Tutorial 5: Combining Multiple Datasets into a Single Dataset

# Author: Gregoire Cattan
#
# https://github.com/plcrodrigues/Workshop-MOABB-BCI-Graz-2019

from pyriemann.classification import MDM
from pyriemann.estimation import ERPCovariances
from sklearn.pipeline import make_pipeline

from moabb.datasets import Cattan2019_VR
from moabb.datasets.braininvaders import BI2014a
from moabb.datasets.compound_dataset import CompoundDataset
from moabb.datasets.utils import blocks_reps
from moabb.evaluations import WithinSessionEvaluation
from moabb.paradigms.p300 import P300

Initialization

This tutorial illustrates how to use the CompoundDataset to: 1) Select a few subjects/sessions/runs in an existing dataset 2) Merge two CompoundDataset into a new one 3) … and finally use this new dataset on a pipeline (this steps is not specific to CompoundDataset)

Let’s define a paradigm and a pipeline for evaluation first.

paradigm = P300()
pipelines = {}
pipelines["MDM"] = make_pipeline(ERPCovariances(estimator="lwf"), MDM(metric="riemann"))

Creation a selection of subject

We are going to great two CompoundDataset, namely CustomDataset1 & 2. A CompoundDataset accepts a subjects_list of subjects. It is a list of tuple. A tuple contains 4 values:

• the original dataset

• the subject number to select

• the sessions. It can be:

  • a session name (‘0’)

  • a list of sessions ([‘0’, ‘1’])

  • None to select all the sessions attributed to a subject

• the runs. As for sessions, it can be a single run name, a list or None` (to select all runs).

class CustomDataset1(CompoundDataset):
    def __init__(self):
        biVR = Cattan2019_VR(virtual_reality=True, screen_display=True)
        runs = blocks_reps([0, 2], [0, 1, 2, 3, 4], biVR.n_repetitions)
        subjects_list = [
            (biVR, 1, "0VR", runs),
            (biVR, 2, "0VR", runs),
        ]
        CompoundDataset.__init__(
            self,
            subjects_list=subjects_list,
            code="CustomDataset1",
            interval=[0, 1.0],
        )


class CustomDataset2(CompoundDataset):
    def __init__(self):
        bi2014 = BI2014a()
        subjects_list = [
            (bi2014, 4, None, None),
            (bi2014, 7, None, None),
        ]
        CompoundDataset.__init__(
            self,
            subjects_list=subjects_list,
            code="CustomDataset2",
            interval=[0, 1.0],
        )

Merging the datasets

We are now going to merge the two CompoundDataset into a single one. The implementation is straight forward. Instead of providing a list of subjects, you should provide a list of CompoundDataset. subjects_list = [CustomDataset1(), CustomDataset2()]

class CustomDataset3(CompoundDataset):
    def __init__(self):
        subjects_list = [CustomDataset1(), CustomDataset2()]
        CompoundDataset.__init__(
            self,
            subjects_list=subjects_list,
            code="CustomDataset3",
            interval=[0, 1.0],
        )

Evaluate and display

Let’s use a WithinSessionEvaluation to evaluate our new dataset. If you already new how to do this, nothing changed: The CompoundDataset can be used as a normal dataset.

datasets = [CustomDataset3()]
evaluation = WithinSessionEvaluation(
    paradigm=paradigm, datasets=datasets, overwrite=False, suffix="newdataset"
)
scores = evaluation.process(pipelines)

print(scores)

/home/runner/work/moabb/moabb/moabb/analysis/results.py:192: H5pyDeprecationWarning: Creating a dataset without passing data or dtype is deprecated. Pass an explicit dtype. Using dtype='f4' will keep the current default behaviour.
  dset.create_dataset(
      score      time  samples  ...         dataset  pipeline  codecarbon_task_name
0  0.548850  4.172955   1356.0  ...  CustomDataset3       MDM
1  0.672500  0.296296    120.0  ...  CustomDataset3       MDM
2  0.602500  0.291793    120.0  ...  CustomDataset3       MDM
3  0.638183  1.892289    768.0  ...  CustomDataset3       MDM

[4 rows x 13 columns]