def _mf(x, Id, grp, self, display, probOccur):
"""cross-validation multifeatures
"""
# Unpack arguments :
y = self._y
cvOut = defCv(y, **self._cvOut).cvr
clfOut = defClf(y, **self._clfOut)
# Repetitions loop :
idxCvOut, da = [], []
for k, i in enumerate(cvOut):
predCv, yTestCv = [], []
for train_index, test_index in i.split(x, y):
# Get training and testing sets:
xTrain, xTest, yTrain, yTest = x[train_index, :], x[
test_index, :], y[train_index], y[test_index]
# Define the classification object :
clfObj = classify(yTrain,
clf=defClf(yTrain, **self._clfIn),
cvtype=defCv(yTrain, **self._cvIn))
# Get the MF model:
MFmeth, MFstr = Id2methods(Id,
yTrain,
clfObj,
p=self._p,
display=display,
direction=self._direction,
nbest=self._nbest,
stat=self._stat,
n_perm=self._n_perm)
# Apply the MF model:
xCv, idxCvIn, grpIn, MFstrCascade = applyMethods(
MFmeth, MFstr, xTrain, grp)
# Select the xTest features:
xTest = xTest[:, idxCvIn]
# Classify:
if xCv.size:
predCv.extend(clfOut.fit(xCv, yTrain).predict(xTest))
else:
predCv.extend(n.array([None] * len(test_index)))
# Keep info:
idxCvOut.extend(idxCvIn), yTestCv.extend(yTest)
# Get the decoding accuracy :
da.append(100 * sum([
1 if predCv[k] == yTestCv[k] else 0 for k in range(0, len(predCv))
]) / len(predCv))
prob = occurProba(idxCvOut, list(range(x.shape[1])), kind=probOccur)
return da, prob, MFstrCascade
def __init__(self, y, Id='0', p=0.05, n_perm=200, stat='bino',
threshold=None, nbest=10, direction='forward', occurence='i%',
clfIn={'clf': 'lda'}, clfOut={'clf': 'lda'},
cvIn={'cvtype': 'skfold', 'n_folds': 10, 'rep': 1},
cvOut={'cvtype': 'skfold', 'n_folds': 10, 'rep': 10}):
self._Id = Id
self._stat = stat
self._y = np.ravel(y)
if threshold is not None:
p = bino_da2p(y, threshold)
self._p = p
self._nbest = nbest
self._n_perm = n_perm
self._direction = direction
self._threshold = threshold
self._occurence = occurence
self._clfIn = clfIn
self._clfOut = clfOut
self._cvIn = cvIn
self._cvOut = cvOut
self.setup = {'Id': Id, 'p': p, 'n_perm': n_perm, 'stat': stat,
'nbest': nbest, 'threshold': threshold,
'direction': direction, 'occurence': occurence,
'clfIn': defClf(y, **clfIn).lgStr,
'clfOut': defClf(y, **clfOut).lgStr,
'cvIn': defCv(y, **cvIn).lgStr,
'cvOut': defCv(y, **cvOut).lgStr}
def _mf(x, Id, grp, self, display, probOccur):
"""cross-validation multifeatures
"""
# Unpack arguments :
y = self._y
cvOut = defCv(y, **self._cvOut).cvr
clfOut = defClf(y, **self._clfOut)
# Repetitions loop :
idxCvOut, da = [], []
for k, i in enumerate(cvOut):
predCv, yTestCv = [], []
for train_index, test_index in i:
# Get training and testing sets:
xTrain, xTest, yTrain, yTest = x[train_index, :], x[
test_index, :], y[train_index], y[test_index]
# Define the classification object :
clfObj = classify(yTrain, clf=defClf(yTrain, **self._clfIn),
cvtype=defCv(yTrain, **self._cvIn))
# Get the MF model:
MFmeth, MFstr = Id2methods(Id, yTrain, clfObj, p=self._p,
display=display,
direction=self._direction,
nbest=self._nbest,
stat=self._stat, n_perm=self._n_perm)
# Apply the MF model:
xCv, idxCvIn, grpIn, MFstrCascade = applyMethods(
MFmeth, MFstr, xTrain, grp)
# Select the xTest features:
xTest = xTest[:, idxCvIn]
# Classify:
if xCv.size:
predCv.extend(clfOut.fit(xCv, yTrain).predict(xTest))
else:
predCv.extend(n.array([None] * len(test_index)))
# Keep info:
idxCvOut.extend(idxCvIn), yTestCv.extend(yTest)
# Get the decoding accuracy :
da.append(100 * sum([1 if predCv[k] == yTestCv[k]
else 0 for k in range(0, len(predCv))
]) / len(predCv))
prob = occurProba(idxCvOut, list(range(x.shape[1])), kind=probOccur)
return da, prob, MFstrCascade
def __init__(self,
y,
Id='0',
p=0.05,
n_perm=200,
stat='bino',
threshold=None,
nbest=10,
direction='forward',
occurence='i%',
clfIn={'clf': 'lda'},
clfOut={'clf': 'lda'},
cvIn={
'cvtype': 'skfold',
'n_folds': 10,
'rep': 1
},
cvOut={
'cvtype': 'skfold',
'n_folds': 10,
'rep': 10
}):
self._Id = Id
self._stat = stat
self._y = np.ravel(y)
if threshold is not None:
p = bino_da2p(y, threshold)
self._p = p
self._nbest = nbest
self._n_perm = n_perm
self._direction = direction
self._threshold = threshold
self._occurence = occurence
self._clfIn = clfIn
self._clfOut = clfOut
self._cvIn = cvIn
self._cvOut = cvOut
self.setup = {
'Id': Id,
'p': p,
'n_perm': n_perm,
'stat': stat,
'nbest': nbest,
'threshold': threshold,
'direction': direction,
'occurence': occurence,
'clfIn': defClf(y, **clfIn).lgStr,
'clfOut': defClf(y, **clfOut).lgStr,
'cvIn': defCv(y, **cvIn).lgStr,
'cvOut': defCv(y, **cvOut).lgStr
}