def test_one_class_empty(self):
w = np.array([ 1, 1, 0, 1])
y = np.array([ 1, 1, 0, 1])
resp1 = balance_masked_weights(y, w)
resp2 = balance_masked_weights(1-y, w)
self.assertTrue((w==resp1).all())
self.assertTrue((w==resp2).all())
input_shape = x.shape[1]
else:
input_shape = [arr.shape for arr in x]
data_shapes = config.get('data_shapes', 'none')
if data_shapes != 'none':
input_shape = data_shapes
for cv_train_idx, cv_val_idx in cv_splits:
train = slice_data(data_stuff, cv_train_idx, N_inputs)
val = slice_data(data_stuff, cv_val_idx, N_inputs)
if is_multitask:
logger.info('Train Y: %s' % str(np.sum(train[1], axis=1)))
for i, vy in enumerate(train[1]):
key = 'out%i' % i
train[2][key] = balance_masked_weights(vy, train[2][key])
val[2][key] = balance_masked_weights(val[1][i], val[2][key])
else:
train[2] = scale_weights(train[1], train[2], args.scale_positive)
val[2] = scale_weights(val[1], val[2], args.scale_positive)
logger.info('Train Y: %s' % str(np.sum(train[1], axis=0)))
model, metric = make_model(input_shape, np.shape(y),
config['model_params'])
logger.info('Model build')
result = model.fit(train[0],
train[1],
sample_weight=train[2],
validation_data=val,
batch_size=batch_size,
epochs=epochs,
shuffle=True,
def test_balance_masked_weights(self):
resp = balance_masked_weights(self.known_values['y'], self.known_values['w'])
msg = '\nGot: %s\nShould be: %s'%( str(resp), str(self.known_values['o']))
self.assertTrue((resp==self.known_values['o']).all(), msg)
def test_source_unchanged(self):
prev = hash(tuple(self.known_values['w']))
new = balance_masked_weights(self.known_values['y'], self.known_values['w'])
after = hash(tuple(self.known_values['w']))
self.assertEqual(prev, after)
def test_detect_is_none(self):
self.assertTrue(balance_masked_weights(self.known_values['y'], None) is None)