Initialize the data directory

dataset_initialize(path) records path as the local mirror of the bucket. The choice is persisted across Python sessions, so subsequent calls on the same machine are unnecessary – Subject / Group loaders pick the configured directory up automatically.

from laion_fmri.config import dataset_initialize, get_data_dir

dataset_initialize("./laion_fmri_data")
get_data_dir()        # → "./laion_fmri_data"

This step writes nothing to S3 and does not require AWS credentials. It only sets up the local destination for future download(...) calls and makes Subject / Group loaders aware of where the data lives.

Requirements

• The directory must already exist. dataset_initialize raises FileNotFoundError if the path is missing – it doesn’t create the directory for you, so a typo can’t silently scatter dataset files in the wrong place. Create it yourself first (e.g. mkdir -p ./laion_fmri_data).

• The path argument must be a string – TypeError otherwise.

What gets created

The package records your license acceptance privately inside the data directory so subsequent download(...) calls don’t re-prompt. If you want to start over with a fresh acceptance flow (e.g. on a new dataset clone), delete the data directory or its hidden configuration folder. The bucket mirror itself is laid out under derivatives/, stimuli/ etc. directly in the data directory once download(...) runs.

Switching data directories

Calling dataset_initialize again with a different path overwrites the persisted choice and points all subsequent loaders at the new location. The old directory is left on disk untouched – laion_fmri never deletes data for you.

dataset_initialize("./scratch_run")    # active until next call
dataset_initialize("./project_run")    # now active; first dir untouched

If no data directory has been initialized yet, get_data_dir() raises DataDirNotSetError with a hint to run dataset_initialize.