TimeSeriesSplitByDate

TimeSeriesSplitByDate is a scikit-learn-compatible cross-validator for machine learning with irregular timestamps.

It is designed for datasets where index-based splits can misrepresent real time windows because observations are unevenly spaced.

Key capabilities:

• Date-based splitting modes: days, weeks, and months
• Configurable week start (monday or sunday) for weekly folds
• Deterministic chronological train/test boundaries from datetime values
• Fallback to sklearn TimeSeriesSplit when no date source is provided

Use this package when you need validation folds that reflect calendar time, not just row counts.