def inference(x, is_training=True):
with tf.variable_scope("layer1"):
filter_weights = weight_variable([1, CHANNELS, CHANNELS_L1], name="weights")
feature_map = tf.nn.conv1d(x, filter_weights, stride=1, padding='SAME')
feature_map = batch_norm(feature_map, decay=decay_bn, center=True, scale=scale_bn,
epsilon=epsilon_bn, activation_fn=None, is_training=is_training)
activation = tf.nn.elu(feature_map)
activation = tf.nn.dropout(activation, keep_prob=keep_prob)
activation = tf.reshape(activation, [-1, CHANNELS_L1, WINDOW_SIZE, 1])
with tf.variable_scope("layer2"):
filter_weights = weight_variable(KERNEL2, name="weights")
feature_map = tf.nn.conv2d(activation, filter_weights, strides=[1, 1, 1, 1], padding='SAME')
feature_map = batch_norm(feature_map, decay=decay_bn, center=True, scale=scale_bn,
epsilon=epsilon_bn, activation_fn=None, is_training=is_training)
activation = tf.nn.elu(feature_map)
activation = tf.nn.max_pool(activation, maxpool_ksize, [1, 1, 1, 1], padding='VALID',
data_format='NHWC', name='maxpool')
activation = tf.nn.dropout(activation, keep_prob=keep_prob)
with tf.variable_scope("layer3"):
filter_weights = weight_variable(KERNEL3, name="weights")
feature_map = tf.nn.conv2d(activation, filter_weights, strides=[1, 1, 1, 1], padding='SAME')
feature_map = batch_norm(feature_map, decay=decay_bn, center=True, scale=scale_bn,
epsilon=epsilon_bn, activation_fn=None, is_training=is_training)
activation = tf.nn.elu(feature_map)
activation = tf.nn.max_pool(activation, maxpool_ksize, [1, 1, 1, 1], padding='VALID',
data_format='NHWC', name='maxpool')
activation = tf.nn.dropout(activation, keep_prob=keep_prob)
with tf.variable_scope("layer4"):
filter_weights = weight_variable(KERNEL_DEEP, name="weights")
feature_map = tf.nn.conv2d(activation, filter_weights, strides=[1, 1, 1, 1], padding='SAME')
feature_map = batch_norm(feature_map, decay=decay_bn, center=True, scale=scale_bn,
epsilon=epsilon_bn, activation_fn=None, is_training=is_training)
activation = tf.nn.elu(feature_map)
activation = tf.nn.max_pool(activation, maxpool_ksize, [1, 1, 1, 1], padding='VALID',
data_format='NHWC', name='maxpool')
activation = tf.nn.dropout(activation, keep_prob=keep_prob)
with tf.variable_scope("output"):
dim = np.prod(activation.get_shape().as_list()[1:])
flattened = tf.reshape(activation, [-1, dim])
weights = weight_variable([dim, 1])
bias = bias_variable([1])
logits = tf.matmul(flattened, weights) + bias
return logits
def make_output_graph(self):
self.n_actions = self.env.action_space.n
with tf.name_scope("policy"):
self.W_policy = weight_variable([self.FC_2_SIZE, self.n_actions],
name="W")
self.b_policy = bias_variable([self.n_actions], name="b")
self.policy_head = tf.matmul(self.fc_2_activation_policy,
self.W_policy) + self.b_policy
with tf.name_scope("value"):
self.W_value = weight_variable([self.FC_2_SIZE, 1], name="W")
self.b_value = bias_variable([1], name="b")
self.value_head = tf.matmul(self.fc_2_activation_value,
self.W_value) + self.b_value
def make_fc_2_graph(self):
with tf.name_scope("fully_connected_2p"):
self.W_fc_2_policy = weight_variable(
[self.FC_1_SIZE, self.FC_2_SIZE], name="W")
