Skip to content

Home

Jump to bottom
Pierre Bellec edited this page Feb 14, 2021 · 8 revisions

Data structures

Parcellations and R2 maps

Main folder

Folder {type_experiment}, where type_experiment is the name of the experiment in:

['friends-s01', 'friends-s01_clean', 'friends-s01_clean_multi_fwhm', 'friends', 'hcptrt', 'movie10']

Subfolders

A series of subfolders named:

dataset-{data}_tasks-{task}_cluster-{n_cluster}_states-{n_state}_batches-{n_batch}_reps-{n_rep}_fwhm-{fwhm}

where:

  • data is the name of the cneuromod dataset, in:
['friends', 'movie10', 'hcptrt']
  • task is the name of the task, and depends on type_experiment
    • for friends-s01*, in:
['s01even', 's01odd']
  • for friends, in:
['s01', 's02']
  • for hcptrt, in:
['restingstate', 'gambling', 'motor', 'social', 'wm', 'emotion', 'language', 'relational']
  • for movie10, in:
['bourne', 'wolf', 'life1', 'life2', 'figures1', 'figures2']
  • n_cluster, n_states are the number of clusters and states, respectively. The parameters depend on the type of experiment.
    • for friends-s01, in:
[20-60, 20-120, 50-150, 50-300, 300-900]
  • for other experiments, in:
[64-256, 64-512, 256-1024]
  • n_batch is the number of batches for dypac. Because dypac scales well enough for cneuromod2020, all experiments are run with n_batch=1.
  • n_rep number of cluster replication per run. Always set to n_rep=100.
  • fwhm is the smoothing parameter for fMRI data. Depends on the experiment:
    • for friends-s01 and friends-s01_clean, in: [5, 8].
    • for all other experiments [5].

Files

Intra-subject

{subject}_{subfolder}_{type_experiment}_r2_scores.hdf5

Where subfolder is the name of the subfolder described above, subject is a subject id (sub-01 to sub-06), and type_experiment is described in the first paragraph. This hdf5 file has an attribute 'validation' and then a list of file names corresponding to different runs of that subject. The variable is a 3D numpy array with a R2 brain map in standard space.

Inter-subject

{subject}_{subfolder}_{type_experiment}_inter_r2_scores.hdf5

Where subfolder is the name of the subfolder described above, subject is a subject id (sub-01 to sub-06), and type_experiment is described in the first paragraph. This hdf5 file has an attribute 'inter' then a list of subject ids attribute, and then a list of file names corresponding to different runs of that subject. The variable is a 3D numpy array with a R2 brain map in standard space.

parcellation model

{subject}_{subfolder}_{type_experiment}.pickle

a pickled model generated using dypac_masker.

Other atlases

Additional r2 maps generated for pre-generated atlases are saved in an additional subfolder other_atlases. It contains a series of files:

{atlas}_fwhm-{fwhm}_r2_score.hdf5

where atlas is the name of one of the following atlases:

['difumo256', 'difumo512', 'difumo1024', 'mist197', 'mist444', 'schaefer', 'smith']

This hdf5 file has a list of subject ids attribute, then a list of file names corresponding to different runs of that subject. The variable is a 3D numpy array with a R2 brain map in standard space.

Clone this wiki locally