Data and weights¶

These commands check, download and prepare the datasets and the model weights. See Data and Models.

data¶

Check which datasets are on this machine, download them, and build optional faster-to-read copies.

Usage: neuroatlas data <action>

Command	Description
`data status`	Show which datasets are on this machine.
`data download`	Download datasets, or print how to get them.
`data prepare`	Build the files a bci_cognitive dataset is read from, or a faster copy of an epilepsy dataset.

data status¶

Show which datasets are on this machine, where they were looked for and how to get the missing ones.

Usage: neuroatlas data status [targets ...] [options]

neuroatlas data status
neuroatlas data status sleep_stage
neuroatlas data status siena -v

Option	Description	Default
`targets`	Benchmarks, domains (epilepsy, sleep, brain_age, bci or all) or dataset names, separated by spaces or commas.	all
`-v`, `--verbose`	Add columns for each path, the setting that moves it and the channel map, and explain every value.
`--format {table,csv,md,json}`	Output format. In csv, md and json the notes under a row go to a `note` column.	`table`

data download¶

Download datasets into the folder each one is read from. MOABB datasets go to $MNE_DATA. For datasets that need an account or a request, it prints where to get them.

Usage: neuroatlas data download datasets [datasets ...] [options]

neuroatlas data download sleep_edf_expanded --mirror aws
neuroatlas data download siena --first 5    # a quick test
neuroatlas data download cfs --dry-run

Option	Description	Default
`datasets`	Dataset names, as listed by `neuroatlas list datasets`.
`--dry-run`	Print the plan and run every check (token, disk space, tools), but download nothing.
`--mirror {physionet,aws}`	Where to get PhysioNet datasets. aws is PhysioNet's open-data copy on AWS, which is usually much faster.	`physionet`
`--keep-archive`	Keep downloaded .zip files after unpacking them.
`--first N`	Download only the first N recordings, for a quick test. `check` works on them, but `run` needs the whole dataset. Works for PhysioNet, NSRR and file-by-file Zenodo downloads.

data prepare¶

Build the two files that dreamer_valence, dreamer_arousal, eegmat or arithmetic_task are read from. Run it once before run. For bonn, epilepsiae, sz1, tuab and tusz it builds an optional faster-to-read copy. Download the raw data first. The files go to <cache root>/prepared.

Usage: neuroatlas data prepare [dataset] [options]

neuroatlas data prepare --list
neuroatlas data prepare bonn --dry-run
neuroatlas data prepare tusz --shard 0/8

Option	Description	Default
`dataset`	Dataset name. --list shows the datasets with a build step.
`--set KEY=VALUE`	Change a dataset setting, as in raw_root=/data.
`--dest PATH`	Where to write the copy.	`<cache root>/prepared`
`--shard K/N`	Build only shard K of N, to split the work across cluster jobs.
`--dry-run`	Print what would be built, without building it.
`--list`	List the datasets that have a build step.

models¶

Check which model weights are on this machine, and download the missing ones.

Usage: neuroatlas models <action>

Command	Description
`models status`	Show which checkpoints have their weights here.
`models download`	Download the weights a selection is missing.

models status¶

Show which checkpoints have their weights on this machine. It does not use the network.

Usage: neuroatlas models status [selector] [options]

neuroatlas models status
neuroatlas models status all_fm -v

Option	Description	Default
`selector`	Checkpoints to check, given as an alias, group, family or ids.	every checkpoint
`-m`, `--models SELECTOR`	The same selection, given with -m as for `run`.
`-v`, `--verbose`	Show where each file is expected, and explain every state.
`--format {table,csv,md,json}`	Output format. In csv, md and json the notes under a row go to a `note` column.	`table`

models download¶

Download the weights that a selection of checkpoints is missing. It exits with status 1 if one could not be downloaded.

Usage: neuroatlas models download [selector] [options]

neuroatlas models download biot_pretrained
neuroatlas models download all_fm

Option	Description	Default
`selector`	Checkpoints to download, given as an alias, group, family or ids such as all_fm.
`-m`, `--models SELECTOR`	The same selection, given with -m as for `run`.

Note

neuroatlas fetch and neuroatlas prepare are other names for neuroatlas data download and neuroatlas data prepare.