Release: Model training improvements, lookback period improvements, improved SSO to support BoxyHQ

✨ Feature Highlights

Model Training Improvements

Optimal hyper parameters for machine learning models are now computed and applied on initial run. Ensuring the first run of the monitor runs optimally. Hyper parameters for each model are also periodically improved and recalculated by sifflet.

Lookback Period Improvements

Monitors now alert on the Lookback Period. The lookback period is used to add a past buffer to monitor runs to account for pipelines who update or backfill past data points. This was previously used only to update graphs and improved the model for better predictions.

What Changed: If a data point in the lookback window was within the expected range at the time of the initial run but has since then been changed the Sifflet monitor will treat that change as a new anomaly

Additionally if a data point was an anomaly when it was first generated but a subsequent run identifies that the point is now in the correct range. It will automatically be marked as Fixed

🛠 Fixes

• Improved SSO to support BoxyHQ

App version: v217