README.en.md

model-debugger (English)

model-debugger is a practical toolkit for post-training diagnostics of classification models.

What it does

• Global metrics: accuracy, precision, recall, f1, error_rate
• Segment diagnostics and top failure slices
• Binary fairness diagnostics (selection rate, TPR, FPR, FNR, precision)
• Calibration diagnostics (ECE, MCE) and calibration curve
• Output artifacts: JSON/CSV/PNG/Markdown report

Quickstart

python3 -m venv .venv
source .venv/bin/activate
pip install -e .

python3 examples/generate_synthetic_data.py --rows 4000 --seed 42 --output examples/sample_predictions.csv
python3 -m model_debugger \
  --input examples/sample_predictions.csv \
  --target y_true \
  --pred y_pred \
  --proba y_proba \
  --segments gender,region,device,age_group \
  --fairness-groups gender,region \
  --output artifacts

Main documentation is in Russian: README.md