def select_perm(x, y, p=0.05, **kwargs):
"""Select <p significant features using the permutations"""
n_perm = round(1/p)
# Classify each features :
_, _, _, pvalue = classify(x, y, n_perm=n_perm, kind='sf', **kwargs)
return [k for k in range(0, len(pvalue)) if pvalue[k] < p]
def select_bino(x, y, p=0.05, **kwargs):
"""Select <p significant features using the binomial law"""
allfeat = [k for k in range(0, x.shape[0])]
# Classify each features :
da, _, _, _ = classify(x, y, kind='sf', **kwargs)
# Get significant features :
signifeat, _ = binofeat(y, da, p)
# Return list of significant fatures:
return list(n.array(allfeat)[signifeat])
def signifeat_permutations(x, y, threshold=None, classifier='lda', kern='rbf', n_folds=10, n_jobs=1, n_knn=10,
n_perm=100,
n_tree=100, cvkind='skfold'):
"Get significant features based on permutation test"
if threshold is None:
threshold = 1 / n_perm
_, nfeat = x.shape
signifeat = n.array(range(0, nfeat))
da, _, _, pvalue = ct.classify(x, y, classifier=classifier, kern=kern, n_folds=n_folds, rep=1,
kind='sf', n_jobs=n_jobs, n_knn=n_knn, n_perm=n_perm, n_tree=n_tree,
cvkind=cvkind)
upper_features = n.squeeze(pvalue < threshold)
return signifeat[upper_features], n.squeeze(pvalue), n.squeeze(da)
def classify_combination(x, y, combination, classifier='lda', kern='rbf', n_folds=10, rep=1, n_jobs=1,
n_knn=3, n_perm=0, n_tree=100,
cvkind='skfold'):
"""Classify each combination and return scores and the best one location."""
# - Get combination size :
n_comb = len(combination)
# - Classify each combination :
da = n.zeros((1, n_comb))
for k in range(0, n_comb):
da[0, k], all_scores, permutation_scores, pvalue = ct.classify(x[:, combination[k]], y, classifier=classifier,
kern=kern,
n_folds=n_folds, rep=rep, kind='mf',
n_jobs=n_jobs,
n_knn=n_knn, n_perm=n_perm, n_tree=n_tree,
cvkind=cvkind)
return da, all_scores, permutation_scores, pvalue
def bidimtransform(x, y, ybi, classifier=0, cvkind='skfold', n_folds=10, p=0):
nbtrans = len(ybi)
nbepoch, nbfeat = x.shape
flist = n.array(range(0, nbfeat))
da = n.zeros((nbtrans, nbfeat))
signifeat = []
for i in range(0, nbtrans):
# First, get index for selected item :
ylocidx = yloc(y, ybi[i])
# Select y and x index :
ynb = y[ylocidx]
xnb = x[ylocidx, :]
# Classify each x :
da[i, :], _, _, _ = ct.classify(xnb, ynb, classifier=classifier, kind='sf', rep=1, cvkind=cvkind,
n_folds=n_folds)
# Get significant features?
if p is not 0:
th = ct.binostat(ynb, p)
signifeat.extend([flist[100 * da[i, :] >= th]])
return da, signifeat
def select_nbest(x, y, nbest=10, **kwargs):
"""Select nbest features"""
# Classify each features :
da, _, _, _ = classify(x, y, kind='sf', **kwargs)
return list(n.ravel(da.T).argsort()[-nbest:][::-1])
