from sklearn.svm import SVC
from pyriemann.estimation import Covariances
from pyriemann.spatialfilters import CSP
from sklearn.model_selection import GridSearchCV
from sklearn.feature_selection import SelectKBest, mutual_info_classif
from moabb.pipelines.utils import FilterBank
from sklearn.pipeline import make_pipeline
import numpy as np
parameters = {'C': np.logspace(-2, 2, 10)}
clf = GridSearchCV(SVC(kernel='linear'), parameters)
fb = FilterBank(make_pipeline(Covariances(estimator='oas'), CSP(nfilter=4)))
pipe = make_pipeline(fb, SelectKBest(score_func=mutual_info_classif, k=10),
clf)
# this is what will be loaded
PIPELINE = {
'name': 'FBCSP + optSVM',
'paradigms': ['FilterBankMotorImagery'],
'pipeline': pipe
}
#
# The CSP implementation from MNE is used. We selected 8 CSP components, as
# usually done in the litterature.
#
# The second pipeline is the filter bank CSP. We use the FilterBank object
# with a CSP estimator. We set up the CSP to 4 components, to compensate for
# the higher dimensionality.
#
# The two pipelines will be applied on two different paradigm, so they have
# their own dict.
pipelines = {}
pipelines['CSP + LDA'] = make_pipeline(CSP(n_components=8), LDA())
pipelines_fb = {}
pipelines_fb['FBCSP + LDA'] = make_pipeline(FilterBank(CSP(n_components=4)),
LDA())
##############################################################################
# Evaluation
# ----------
#
# Since two different preprocessing will be applied, we have two different
# paradigm objects. We have to make sure their filter matchs so the comparison
# will be fair.
#
# The first one is a standard `LeftRightImagery` with a 8 to 35 Hz broadband
# filter.
#
# The second is a `FilterBankLeftRightImagery` with a bank of 6 filter, ranging
# from 8 to 35 Hz.