Humanloop CLI
Reference for the Humanloop CLI

humanloop init <directory>

Creates a new Humanloop Programmatic project in the specified directory. This will initialise that directory with a "main.py" file where you can control what data is loaded and manage any data pre-processing that you want to happen before staring the Programmatic app.

humanloop run <directory>

Starts the Humanloop application and loads the data and previous labelling functions found in <directory>.
If no directory is specified this command will raise an error.

humanloop run-all-functions <directory>

Runs all labelling functions found in <directory>, regardless of their current status.
If no directory is specified this command will raise an error.

humanloop aggregate <directory>

Aggregate and denoise results found in <directory>.
Note that Programmatic will first run any outstanding labelling functions (e.g. labelling functions that have not been run, or have been modified since they were last run) before aggregating results.
If no directory is specified this command will raise an error.

humanloop export --output <file> <directory>

Export results found in <directory> to <file>.
The export will be a JSON that includes all datapoints and their associated results, aggregate results, and ground truths.
For large projects with many datapoints, we recommend using the --ndjson flag. This avoids loading all results into memory and can thus avoid out-of-memory errors.
If the --ndjson argument is specified, Programmatic will create a folder with the name specified by --output, excluding the .json extension if provided. This folder will contain the following files: labels.ndjson, functions.ndjson, and datapoints.ndjson. (Note that the "ndjson" specification is effectively the same as "jsonl" or "jsonlines".)
If no directory is specified this command will raise an error.

humanloop info

Print information about the Programmatic installation.