Command line

fitrank [--verbose | --quiet] COMMAND [ARGS]
fitrank --version
fitrank COMMAND --help

Command	What it does
fitrank run CONFIG	Evaluates every candidate, shows the leaderboard and saves the results folder.
fitrank validate CONFIG	Checks the config and the data, trains nothing.
fitrank init	Writes a commented starter config.
fitrank predict MODEL DATA	Predicts new data with a saved model.
fitrank list models	Lists the model names, for each task.
fitrank list reductions	Lists the reduction names.

run

fitrank run examples/iris.yaml

Shows a progress bar, then the leaderboard on the standard output, and saves the results in <output_dir>/<name>_<date>_<time>/. Everything except the leaderboard (progress, warnings, errors) goes to the error output, so fitrank --quiet run config.yaml > leaderboard.txt keeps only the table.

validate

fitrank validate examples/titanic.yaml

Config OK: examples/titanic.yaml
Data OK: 1309 rows, 7 features, target 'Survived'
8 candidates: 2 reduction(s) x 4 model(s), 5-fold cross-validation

Reads the data and runs every check of run, without training. All the problems are listed at once.

init

fitrank init --data data.csv --target price --task regression

Option	Default	Description
-o, --output	config.yaml	The file to write.
--task	classification	classification or regression.
--data	—	A data file: fills data.path, the feature list and suggested preprocessing groups.
--target	—	The column to predict.
--force	off	Overwrites an existing file.

The suggested preprocessing groups are commented out: nothing is applied until you uncomment them.

predict

fitrank predict runs/iris_2026-11-01_15-36-06/best_model.joblib new_flowers.csv -o predictions.csv

Adds a prediction column to the data and writes it as CSV, to the standard output or to -o FILE. The data needs the same feature columns as for training. A warning is shown when the model was saved with another scikit-learn version.

list

fitrank list models --task regression
fitrank list reductions

Global options

Option	Description
--quiet, -q	Only the leaderboard, for scripts.
--verbose, -v	One line per candidate as soon as it is done, and tracebacks on errors.
--version	Shows the FitRank version.

Exit codes

Code	Meaning
0	Success (warnings included).
1	Configuration error: unknown key, model, step or metric, wrong value.
2	Data error: file not found, missing column, missing values without an imputer...
3	Every candidate failed.

Errors are shown in red, without a traceback: what is wrong, where, and how to fix it. Warnings are shown in yellow and never stop the command.