Command-line interface
The acteval command evaluates prediction columns directly from a CSV file.
It uses the same validation, task defaults, metrics, and result metadata as the
Python API.
Input
The shortest form expects columns named y_true and y_pred:
acteval evaluate predictions.csv --task claim_frequency
ActEval requires the task because frequency, severity, and pure-premium evaluations have different valid metrics and defaults. Use custom column names and select metrics explicitly when needed:
acteval evaluate predictions.csv \
--task claim_frequency \
--observed actual_claim_count \
--predicted predicted_frequency \
--metric rmse \
--metric poisson_deviance \
--metric ae_ratio
For rate inputs weighted by policy exposure, name the exposure column and declare the scale explicitly:
acteval evaluate predictions.csv \
--task claim_frequency \
--exposure policy_years \
--input-scale rate
Add --sample-weight COLUMN for a separate weighting column. When exposure and
sample weight are both present, their product is the effective weight, matching
the Python API.
Output
Without an output path, the command prints a metric table. The output file extension selects CSV, structured JSON, or a standalone HTML report:
acteval evaluate predictions.csv --task pure_premium --output report.json
acteval evaluate predictions.csv --task pure_premium --output report.csv
acteval evaluate predictions.csv --task pure_premium --output report.html
Existing outputs are protected by default. Pass --force to replace one. The
input CSV itself is never accepted as the output path.
Run acteval evaluate --help for every option. python -m acteval is an
equivalent entry point.