tensorflow basic operation

创建Session

import tensorflow as tf
import matplotlib.pyplot as plt
n = 32
x = tf.linspace(-3.0, 3.0, n)

普通Session

sess = tf.Session()

交互式Session

import tensorflow as tf
sess = tf.InteractiveSession()

在sess内执行op

方法1

sess.run(tf.global_variables_initializer())
sess.run(op)
代码示例

sess.run(tf.global_variables_initializer())
result  = sess.run(x)

方法2

tf.global_variables_initializer().run()
sess.run(op)
op.eval()
代码示例

tf.global_variables_initializer().run()
x.eval(session=sess)
sess.close()

新op添加到默认图上

import tensorflow as tf
sigma = 1.0
mean = 0.0
# 和x的shape是一样的
z = (tf.exp(tf.negative(tf.pow(x - mean, 2.0) /
                        (2.0 * tf.pow(sigma, 2.0)))) *
     (1.0 / (sigma * tf.sqrt(2.0 * 3.1415))))
print(type(z))
print(z.graph is tf.get_default_graph())

plt.plot(z.eval())
plt.show()

查看shape

import tensorflow as tf
print(z.shape)
print(z.get_shape())
print(z.get_shape().as_list())
print(tf.shape(z).eval())

常用function

tf.stack

import tensorflow as tf
print(tf.stack([tf.shape(z),tf.shape(z),[3]]).eval())
# tf.reshape, tf.matmul
z_ = tf.matmul(tf.reshape(z, (n, 1)), tf.reshape(z, (1, n)))
plt.imshow(z_.eval()) plt.show()

tf.ones_like, tf.multiply

tf.ones_like返回与输入tensor具有相同shape的tensor

import tensorflow as tf
x = tf.reshape(tf.sin(tf.linspace(- 3.0, 3.0, n)), (n, 1))
print(x.shape)
y = tf.reshape(tf.ones_like(x), (1, n))
print(y.shape)
print(y.eval())
z = tf.multiply(tf.matmul(x,y), z_)
print(z.shape)
plt.imshow(z.eval())
plt.show()

列出graph中所有操作

import tensorflow as tf
ops = tf.get_default_graph().get_operations()
print([op for op in ops])

代码

完整地址

参考文献