Skip to content

CLI Reference

Jump to bottom
Giacomo Saccaggi edited this page Jun 19, 2026 · 1 revision

CLI Reference

scomp-link provides 13 commands. Every command supports --help for full options.

Commands

Command Description
init Scaffold a new project
run Train and evaluate a complete ML pipeline
predict Generate predictions from a .scomp artifact
explain Compute SHAP feature importance
engineer Apply automated feature engineering
forecast Time series forecasting
anomaly Multi-method anomaly detection
drift Detect data drift (PSI + KS)
fairness Check bias metrics
quality Generate data quality report
report Generate interactive HTML report (EDA or model evaluation)
compare Compare multiple artifacts
info Inspect a .scomp artifact

init

scomp-link init my_project [--force]

Creates a project directory with pipeline.py, config.yaml, README.md, .gitignore, and data/, models/, reports/ folders.

run

scomp-link run --data train.csv --target price --task regression \
  [--features col1,col2] [--model-hint numerical_prediction] \
  [--test-size 0.2] [--save-artifact model.scomp] [--output results.json] \
  [--engineer] [--ensemble voting|stacking] [--advanced-cv] [--silent]
Flag Description
--engineer Apply feature engineering before training
--ensemble voting|stacking Enable ensemble learning
--advanced-cv Run LOOCV + Bootstrap validation

predict

scomp-link predict --artifact model.scomp --data new_data.csv [--output predictions.csv] [--silent]

Outputs all input columns + a prediction column. Automatically drops the target column if configured.

explain

scomp-link explain --artifact model.scomp --data test.csv [--n-samples 100] [--output importance.csv] [--silent]

engineer

scomp-link engineer --data raw.csv --target y \
  [--interactions] [--log-transform] [--date-features] \
  [--target-encode] [--auto-bin] [--n-bins 5] \
  [--output engineered.csv] [--silent]

Each flag enables a specific transformation. With no flags, outputs unchanged data.

forecast

scomp-link forecast --data series.csv --column value \
  [--horizon 10] [--method auto|arima|sarima|exp_smoothing] \
  [--seasonal-period 12] [--cv-splits 5] [--output forecast.csv] [--silent]

Output includes forecast, lower, and upper (95% confidence interval).

anomaly

scomp-link anomaly --data data.csv \
  [--features col1,col2] [--methods iforest,lof] \
  [--contamination 0.05] [--consensus 2] [--output anomalies.csv] [--silent]

Available methods: iforest, lof, tabnet, transformer.

drift

scomp-link drift --reference train.csv --current production.csv \
  [--features col1,col2] [--threshold 0.2] [--output drift_report.csv] [--silent]

fairness

scomp-link fairness --data preds.csv --target y_true --predicted y_pred --sensitive gender \
  [--output report.json] [--silent]

quality

scomp-link quality --data raw_data.csv [--output report.html] [--silent]

report

# EDA report
scomp-link report --data train.csv --output eda.html

# Model evaluation report
scomp-link report --artifact model.scomp --data test.csv --output eval.html

compare

scomp-link compare --artifacts v1.scomp v2.scomp v3.scomp [--output comparison.csv]

info

scomp-link info --artifact model.scomp

Output (JSON): config, metrics, model type, n_features, metadata.

Supported File Formats

Extension Read Write
.csv ✅ ✅
.tsv ✅ —
.parquet ✅ ✅
.json — ✅
.html — ✅

Clone this wiki locally