def train(use_fisher):
x, y = loader.load()
if use_fisher:
x = preprocessing.to_fisher(x)
x_train, y_train, x_val, y_val = loader.split_data(x, y)
x_train = x_train.reshape(x_train.shape[0], -1)
x_val = x_val.reshape(x_val.shape[0], -1)
clf = svm.SVC(class_weight='balanced')
clf.fit(x_train, y_train)
evaluate.evaluate(clf, x_val, y_val)
def train(args):
x, y = loader.load()
if args.model == 'mlp_fv':
x = preprocessing.to_fisher(x)
nb_samples, nb_landmarks, l = x.shape
input_shape = (nb_landmarks, l)
else:
nb_samples, nb_frames, nb_landmarks, _ = x.shape
input_shape = (nb_frames, nb_landmarks, 3)
# x = loader.compact_frames(x, window_size=5, step_size=2)
x_train, y_train, x_val, y_val = loader.split_data(x, y)
model = get_model(args.model, input_shape=input_shape)
print("Input shape: {}".format(x.shape))
print(model.summary())
model.compile(optimizer='adam',
loss=losses.binary_crossentropy,
metrics=['accuracy'])
checkpointer = callbacks.ModelCheckpoint(filepath="data/weights.hdf5", verbose=1, save_best_only=True)
early_stopping = callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=2)
class_weights = get_class_weights(y_train)
model.fit(x_train, y_train,
epochs=50,
batch_size=8,
validation_data=(x_val, y_val),
callbacks=[
checkpointer,
early_stopping,
],
class_weight=class_weights,
)
# Load best model
model.load_weights('data/weights.hdf5')
# Print evaluation matrix
train_score = model.evaluate(x_train, y_train)
val_score = model.evaluate(x_val, y_val)
print(model.metrics_names, train_score, val_score)
evaluate.evaluate(model, x_train, y_train)
evaluate.evaluate(model, x_val, y_val)
def test_load(self):
x, y = loader.load(shuffle=False)
nb_x = x.shape[0]
nb_y = len(y)
# Test that number of samples match
self.assertEqual(nb_x, nb_y)
# Test that labels match csv
with open('data/bluff_data.csv') as fp:
reader = csv.DictReader(fp)
labels = []
for row in reader:
if row['error'] != '':
continue
labels.append(float(row['isBluffing']))
np.testing.assert_array_equal(y, np.array(labels))
def setUp(self):
self.x, self.y = loader.load(shuffle=False)
def test_fisher_vector(self):
x, y = loader.load(shuffle=False)
fv = preprocessing.to_fisher(x)