Flow Forecast
Fiscal Host: Open Source Collective
A deep learning for time series forecasting framework built in PyTorch
Contribute
Become a financial contributor.
Financial Contributions
Recurring contribution
Backer
For $5/month you can help defray our cloud storage costs. Flow Forecast stores several large publicly accessible datasets including FlowDB (publish... Read more
Starts at
$5 USD / month
Flow Forecast is all of us
Our contributors 1
Thank you for supporting Flow Forecast.
About
Flow Forecast is a deep learning for time series forecasting, classification, and anomaly detection framework built in PyTorch. Flow forecast aims to help both researchers and data scientists in industry to train, architect and deploy state of the art deep learning models to a variety of temporal tasks.
On the research side Flow Forecast allows researchers to rapidly experiments with new models, easily benchmark their results across many datasets, and reproduce their results. Flow Forecast's modular code allows researchers to minutely tweak their models and test out hundreds of different configurations with writing just a few lines of code. Flow forecast also supports upcoming areas of research with modules that provide support for multiple loss functions and many data modalities.
At the moment many deep learning techniques are simply published at research conferences but then never utilized or only leveraged by large companies the can afford full-time research labs 😞. Flow Forecast aims to democratize the power of deep learning for time series forecasting by creating easy to use APIs, clear explanations of predictions, hyper-parameter sweeps, native cloud integration, and modules to help with deployment. This way even data scientists and machine learning engineers lacking in direct deep learning experience can leverage state of the art techniques to solve their particular use case and positively impact their business.
Flow forecast also directly powers several major AI4Good time series forecasting initiatives such as the effort to forecast 🦠 COVID-19 cases and the effect of policy interventions at CoronaWhy, 🏞️ predicting flash floods along major US rivers, and forecasting patient vitals in the ICU 🤒.
All of AIStream's work is open, transparent and well documented. Sponsorship is important as it helps pay for cloud infrastructure for experiments, living expenses of core contributors (many of whom work nearly full-time on our projects), and funding sprints/other developer events.
On the research side Flow Forecast allows researchers to rapidly experiments with new models, easily benchmark their results across many datasets, and reproduce their results. Flow Forecast's modular code allows researchers to minutely tweak their models and test out hundreds of different configurations with writing just a few lines of code. Flow forecast also supports upcoming areas of research with modules that provide support for multiple loss functions and many data modalities.
At the moment many deep learning techniques are simply published at research conferences but then never utilized or only leveraged by large companies the can afford full-time research labs 😞. Flow Forecast aims to democratize the power of deep learning for time series forecasting by creating easy to use APIs, clear explanations of predictions, hyper-parameter sweeps, native cloud integration, and modules to help with deployment. This way even data scientists and machine learning engineers lacking in direct deep learning experience can leverage state of the art techniques to solve their particular use case and positively impact their business.
Flow forecast also directly powers several major AI4Good time series forecasting initiatives such as the effort to forecast 🦠 COVID-19 cases and the effect of policy interventions at CoronaWhy, 🏞️ predicting flash floods along major US rivers, and forecasting patient vitals in the ICU 🤒.
All of AIStream's work is open, transparent and well documented. Sponsorship is important as it helps pay for cloud infrastructure for experiments, living expenses of core contributors (many of whom work nearly full-time on our projects), and funding sprints/other developer events.