search

medic / chic-ts-outlierdetect

Time series anomaly detection module
Apache License 2.0
1 stars 1 forks source link

Implement better train/test split functionality #1

Closed amrdixon closed 2 years ago

amrdixon commented 2 years ago

At a high-level, it would be great to implement a train/test split feature.

Currently, the code performs a fit over an entire signal. When implementing the .fit() method, you can choose a subset to fit the model on or, as is the case with the TimeSeriesUnivariateRollingWindow class, you can implement a rolling window fit function. Outlier detection can be done using the entire residual signal or with a rolling window.

It would be nice to have functionality to perform any fit function on the first n data points and test on the remaining.

amrdixon commented 2 years ago

Repo instead focuses on either fitting the entire signal or fitting/analyzing using a rolling window. This functionality is ultimately not needed here. Closing issue.