def GCN_check(name, adj, weights, layer_config):
num_layer = len(layer_config)
model = Network()
for i in range(num_layer - 2):
model.add(Aggregate('A{}'.format(i), adj))
model.add(
Linear('W{}'.format(i), layer_config[i], layer_config[i + 1],
'xavier').set_W(weights[i]))
model.add(Tanh('Tanh{}'.format(i)))
model.add(Aggregate('A{}'.format(num_layer - 2), adj))
model.add(
Linear('W{}'.format(num_layer - 2), layer_config[-2], layer_config[-1],
'xavier').set_W(weights[-1]))
loss = SoftmaxCrossEntropyLoss(name='loss')
# loss = EuclideanLoss(name='loss')
print("Model " + name)
for layer in model.layer_list:
print(":\t" + repr(layer))
print(':\t' + repr(loss))
print('Forward Computation: ', model.str_forward('X'))
print('Backward Computation:', model.str_backward('Z-Y'))
print()
model.str_update()
print()
return model, loss
def __call__(self, q_input, a_input, *args, **kwargs):
# convolve input feature maps with filters
q_conv_out = conv2d(input=q_input,
filters=self.W,
filter_shape=self.filter_shape)
a_conv_out = conv2d(input=a_input,
filters=self.W,
filter_shape=self.filter_shape)
# add the bias term. Since the bias is a vector (1D array), we first
# reshape it to a tensor of shape (1, n_filters, 1, 1). Each bias will
# thus be broadcasted across mini-batches and feature map
# width & height
if self.non_linear == "tanh":
q_conv_out_tanh = Tanh(q_conv_out +
self.b.dimshuffle('x', 0, 'x', 'x'))
a_conv_out_tanh = Tanh(a_conv_out +
self.b.dimshuffle('x', 0, 'x', 'x'))
q_output = pool.pool_2d(input=q_conv_out_tanh,
ws=self.pool_size,
ignore_border=True) # max
a_output = pool.pool_2d(input=a_conv_out_tanh,
ws=self.pool_size,
ignore_border=True)
elif self.non_linear == "relu":
q_conv_out_relu = ReLU(q_conv_out +
self.b.dimshuffle('x', 0, 'x', 'x'))
a_conv_out_relu = ReLU(a_conv_out +
self.b.dimshuffle('x', 0, 'x', 'x'))
q_output = pool.pool_2d(input=q_conv_out_relu,
ws=self.pool_size,
ignore_border=True)
a_output = pool.pool_2d(input=a_conv_out_relu,
ws=self.pool_size,
ignore_border=True)
else:
q_output = pool.pool_2d(input=q_conv_out,
ws=self.pool_size,
ignore_border=True)
a_output = pool.pool_2d(input=a_conv_out,
ws=self.pool_size,
ignore_border=True)
return q_output, a_output
def forward(self, x, hidden_state_prev, params):
assert len(x.shape) == 2
affine_hidden, affine_input, affine_output, tanh = Affine(), Affine(
), Affine(), Tanh()
hidden_state_raw = affine_hidden(hidden_state_prev, params['h2h'],
params['h2h_b'])
hidden_state_raw += affine_input(x, params['i2h'], params['i2h_b'])
hidden_state = tanh(hidden_state_raw)
logits = affine_output(hidden_state, params['h2o'], params['h2o_b'])
self.cache = (affine_hidden, affine_input, affine_output, tanh, params)
return hidden_state, logits
def test_TwoDifferentModelsShouldHaveDifferentGradients(self):
x = np.random.rand(5)
real_model = Seq([
Linear(5, 3, initialize='ones'),
Tanh(),
Linear(3, 5, initialize='ones'),
Tanh()
])
y = real_model.forward(x)
real_grad = real_model.backward(np.ones(5))
num_model = Seq([
Linear(5, 3, initialize='ones'),
Relu(),
Linear(3, 5, initialize='ones'),
Relu()
])
num_grad = numerical_gradient.calc(num_model.forward, x)
num_grad = np.sum(num_grad, axis=1)
self.assertFalse(numerical_gradient.are_similar(real_grad, num_grad))
def test_TwoLinearLayersTanh(self):
x = np.random.rand(5)
real_model = Seq([
Linear(5, 3, initialize='ones'),
Tanh(),
Linear(3, 5, initialize='ones'),
Tanh()
])
y = real_model.forward(x)
real_grad = real_model.backward(np.ones(5))
num_model = Seq([
Linear(5, 3, initialize='ones'),
Tanh(),
Linear(3, 5, initialize='ones'),
Tanh()
])
num_grad = numerical_gradient.calc(num_model.forward, x)
num_grad = np.sum(num_grad, axis=1)
self.assertTrue(numerical_gradient.are_similar(real_grad, num_grad))
def test_CheapTanh(self):
x = np.random.rand(10)
cheap_tanh = CheapTanh()
def f1():
cheap_tanh.forward(x)
tanh_layer = Tanh()
def f2():
tanh_layer.forward(x)
t1 = timeit.timeit(f1, number=10000)
t2 = timeit.timeit(f2, number=10000)
self.assertGreater(t2, t1)
from train import train, evaluate
from keras.utils import to_categorical
import numpy as np
iris = load_iris()
X = iris.data
y = iris.target
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
y_train_one_hot = to_categorical(y_train).astype(np.int)
y_test_one_hot = to_categorical(y_test).astype(np.int)
net = NetWork([
Linear(input_dim=4, output_dim=5),
Tanh(),
Linear(input_dim=5, output_dim=3),
Dropout(p=0.3),
Softmax(input_dim=3)
])
svc = SVC()
lr = LogisticRegression()
# 训练SVM分类器
svc.fit(X_train, y_train)
# 训练逻辑斯蒂回归分类器
lr.fit(X_train, y_train)
# 训练神经网络
FullyConnected(32,
config.NUM_CLASSES,
xavier_uniform_init,
use_weight_norm=True,
use_bias=False)
]
tanh_model = [
FullyConnected(config.INPUT_DIM,
256,
xavier_uniform_init,
use_weight_norm=True,
use_bias=False),
Dropout(0.4),
BatchNorm(input_dim=256),
Tanh(),
FullyConnected(256,
64,
xavier_uniform_init,
use_weight_norm=True,
use_bias=False),
Dropout(0.4),
BatchNorm(input_dim=64),
Tanh(),
FullyConnected(64,
32,
xavier_uniform_init,
use_weight_norm=True,
use_bias=False),
Dropout(0.1),
BatchNorm(input_dim=32),