self.b_fc_2_policy = bias_variable([self.FC_2_SIZE], name="b")
z = tf.matmul(self.fc_1_activation,
self.W_fc_2_policy) + self.b_fc_2_policy
self.fc_2_activation_policy = tf.sigmoid(z)
with tf.name_scope("fully_connected_2v"):
self.W_fc_2_value = weight_variable(
[self.FC_1_SIZE, self.FC_2_SIZE], name="W")
self.b_fc_2_value = bias_variable([self.FC_2_SIZE], name="b")
z = tf.matmul(self.fc_1_activation,
self.W_fc_2_value) + self.b_fc_2_value
self.fc_2_activation_value = tf.sigmoid(z)
def make_output_graph(self):
self.n_outputs = self.env.action_space.n
with tf.name_scope("output"):
self.W_output = weight_variable([self.HIDDEN_SIZE, self.n_outputs],
name="W")
self.b_output = bias_variable([self.n_outputs], name="b")
z = tf.matmul(self.hidden_activation,
self.W_output) + self.b_output
self.output = z
self.action = tf.argmax(self.output, 1)
def make_hidden_graph(self):
with tf.name_scope("flatten"):
flat_input = tf.reshape(self.box_input, [-1, self.n_inputs - 1])
flat_input = tf.concat([flat_input, self.chain_input], 1)
with tf.name_scope("hidden"):
self.W_hidden = weight_variable([self.n_inputs, self.HIDDEN_SIZE],
name="W")
self.b_hidden = bias_variable([self.HIDDEN_SIZE], name="b")
z = tf.matmul(flat_input, self.W_hidden) + self.b_hidden
self.hidden_activation = tf.sigmoid(z)
def make_convolution_graph(self):
with tf.name_scope("convolution"):
self.W_conv = weight_variable([
self.KERNEL_SIZE, self.KERNEL_SIZE, self.box_shape[0],
self.NUM_FEATURES
],
name="W")
self.b_conv = bias_variable([self.NUM_FEATURES], name="b")
self.box_activations = []
for box in self.box_inputs:
z = conv2d(box, self.W_conv) + self.b_conv
self.box_activations.append(tf.sigmoid(z))
def ConvLayer(Input, FilterIn, FilterOut, Training, Scope):
with tf.variable_scope(Scope):
Weight = weight_variable([3, 3, FilterIn, FilterOut])
if cfg.LeakyReLU == True:
return tf.nn.leaky_relu(
batch_norm_conv(conv2d(Input, Weight), FilterOut, Training))
else:
return tf.nn.relu(
batch_norm_conv(conv2d(Input, Weight), FilterOut, Training))
def Dang_FNN(demand, topo):
x_demand = tf.reshape(demand, [-1,20,20,1])
x_topo = tf.reshape(topo, [-1,20,20,1])
x_in = tf.concat([ x_demand, x_topo], 3)
x_in_all = tf.reshape(x_in ,[-1,800])
# y
with tf.name_scope('D_fully_1_layer'):
with tf.name_scope('Weight'):
D_W_fc1 = util.weight_variable([800, 512])
with tf.name_scope('biases'):
D_b_fc1 = util.bias_variable([512])
with tf.name_scope('relu'):
D_h_fc1 = tf.nn.relu(tf.matmul(x_in_all, D_W_fc1) + D_b_fc1)
with tf.name_scope('D_fully_1_layer'):
with tf.name_scope('Weight'):
D_W_fc2 = util.weight_variable([512, 128])
with tf.name_scope('biases'):
D_b_fc2 = util.bias_variable([128])
with tf.name_scope('relu'):
D_h_fc2 = tf.nn.relu(tf.matmul(D_h_fc1, D_W_fc2) + D_b_fc2)
# dropoutd
with tf.name_scope('dropout'):
keep_prob1 = tf.placeholder(tf.float32, name = "drop_in")
h_fc1_drop = tf.nn.dropout(D_h_fc2, keep_prob1)
# readout
with tf.name_scope('fully_layer'):
with tf.name_scope('Weight'):
W_out = util.weight_variable([128, 1])
with tf.name_scope('biases'):
b_out = util.bias_variable([1])
y_out = tf.matmul(h_fc1_drop, W_out) + b_out
