CLI Reference¶

Complete reference for all Pilz command-line commands.

Commands Overview¶

Global Options¶

Every command accepts the standard Typer options:

train¶

Train Pilz models for every target value in the datacard.

pilz train --datacard <file.yaml> --trainsettings <file.yaml>

Options¶

Example¶

pilz train --datacard iris_dc.yaml --trainsettings train.yaml

eval¶

Evaluate trained models and generate ROC plots and metrics.

pilz eval --datacard <file.yaml> --evalsettings <file.yaml>

Options¶

Example¶

pilz eval --datacard iris_dc.yaml --evalsettings eval.yaml

Output Files¶

eval writes scores and plots. It does not emit SQL rules or predictions as SQL.

• {target}_roc.html - ROC curve per target value; always written unless infer is used
• all_roc.html - Combined ROC curves; only when not using infer
• multi_class_result.html - Per-class accuracy; only for multi-class targets (len(target.values) > 1) and not using infer
• {out_file} - Prediction scores (CSV/Parquet); only when the out_file setting is set

infer¶

Apply trained models to new data and write the predictions.

pilz infer --datacard <file.yaml> --evalsettings <file.yaml>

Unlike eval, no plots or accuracy metrics are produced, only the predicted scores (and, for multi-class targets, the predicted label). No HTML files are written; {out_file} is written when the out_file setting is set.

Options¶

Example¶

pilz infer --datacard iris_dc.yaml --evalsettings eval.yaml

Use Cases¶

• Fast prediction on new data
• Batch scoring in production
• When you only need scores, not metrics

create-dc¶

Interactively generate a datacard YAML from a source data file.

pilz create-dc --src <path> --out <file.yaml>

Options¶

Example¶

pilz create-dc --src data.csv --out datacard.yaml

Interactive Prompts¶

1. Prompts for the target column until a valid column name is entered
2. Derives each feature's type and statistical automatically from the column dtype (no user prompt)
3. Prompts once for a replacement missing_value for any column containing nulls or float NaN
4. Prompts for "Infos about the dataset" and writes the answer as infos: {bla: <answer>}

convert¶

Rewrite legacy cut-string trees to the compact structured form.

pilz convert --src <folder> --out <folder>

Trees trained before the structured cond representation stored every leaf condition as rendered SQL text; fan-out trees could reach megabyte-sized leaves. This command parses each condition back into the structured cond form and writes the trees to out. Tree structure and leaf scores are unchanged, so eval results are identical — only evaluation becomes fast.

Options¶

Example¶

pilz convert --src iris_model --out iris_model_cond

Quick Reference¶