openstef_beam.evaluation#

Organizes forecasting results into structured performance reports.

After running backtests, you get lots of predictions and actual values. This module helps you make sense of them by calculating metrics across different time periods, filtering for specific conditions (like weekends or peak hours), and organizing everything into clear reports.

What it handles:

  • Time windows: Compare performance across days, weeks, seasons

  • Lead times: Evaluate how accuracy changes from 1-hour to 48-hour forecasts

  • Data filtering: Focus on specific conditions (peaks, weekdays, etc.)

  • Metric calculation: Apply the right metrics to the right data subsets

  • Report structure: Organize results for easy analysis and comparison

The evaluation produces raw numerical data (metrics, timestamps, values) that can be fed into the analysis module for visualization and further interpretation.

Classes#

EvaluationConfig(**data)

Configuration for the evaluation pipeline.

EvaluationPipeline(config, quantiles, ...)

Pipeline for evaluating probabilistic forecasting models.

EvaluationReport(**data)

Complete evaluation report containing results for all data subsets.

EvaluationSubsetReport(**data)

Container for evaluation results on a specific data subset.

SubsetMetric(**data)

Container for evaluation metrics computed on a data subset.

Window(lag, size[, stride, minimum_coverage])

Represents a time window with lag, size, and stride parameters.