#return y_out, keep_prob1, keep_prob2, keep_prob_d, keep_prob_t
return y_out, keep_prob1
def make_fc_1_graph(self):
conv_layer_size = self.NUM_FEATURES * self.box_shape[
1] * self.box_shape[2]
with tf.name_scope("flatten"):
flat_input = self.chain_inputs[:]
for activation in self.box_activations:
flat_input.append(tf.reshape(activation,
[-1, conv_layer_size]))
flat_input = tf.concat(flat_input, 1)
n_flat = self.HISTORY_SIZE * (1 + conv_layer_size)
with tf.name_scope("fully_connected_1"):
self.W_fc_1 = weight_variable([n_flat, self.FC_1_SIZE], name="W")
self.b_fc_1 = bias_variable([self.FC_1_SIZE], name="b")
z = tf.matmul(flat_input, self.W_fc_1) + self.b_fc_1
self.fc_1_activation = tf.sigmoid(z)
# 输入层，这里规定的是输入数据的格式
x_raw = tf.placeholder(tf.float32, shape=[None,
784]) # 784 = 28x28，是图片的尺寸
y = tf.placeholder(tf.float32, shape=[None, 10]) # 10是图片的类别数，分别是 0~9
with tf.name_scope('hidden0'):
# 第一层，这里是为了处理数据，使得模型可以被迁移，在这里，只是简单地将数据reshape，后期可以做其他的尝试
l_pool0 = tf.reshape(x_raw, shape=[-1, 28, 28,
1]) # 最后的 1，是输出数目，可以理解为通道数
tf.summary.image('x_input', l_pool0, max_outputs=10)
with tf.name_scope('hidden1'):
# 第一个卷积层，从这一层开始到最后输出层为止，是要保留的模型
W_conv1 = util.weight_variable(
[5, 5, 1, 32],
name='W_conv1') # [5,5,1,32]中的1对应上一层的输出数目，而这里的输出数目是32
b_conv1 = util.bias_variable([32], name='b_conv1') # 32要跟上一行定义的32对应
l_conv1 = tf.nn.relu(
tf.nn.conv2d(
l_pool0, W_conv1, strides=[1, 1, 1, 1], padding='SAME') +
b_conv1) # 设置激活函数
# 池化层，这里[1,2,2,1]的步长会使输入数据的长宽分别减半， 最后是 [?, 14, 14, 32]
l_pool1 = tf.nn.max_pool(l_conv1,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME')
tf.summary.histogram('W_con1', W_conv1)
tf.summary.histogram('b_con1', b_conv1)
def SCNN(demand,topo) :
x_demand = tf.reshape(demand, [-1,8,8,1])
x_topo = tf.reshape(topo, [-1,20,20,1])
# convolution 1
with tf.name_scope('D_cnn_1_layer'):
with tf.name_scope('Weight'):
W_conv1_d = util.weight_variable([3, 3, 1, 16])
with tf.name_scope('biases'):
b_conv1_d = util.bias_variable([16])
# h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
d_conv1 = tf.nn.relu(util.conv2d(x_demand, W_conv1_d) + b_conv1_d)
#d_conv1 = util.conv2d(x_demand, W_conv1_d) + b_conv1_d
# h_pool1 = max_pool_2x2(h_conv1)
# convolution 2
with tf.name_scope('D_cnn_2_layer'):
with tf.name_scope('Weight'):
W_conv2_d = util.weight_variable([3, 3, 16, 32])
with tf.name_scope('biases'):
b_conv2_d = util.bias_variable([32])
# h_conv2 = tf.nn.relu(conv2d(h_conv1, W_conv2) + b_conv2)
d_conv2 = tf.nn.relu(util.conv2d(d_conv1, W_conv2_d) + b_conv2_d)
#d_conv2 = util.conv2d(d_conv1, W_conv2_d) + b_conv2_d
#d_conv2_flat = tf.reshape(d_conv2, [-1, 8 * 8 * 32])
# h_pool2 = max_pool_2x2(h_conv2)
# dropout
#with tf.name_scope('dropout'):
# keep_prob_d = tf.placeholder(tf.float32, name = "drop_in")
# d_drop = tf.nn.dropout(d_conv2, keep_prob_d)
# convolution 1
with tf.name_scope('T_cnn_1_layer'):
with tf.name_scope('Weight'):
W_conv1_t = util.weight_variable([5, 5, 1, 16])
with tf.name_scope('biases'):
b_conv1_t = util.bias_variable([16])
# h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
t_conv1 = tf.nn.relu(util.conv2d(x_topo, W_conv1_t) + b_conv1_t)
