def test_n_for_b(self):
p_x_train, p_x_test, x_train, x_test, y_train, y_test, im, om = \
util.set_sample(1, 1, 100, 40)
bc = b_classfy.b_classfy([im, 200, 150, om], isClassification=True)
bc.pre_training(p_x_train, p_x_test)
self.assertGreater(bc.learn(x_train, y_train, x_test, y_test), 0.8)
def test_n_for_b(self):
p_x_train, p_x_test, x_train, x_test, y_train, y_test, im, om = \
util.set_sample(1, 1, 100, 40)
bc = b_classfy.b_classfy([im, 200, 150, om], isClassification=True)
bc.pre_training(p_x_train, p_x_test)
self.assertGreater(bc.learn(x_train, y_train, x_test, y_test), 0.8)
def test_auto_encoder(self):
p_x_train, p_x_test, x_train, x_test, y_train, y_test, im, om = \
util.set_sample(60, 1, 40, 20, split_mode='pp', offset_cancel=True, same_sample=10)
bc = auto_encoder.Autoencoder([im, 200, 150, im], epoch=40, is_classification=False, nobias=False)
bc.pre_training(p_x_train, p_x_test)
bc.learn(x_train, x_train, x_test, x_test)
#bc.disp_w()
self.assertLess(bc.final_test(x_test[0:9], x_test[0:9], False), 0.1)
def test_auto_encoder(self):
p_x_train, p_x_test, x_train, x_test, y_train, y_test, im, om = \
util.set_sample(60, 1, 40, 20, split_mode='pp', offset_cancel=True, same_sample=10)
bc = auto_encoder.Autoencoder([im, 200, 150, im],
epoch=40,
is_classification=False,
nobias=False)
bc.pre_training(p_x_train, p_x_test)
bc.learn(x_train, x_train, x_test, x_test)
#bc.disp_w()
self.assertLess(bc.final_test(x_test[0:9], x_test[0:9], False), 0.1)
epoch=10,
**keywords):
AbstractChain.__init__(self,
n_units=n_units,
is_classification=is_classification,
epoch=epoch,
batch_size=batch_size,
visualize=True,
**keywords)
#configuration
self.plot_enable = plot_enable
self.save_as_png = save_as_png
if save_as_png and not os.path.exists(util.IMAGE_DIR):
os.mkdir(util.IMAGE_DIR)
# Save final self.model
if pickle_enable:
pickle.dump(self.model, open('self.model', 'w'), -1)
if __name__ == '__main__':
p_x_train, p_x_test, x_train, x_test, y_train, y_test, im, om = \
util.set_sample(1, 1, 100, 40)
bc = b_classfy([im, 150, 150, om], isClassification=True)
bc.pre_training(p_x_train, p_x_test)
bc.learn(x_train, y_train, x_test, y_test)
#bc.disp_w()
bc.final_test(x_test[0:9], y_test[0:9])
if not isinstance(x_data, Variable):
x_data = Variable(x_data)
for i, model in enumerate(self):
if i == len(self) - 1 and self.pre_trained:
x_data = model(x_data)
else:
x_data = F.dropout(F.relu(model(x_data)), train=train)
return x_data
if __name__ == '__main__':
p_x_train0, p_x_test0, x_train0, x_test0, y_train0, y_test0, im, om = \
util.set_sample(60, 1, 40, 20,
split_mode='pp',
#offset_cancel=True,
first_cancel=True,
#normal_constant=1709.0,
same_sample=70,
spec_target=0,
)
p_x_train1, p_x_test1, x_train1, x_test1, y_train1, y_test1, im, om = \
util.set_sample(60, 1, 40, 20,
split_mode='pp',
first_cancel=True,
div_reference=True,
#normal_constant=1709.0,
)
bc = Autoencoder([im, 150, 120, im],
epoch=100,
is_classification=False,
nobias=True)