forecastEval 0.2.4

A lightweight Python framework for rigorous and statistically grounded forecast evaluation, with baseline comparison, horizon-stratified analysis, and Diebold–Mariano testing.

forecastEval

forecastEval is an open-source Python library that provides a lightweight and unified framework for rigorous evaluation of time-series forecasts.

The library is designed to address common shortcomings in forecasting practice, including insufficient baseline comparison, over-reliance on aggregated error metrics, and lack of statistical validation of performance differences.

---

Key Features

• Baseline-aware evaluation

  • Automatic comparison against persistence (naïve) and seasonal naïve baselines
  • Mean Absolute Scaled Error (MASE) and skill scores for interpretable performance assessment

• Horizon-stratified analysis

  • User-defined forecast horizon windows
  • Detection of horizon-dependent performance degradation

• Statistical validation

  • Diebold–Mariano test for forecast comparison
  • Autocorrelation-adjusted variance estimation

• Interpretative reporting

  • Clear PASS / FAIL recommendations for model deployment
  • Human-readable console summaries

• Interactive HTML reports

  • Collapsible sections, visual summaries, and horizon-wise breakdowns

---

Installation

pip install forecastEval