#t_conv1 = util.conv2d(x_topo, W_conv1_t) + b_conv1_t
# h_pool1 = max_pool_2x2(h_conv1)
# convolution 2
with tf.name_scope('T_cnn_2_layer'):
with tf.name_scope('Weight'):
W_conv2_t = util.weight_variable([5, 5, 16, 32])
with tf.name_scope('biases'):
b_conv2_t = util.bias_variable([32])
# h_conv2 = tf.nn.relu(conv2d(h_conv1, W_conv2) + b_conv2)
t_conv2 = tf.nn.relu(util.conv2d(t_conv1, W_conv2_t) + b_conv2_t)
#t_conv2 = util.conv2d(t_conv1, W_conv2_t) + b_conv2_t
#t_conv2_flat = tf.reshape(t_conv2, [-1, 8 * 8 * 32])
# h_pool2 = max_pool_2x2(h_conv2)
#dropout
# with tf.name_scope('dropout'):
# keep_prob_t = tf.placeholder(tf.float32, name = "drop_in")
# t_drop = tf.nn.dropout(t_conv2, keep_prob_t)
with tf.name_scope('Contact'):
D_conv2_seq = tf.reshape(d_conv2,[-1,8 * 8,32])
T_conv2_seq = tf.reshape(t_conv2,[-1,20 * 20,32])
Demand_Topo = tf.concat([D_conv2_seq, T_conv2_seq], 1)
# full-connected 1
Demand_Topo_flat = tf.reshape(Demand_Topo, [-1, (64 + 400) * 32])
with tf.name_scope('fully_1_layer'):
with tf.name_scope('Weight'):
W_fc1 = util.weight_variable([ (64 + 400)* 32, 512])
with tf.name_scope('biases'):
b_fc1 = util.bias_variable([512])
with tf.name_scope('relu'):
h_fc1 = tf.nn.relu(tf.matmul(Demand_Topo_flat, W_fc1) + b_fc1)
# dropout
with tf.name_scope('dropout'):
keep_prob1 = tf.placeholder(tf.float32, name = "drop_in")
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob1)
# full-connected 2
with tf.name_scope('fully_2_layer'):
with tf.name_scope('Weight'):
W_fc2 = util.weight_variable([512, 128])
with tf.name_scope('biases'):
b_fc2 = util.bias_variable([128])
with tf.name_scope('relu'):
h_fc2 = tf.nn.relu(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)
# dropout
with tf.name_scope('dropout'):
keep_prob2 = tf.placeholder(tf.float32, name = "drop_in")
h_drop = tf.nn.dropout(h_fc2, keep_prob2)
# readout
with tf.name_scope('fully_layer'):
with tf.name_scope('Weight'):
W_out = util.weight_variable([128, 1])
with tf.name_scope('biases'):
b_out = util.bias_variable([1])
y_out = tf.matmul(h_drop, W_out) + b_out
#return y_out, keep_prob1, keep_prob2, keep_prob_d, keep_prob_t
return y_out, keep_prob1, keep_prob2
def SFNN(demand,topo) :
x_demand = tf.reshape(demand, [-1,8 * 8])
x_topo = tf.reshape(topo, [-1,20 * 20])
# Demand Fully Connected 1
with tf.name_scope('D_fully_1_layer'):
with tf.name_scope('Weight'):
D_W_fc1 = util.weight_variable([64, 128])
with tf.name_scope('biases'):
D_b_fc1 = util.bias_variable([128])
with tf.name_scope('relu'):
D_h_fc1 = tf.nn.relu(tf.matmul(x_demand, D_W_fc1) + D_b_fc1)
# Demand Fully Connected 1
with tf.name_scope('D_fully_2_layer'):
with tf.name_scope('Weight'):
D_W_fc2 = util.weight_variable([ 128, 64])
with tf.name_scope('biases'):
D_b_fc2 = util.bias_variable([64])
with tf.name_scope('relu'):
D_h_fc2 = tf.nn.relu(tf.matmul(D_h_fc1, D_W_fc2) + D_b_fc2)
# Topo Fully Connected 1
with tf.name_scope('T_fully_1_layer'):
with tf.name_scope('Weight'):
T_W_fc1 = util.weight_variable([400, 512])
with tf.name_scope('biases'):
T_b_fc1 = util.bias_variable([512])
with tf.name_scope('relu'):
T_h_fc1 = tf.nn.relu(tf.matmul(x_topo, T_W_fc1) + T_b_fc1)
