Support time series anomaly algorithms

dotnet / machinelearning

ML.NET is an open source and cross-platform machine learning framework for .NET.

https://dot.net/ml

MIT License

9.03k stars 1.88k forks source link

Support time series anomaly algorithms #163

Closed vejadu closed 6 years ago

vejadu commented 6 years ago

Issue

Would love to have/add support for time series forecasting/anomaly detection. There's some pretty common scenarios such as

different types of forecasting such as HW (TES and DES)
online aspect of these algorithms (performant; so no retraining/evaluation is necessary to be integrated in a streaming fashion)
PScore/ZScore/threadhold evaluation for anomalies

GalOshri commented 6 years ago

Absolutely! This is on the roadmap. Are there any specific scenarios you are trying to enable first?

vejadu commented 6 years ago

Effectively HW TES w/ PScore thresholding w/ auto tuning (alpha/beta/gamma) and retraining every N intervals w/ the ability to serialize/deserialize the model to pick up where we left off in the forecasting. (it's what I currently have working in my current usage scenario.)

veikkoeeva commented 6 years ago

Adding here for the benfit of others: https://github.com/Spreads/Spreads.

abgoswam commented 6 years ago

DRI RESPONSE : Adding support for time series is being tracked using issue #978. Planning to close this issue (duplicate of #978)