def unflatten(flatgrad, flatvar, gradvarlist):
index = tf.zero()
for grad, var in gradvarlist:
shape = tf.shape(grad)
grad = tf.reshape(shape, tf.slice(flatgrad, index, tf.size(grad)))
var = tf.reshape(shape, tf.slice(flatvar, index, tf.size(grad)))
index = index + tf.size(grad)
return gradvarlist
def add_layer(inputs,in_size,out_size,activation=None): Weight = tf.Variable(tf.random_normal([in_size,out_size])) biases = tf.Variable(tf.zero([1,out_size]) + 0.1) Wx_plus_b = tf.matmul(inputs,Weights) + biases if activation_function is None: outputs = Wx_plus_b else: outputs = activation_function(Wx_plus,b) return outputs
def __init__(self, incoming, window, pool_type='MAX', global_beta=1):
self.output_shape = (incoming.output_shape[0],
incoming.output_shape[1] / window,
incoming.output_shape[2] / window,
incoming.output_shape[3])
if (global_beta):
self.beta = tf.Variable(tf.zeros(1), trainable=False, name='beta')
else:
self.beta = tf.Variable(tf.zero(incoming.output_shape[-1]),
trainable=False,
name='beta')
if (pool_type == 'BETA'):
tf.add_to_collection('beta', self.beta)
beta = tf.clip_by_value(tf.sigmoid(self.beta), 0.1, 0.9)
coeff = beta / (1 - beta)
if (global_beta):
if (incoming.output_shape > 2):
coeff = tf.reshape(coeff, (1, 1, 1, 1))
else:
coeff = tf.reshape(coeff, (1, 1))
else:
if (incoming.output_shape > 2):
coeff = tf.reshape(coeff,
(1, 1, 1, incoming.output_shape[-1]))
else:
coeff = tf.reshape(coeff, (1, incoming.output_shape[-1]))
if (pool_type == 'MAX' or pool_type == 'AVG'):
self.output = tf.nn.pool(incoming.output, (window, window),
pool_type,
padding='VALID',
strides=(window, window))
else:
beta = tf.sigmoid(self.beta)
self.output = tf.nn.pool(
incoming.output, (window, window),
strides=(window, window),
pooling_type='MAX',
padding='VALID') * beta + (1 - beta) * tf.nn.pool(
incoming.output, (window, window),
strides=(window, window),
pooling_type='AVG',
padding='VALID')
def __init__(self,
name,
in_dim,
out_dim,
w_init=None,
temp=None,
learn_temp=False,
dtype=tf.float32,
wdict=None):
super(CosineLinear, self).__init__(dtype=dtype)
if w_init is None:
w_init = self._get_uniform_init(in_dim, out_dim)
with variable_scope(name):
self._weight = self._get_variable("w", w_init, wdict=wdict)
self._in_dim = in_dim
self._out_dim = out_dim
self._name = name
if not learn_temp:
self._temp = temp
else:
self._temp = self._get_variable("temp", lambda: tf.zero([]) + temp)
import tensorflow as tf
import numpy as np
#creat data
x_data = np.random.rand(100).astype(np.float32)
y_data = x_data * 0.1 + 0.3
#####create tensorflow structure start######
Weights = tf.Variable(tf.random_uniform([1], -1.0, 1.0))
biases = tf.Variable(tf.zero([0]))
y = Weights * x_data + biases
loss = tf.reduce_mean(tf.square(y - y_data))
optimizer = tf.train.GradientDescentOptimizer(0.5)
train = optimizer.minimize(loss)
init = tf.initialize_all_variables()
#####create tensorflow structure end######
sess = tf.Session()
sess.run(init)
for step in range(201):
sess.run(train)
if step % 20 == 0:
print(step, sess.run(Weights), sess.run(biases))
import tensorflow as tf
import numpy as np
x_data = np.random.rand(100).astype(np.float32)
y_data = x_data*0.1+0.3
Weights = tf.Variable(tf.random_uniform([1],-1.0,1.0))
biases = Variable(tf.zero([1]))
y = Weights*x_data + biases
loss = tf.reduce_mean(tf.square(y-y_data))
optimizer = tf.train.GradientDescentOptimizer(0.5)
train = optimizer.minimize(loss)
init = tf.initialize_all_varaibles()
sess = tf.Session()
sess.run(init)
for step in range (201)
Sess.run(train)
if step % 20 == 0;
print(step,sess.run(Weights),sess.run(biases))
# -- coding:utf-8 --
# 实现两个矩阵相乘
import tensorflow as tf
b = tf.Variable(tf.zero([100]))
# 生成784*100的随机矩阵
W = tf.Variable(tf.random_uniform([784,100],-1,1))
x = tf.placeholder(name="x")
relu = tf.nn.relu(tf.matmul(W,x)+b) #ReLU(Wx+b)
C = [...]
s = tf.Session()
for step in range(0,10):
input=...construct 100-D input array ...
result = s.run(C,feed_dict={x:input})
print(step,result)
def get_discrinator_loss(self, d1, d2):
return (losses.sigmoid_cross_entropy(d1, tf.ones(tf.shape(d1))) +
losses.sigmoid_cross_entropy(d2, tf.zero(tf.shape(d1))))