# Topo Fully Connected 2
with tf.name_scope('T_fully_2_layer'):
with tf.name_scope('Weight'):
T_W_fc2 = util.weight_variable([ 512, 256])
with tf.name_scope('biases'):
T_b_fc2 = util.bias_variable([256])
with tf.name_scope('relu'):
T_h_fc2 = tf.nn.relu(tf.matmul(T_h_fc1, T_W_fc2) + T_b_fc2)
with tf.name_scope('Contact'):
Demand_Topo = tf.concat([D_h_fc2, T_h_fc2], 1)
# full-connected 1
Demand_Topo_flat = tf.reshape(Demand_Topo, [-1, 256 + 64])
with tf.name_scope('fully_1_layer'):
with tf.name_scope('Weight'):
W_fc1 = util.weight_variable([ 256 + 64, 512])
with tf.name_scope('biases'):
b_fc1 = util.bias_variable([512])
with tf.name_scope('relu'):
h_fc1 = tf.nn.relu(tf.matmul(Demand_Topo_flat, W_fc1) + b_fc1)
# full-connected 2
with tf.name_scope('fully_2_layer'):
with tf.name_scope('Weight'):
W_fc2 = util.weight_variable([512, 128])
with tf.name_scope('biases'):
b_fc2 = util.bias_variable([128])
with tf.name_scope('relu'):
h_fc2 = tf.nn.relu(tf.matmul(h_fc1, W_fc2) + b_fc2)
# dropout
with tf.name_scope('dropout'):
keep_prob1 = tf.placeholder(tf.float32, name = "drop_in")
h_drop = tf.nn.dropout(h_fc2, keep_prob1)
# readout
with tf.name_scope('fully_layer'):
with tf.name_scope('Weight'):
W_out = util.weight_variable([128, 1])
with tf.name_scope('biases'):
b_out = util.bias_variable([1])
y_out = tf.matmul(h_drop, W_out) + b_out
return y_out, keep_prob1
sess = tf.InteractiveSession(config=config_opt)
# TensorBoard debug view.
file_writer = tf.summary.FileWriter('LOGS', sess.graph)
# Create placeholders for independent and dependant variables once batch has been selected.
with tf.name_scope('Input_Image'):
x = tf.placeholder(tf.float32, shape=[None, None, None, 1], name='Image') # Independent variables.
# Reshape to amenable shape.
# x_image = tf.reshape(x, [-1, windowSize[0], windowSize[1], 1])
with tf.name_scope('Input_Synapse'):
y_syn = tf.placeholder(tf.float32, shape=[None, 2]) # Target values.
with tf.name_scope('First_Layer'):
# Create first convolutional layer. (No pooling.)
W_conv1 = util.weight_variable(firstLayerDimensions, "w_conv_1") # Weights in first layer.
b_conv1 = util.bias_variable([firstLayerDimensions[3]], "b_conv_1") # Biases in first layer.
h_conv1 = tf.nn.relu(util.conv2d(x, W_conv1, valid=True, stride=1) + b_conv1) # Perform convolution (with zero padding) and apply ReLU.
h_pool1 = util.max_pool(h_conv1, 1) #, kernelWidth=2)
with tf.name_scope('Second_Layer'):
# Create first convolutional layer. (No pooling.)
W_conv2 = util.weight_variable(secondLayerDimensions, "w_conv_2") # Weights in first layer.
b_conv2 = util.bias_variable([secondLayerDimensions[3]], "b_conv_2") # Biases in first layer.
h_conv2 = tf.nn.relu(util.atrous_conv2d(h_pool1, W_conv2, valid=True, rate=2) + b_conv2) # Perform convolution (with zero padding) and apply ReLU.
h_pool2 = util.atrous_max_pool(h_conv2, mask_size=2, rate=2)
with tf.name_scope('Third_Layer'):
# Create first convolutional layer. (No pooling.)
W_conv3 = util.weight_variable(thirdLayerDimensions, "w_conv_3") # Weights in first layer.
b_conv3 = util.bias_variable([thirdLayerDimensions[3]], "b_conv_3") # Biases in first layer.
