledge,

30+

Β 

Add MASE loss function

LOSS FUNCTIONS

PyTorch, Statistics

5-10 hours

Β 

Increase GPU Utilization

INfrastructure

PyTorch (data-loaders), GPU settings, Wandb

10 hours?

Β 

Create Flow Training Docker Image

INfrastructure

Docker, Python

Β 

Β 

Add more tutorials

Tutorials

PyTorch

Β 

Β 

Jira Issue Status Board