Note
Go to the end to download the full example code.
Tutorial 5: Creating a dataset class#
# 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)
CustomDataset3-WithinSession: 0%| | 0/4 [00:00<?, ?it/s]No hdf5_path provided, models will not be saved.
CustomDataset3-WithinSession: 25%|██▌ | 1/4 [00:30<01:32, 30.82s/it]No hdf5_path provided, models will not be saved.
CustomDataset3-WithinSession: 50%|█████ | 2/4 [00:37<00:33, 16.55s/it]
0%| | 0.00/46.4M [00:00<?, ?B/s]
0%| | 12.3k/46.4M [00:00<08:17, 93.3kB/s]
0%| | 119k/46.4M [00:00<01:54, 406kB/s]
1%|▍ | 526k/46.4M [00:00<00:36, 1.25MB/s]
3%|█ | 1.38M/46.4M [00:00<00:14, 3.19MB/s]
6%|██▍ | 3.00M/46.4M [00:00<00:06, 6.81MB/s]
12%|████▌ | 5.68M/46.4M [00:00<00:03, 12.5MB/s]
19%|███████ | 8.84M/46.4M [00:00<00:02, 18.0MB/s]
27%|██████████▏ | 12.7M/46.4M [00:01<00:01, 22.4MB/s]
36%|█████████████▎ | 16.7M/46.4M [00:01<00:01, 27.3MB/s]
43%|███████████████▉ | 20.0M/46.4M [00:01<00:00, 29.0MB/s]
51%|██████████████████▉ | 23.7M/46.4M [00:01<00:00, 31.2MB/s]
58%|█████████████████████▌ | 27.1M/46.4M [00:01<00:00, 32.1MB/s]
66%|████████████████████████▍ | 30.7M/46.4M [00:01<00:00, 33.1MB/s]
74%|███████████████████████████▌ | 34.5M/46.4M [00:01<00:00, 34.6MB/s]
82%|██████████████████████████████▎ | 38.0M/46.4M [00:01<00:00, 34.7MB/s]
89%|█████████████████████████████████ | 41.5M/46.4M [00:01<00:00, 33.8MB/s]
97%|███████████████████████████████████▊ | 45.0M/46.4M [00:01<00:00, 34.1MB/s]
0%| | 0.00/46.4M [00:00<?, ?B/s]
100%|██████████████████████████████████████| 46.4M/46.4M [00:00<00:00, 183GB/s]
No hdf5_path provided, models will not be saved.
CustomDataset3-WithinSession: 75%|███████▌ | 3/4 [00:51<00:15, 15.66s/it]
0%| | 0.00/74.3M [00:00<?, ?B/s]
0%| | 15.4k/74.3M [00:00<10:36, 117kB/s]
0%| | 109k/74.3M [00:00<02:42, 456kB/s]
0%|▏ | 250k/74.3M [00:00<01:47, 690kB/s]
1%|▍ | 898k/74.3M [00:00<00:29, 2.52MB/s]
3%|▉ | 1.93M/74.3M [00:00<00:14, 4.94MB/s]
5%|█▋ | 3.46M/74.3M [00:00<00:08, 8.11MB/s]
8%|███ | 6.05M/74.3M [00:00<00:05, 13.5MB/s]
13%|████▋ | 9.53M/74.3M [00:00<00:03, 18.9MB/s]
18%|██████▌ | 13.2M/74.3M [00:01<00:02, 24.1MB/s]
23%|████████▍ | 16.8M/74.3M [00:01<00:02, 27.7MB/s]
27%|██████████ | 20.3M/74.3M [00:01<00:02, 26.0MB/s]
31%|███████████▋ | 23.4M/74.3M [00:01<00:01, 27.4MB/s]
36%|█████████████▍ | 26.9M/74.3M [00:01<00:01, 29.6MB/s]
41%|███████████████ | 30.2M/74.3M [00:01<00:01, 30.4MB/s]
45%|████████████████▋ | 33.4M/74.3M [00:01<00:01, 31.0MB/s]
50%|██████████████████▍ | 36.9M/74.3M [00:01<00:01, 32.2MB/s]
54%|████████████████████ | 40.3M/74.3M [00:01<00:01, 32.6MB/s]
59%|█████████████████████▋ | 43.6M/74.3M [00:01<00:00, 32.3MB/s]
63%|███████████████████████▍ | 47.0M/74.3M [00:02<00:00, 33.0MB/s]
69%|█████████████████████████▎ | 50.9M/74.3M [00:02<00:00, 29.2MB/s]
74%|███████████████████████████▏ | 54.7M/74.3M [00:02<00:00, 30.4MB/s]
79%|█████████████████████████████▎ | 58.7M/74.3M [00:02<00:00, 33.1MB/s]
84%|██████████████████████████████▉ | 62.1M/74.3M [00:02<00:00, 33.0MB/s]
88%|████████████████████████████████▋ | 65.6M/74.3M [00:02<00:00, 33.4MB/s]
93%|██████████████████████████████████▍ | 69.2M/74.3M [00:02<00:00, 29.5MB/s]
98%|████████████████████████████████████▍| 73.0M/74.3M [00:02<00:00, 31.7MB/s]
0%| | 0.00/74.3M [00:00<?, ?B/s]
100%|██████████████████████████████████████| 74.3M/74.3M [00:00<00:00, 339GB/s]
No hdf5_path provided, models will not be saved.
CustomDataset3-WithinSession: 100%|██████████| 4/4 [01:20<00:00, 20.75s/it]
CustomDataset3-WithinSession: 100%|██████████| 4/4 [01:20<00:00, 20.13s/it]
score time samples ... n_sessions dataset pipeline
0 0.635000 0.327744 120.0 ... 1 CustomDataset3 MDM
1 0.582500 0.321615 120.0 ... 1 CustomDataset3 MDM
2 0.628115 2.076844 768.0 ... 1 CustomDataset3 MDM
3 0.575290 4.496056 1356.0 ... 1 CustomDataset3 MDM
[4 rows x 9 columns]
Total running time of the script: (1 minutes 21.116 seconds)
Estimated memory usage: 671 MB