def best_svm(tracks, feature_names, n_iter=200, save=False):
clf = machine_learning.Classifier('rbfsvm')
X, Y = machine_learning.shape_features(tracks, feature_names)
param_dist = {
'C': scipy.stats.expon(scale=1000),
'class_weight': ['auto'],
#'loss': ['squared_hinge'],
#'penalty': ['l2'],
#'dual': [False],
'tol': scipy.stats.expon(scale=0.1),
}
logging.info('Optimizing parameters: {}'.format(param_dist))
random_search = sklearn.grid_search.RandomizedSearchCV(
clf.clf,
param_distributions=param_dist,
n_iter=n_iter,
verbose=10,
)
random_search.fit(X, Y)
for score in random_search.grid_scores_:
print(score)
print('Best Score: {}'.format(random_search.best_score_))
print('Best Params: {}'.format(random_search.best_params_))
if save:
logging.info('Saving classifier to disk...')
joblib.dump(random_search.best_estimator_, save, compress=True)
return random_search.best_estimator_
def best_svm(tracks, feature_names, n_iter=200, save=False):
clf = machine_learning.Classifier('rbfsvm')
X, Y = machine_learning.shape_features(tracks, feature_names)
param_dist = {
'C': scipy.stats.expon(scale=1000),
'class_weight': ['auto'],
#'loss': ['squared_hinge'],
#'penalty': ['l2'],
#'dual': [False],
'tol': scipy.stats.expon(scale=0.1),
}
logging.info('Optimizing parameters: {}'.format(param_dist))
random_search = sklearn.grid_search.RandomizedSearchCV(
clf.clf,
param_distributions=param_dist,
n_iter=n_iter,
verbose=10,
)
random_search.fit(X, Y)
for score in random_search.grid_scores_:
print(score)
print('Best Score: {}'.format(random_search.best_score_))
print('Best Params: {}'.format(random_search.best_params_))
if save:
logging.info('Saving classifier to disk...')
joblib.dump(random_search.best_estimator_, save, compress=True)
return random_search.best_estimator_
def cross_val_score(tracks, feature_names, folds=5):
X, Y = machine_learning.shape_features(tracks, feature_names)
clf = sklearn.svm.LinearSVC(class_weight='auto')
scores = cross_validation.cross_val_score(clf,
X,
Y,
cv=folds,
scoring='f1_weighted')
return scores
def cross_val_score(tracks, feature_names, folds=5):
X, Y = machine_learning.shape_features(tracks, feature_names)
clf = sklearn.svm.LinearSVC(class_weight='auto')
scores = cross_validation.cross_val_score(
clf,
X,
Y,
cv=folds,
scoring='f1_weighted'
)
return scores
def kfold(tracks, feature_names, folds=5, shuffle=True, **kwargs):
labels = [track['label'] for track in tracks]
kf = cross_validation.StratifiedKFold(labels, n_folds=folds, shuffle=shuffle)
for train, test in kf:
train_tracks = [tracks[i] for i in train]
test_tracks = [tracks[i] for i in test]
clf = machine_learning.Classifier(**kwargs)
clf = machine_learning.train_tracks(clf, train_tracks, feature_names)
predicted_all = []
Y_test_all = []
for track in test_tracks:
X_test, Y_test = machine_learning.shape_features([track], feature_names)
predicted = machine_learning.predict(X_test, clf)
track['sample_predictions'] = predicted
track['prediction'], track['predictions'] = util.most_common(predicted)
predicted_all.extend(predicted)
Y_test_all.extend(Y_test)
yield test_tracks
def kfold(tracks, feature_names, folds=5, shuffle=True, **kwargs):
labels = [track['label'] for track in tracks]
kf = cross_validation.StratifiedKFold(labels,
n_folds=folds,
shuffle=shuffle)
for train, test in kf:
train_tracks = [tracks[i] for i in train]
test_tracks = [tracks[i] for i in test]
clf = machine_learning.Classifier(**kwargs)
clf = machine_learning.train_tracks(clf, train_tracks, feature_names)
predicted_all = []
Y_test_all = []
for track in test_tracks:
X_test, Y_test = machine_learning.shape_features([track],
feature_names)
predicted = machine_learning.predict(X_test, clf)
track['sample_predictions'] = predicted
track['prediction'], track['predictions'] = util.most_common(
predicted)
predicted_all.extend(predicted)
Y_test_all.extend(Y_test)
yield test_tracks
