User guide

These pages cover the CLI task by task. Each one is self-contained, so you can read the one that matches what you’re doing.

Every command is shown with the output it produces. Long listings are cut, marked with ..., and the result count at the end of the block is the real one. Counts that depend on your sktime version or your installed packages differ in your terminal.

Finding estimators

Search sktime’s registry by scitype, name, and capability tag, and read an estimator’s parameters, tags, and dependencies before you use it.

Working with data

Load built-in and remote datasets, inspect an unfamiliar file, convert between formats, and split a series for backtesting. Includes the index and format conventions the CLI applies.

Fitting and evaluating models

Spec strings, pipelines and ensembles, parameter overrides, saved model artifacts, and backtesting with run evaluate.

Scoring, checking, and catalogues

Score predictions against observations, validate an object against sktime’s API contract, and browse the benchmark catalogues sktime ships.

Output formats and errors

The five output formats, the stdout and stderr split, the structured error record, and the exit code table.

Using sktime-cli from an agent

The JSON contract an agent should follow, and the skill file that teaches an agent to use the CLI.