def test_set_params_from_file(self):
dir, _ = os.path.split(os.path.abspath(__file__))
file_path = os.path.join(dir, 'data', 'net_params.txt')
net = GraphLabNeuralNetBuilder()
net.set_params_from_file(file_path)
self.assertEqual(net['learning_rate'], 0.1)
self.assertEqual(net['batch_size'], 256)
self.assertEqual(net['learning_rate_schedule'], 'constant')
self.assertEqual(net['subtract_mean'], False)
print net.net
def test_convolution_layer(self):
nn = GraphLabNeuralNetBuilder()
nn.add_convolution_layer(3, 1, 10)
nn.add_convolution_layer(3, 1, 10, **{'padding': 1})
return
max_iterations = int(sys.argv[4])
df = pd.read_pickle(data_path)
cond_happy = df['django_expression'] == 3
cond_sad = df['django_expression'] == 4
cond_surprise = df['django_expression'] == 5
df = df[cond_happy | cond_sad | cond_surprise]
x = np.array(df['pixels'].tolist())
y = np.array(df['django_expression'].values)
xtrain, xtest, ytrain, ytest = train_test_split(x, y, train_size=.8)
net_builder = GraphLabNeuralNetBuilder()
net_builder.layers = net.layers
net_builder.verify()
model = GraphLabClassifierFromNetBuilder(net_builder,
chkpt_dir=check_point_path,
max_iterations=max_iterations,
train_frac=.8)
model.fit(xtrain, ytrain)
eval = model.evaluate(xtest, ytest, metric=['accuracy', 'confusion_matrix',
'recall@1', 'recall@2'])
ypred = np.array(model.predict(xtest))
ytest = np.array(ytest)
def test_net_params(self):
nn = GraphLabNeuralNetBuilder()
nn['momentum'] = 0.5
self.assertEqual(nn['momentum'], 0.5)
def test_drop_out_layer(self):
nn = GraphLabNeuralNetBuilder()
nn.add_dropout_layer()
nn.add_dropout_layer(.8)
def test_get_net_bad(self):
nn_bad = GraphLabNeuralNetBuilder()
self.assertRaises(ValueError, nn_bad.get_net) # Empty net
def test_activation_layers(self):
nn = GraphLabNeuralNetBuilder()
nn.add_relu_layer()
nn.add_sigmoid_layer()
nn.add_tanh_layer()
nn.add_soft_max_layer()
def test_full_connection_layer(self):
nn = GraphLabNeuralNetBuilder()
nn.add_flatten_layer()
nn.add_full_connection_layer(10)
nn.add_full_connection_layer(100)
def test_flatten_layer(self):
nn = GraphLabNeuralNetBuilder()
nn.add_flatten_layer()
def test_avg_pooling_layer(self):
nn = GraphLabNeuralNetBuilder()
nn.add_avg_pooling_layer(3)
nn.add_avg_pooling_layer(3, padding=0)
nn.add_avg_pooling_layer(3, 4, 5)
max_iterations = int(sys.argv[4])
df = pd.read_pickle(data_path)
cond_happy = df['django_expression'] == 3
cond_sad = df['django_expression'] == 4
cond_surprise = df['django_expression'] == 5
df = df[cond_happy | cond_sad | cond_surprise]
x = np.array(df['pixels'].tolist())
y = np.array(df['django_expression'].values)
xtrain, xtest, ytrain, ytest = train_test_split(x, y, train_size=.8)
net_builder = GraphLabNeuralNetBuilder()
net_builder.layers = net.layers
net_builder.verify()
model = GraphLabClassifierFromNetBuilder(net_builder,
chkpt_dir=check_point_path,
max_iterations=max_iterations,
train_frac=.8)
model.fit(xtrain, ytrain)
eval = model.evaluate(
xtest,
ytest,
metric=['accuracy', 'confusion_matrix', 'recall@1', 'recall@2'])
ypred = np.array(model.predict(xtest))