def test_score(self):
dnet = DNetPRO(estimator=GaussianNB(), max_chunk=4)
Nprobe, Nsample = (5, 4)
X = np.arange(Nprobe * Nsample).reshape((Nsample, Nprobe))
y = np.array(['A', 'A', 'B', 'B'])
dnet.fit(X, y)
score = dnet.score(X, y)
assert score == 0
# print(' took {:.3f} seconds'.format(kbest_toc - dnet_tic), flush=True)
# print(dnet_perf[['feature_1', 'feature_2', 'ct', 'feature_1_ct', 'feature_2_ct', 'weights']].head(n=20))
single_perf.sort_values(by='feature_2_ct',
inplace=True,
ascending=False)
# print(single_perf.head(n=20))
Dnet_data = dnet.fit_transform(X_train, y_train)
new_sample, new_probe = Dnet_data.shape
dnet_signature = dnet.selected_signature
# DNetPRO parameters to save
dnet_score = dnet.score(X_test, y_test)
dnet_informative = len(
[x for x in dnet_signature if x < Ninformative])
dnet_size = len(dnet_signature)
Kbest_data = single_perf.iloc[:dnet_size].feature_2.values.tolist()
kbest_score = classifier.fit(X_train[:, Kbest_data],
y_train).score(
X_test[:, Kbest_data], y_test)
kbest_informative = len(
[x for x in Kbest_data if x < Ninformative])
common_features = len(set(Kbest_data) & set(dnet_signature))
# join string
# Create the DNetPRO feature selection object with GaussianNB classifier
dnet = DNetPRO(estimator=classifier,
scoring='accuracy',
n_jobs=4,
verbose=False)
# extract the filtered dataset as the signature with highest score in the training set
Dnet_data = dnet.fit_transform(X_train, y_train)
new_sample, new_probe = Dnet_data.shape
# Best DNetPRO signature
dnet_signature = dnet.selected_signature
# print some informations
print('Signature DNetPRO: {}'.format(sorted(dnet_signature)))
print('DNetPRO score: {:.3f}'.format(dnet.score(X_test, y_test)))
print('Informative found: {:d} / {:d}'.format(
len([x for x in dnet_signature if x < Ninformative]), Ninformative))
# Compare the obtained results against the Kbest features with K=number of feature in the DNetPRO signature
filter_kbest = SelectKBest(k=new_probe)
# extract the filtered datasets
Kbest_data = filter_kbest.fit_transform(X_train, y_train)
# set to zero the other features
Kbest_filtered = filter_kbest.inverse_transform(Kbest_data)
# now it is easy to extract the selected features as non-zero columns
Kbest_signature = set(np.nonzero(Kbest_filtered)[1])
# Just to be sure that everything goes right...
if not (len(Kbest_signature) == len(dnet_signature)):
raise ValueError(