tensorflow 学习纪录（持续更新）

import tensorflow as tf
import numpy as np

#tensor = tf.constant([[1,2,3,4,5,6,7,8],[1,2,3,4,5,6,7,8]])
tensor = tf.placeholder(tf.int32, [2,8])

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print sess.run(tensor,feed_dict={tensor:[[1,2,3,4,5,6,7,8],[1,2,3,4,5,6,7,8]]})
    print sess.run(tensor)
    tensorReshape = tf.reshape(tensor,[-1,4])
    print sess.run(tensorReshape)

print sess.run(tensor) 会报错，

InvalidArgumentError (see above for traceback): You must feed a value for placeholder tensor 'Placeholder' with dtype int32 and shape [2,8]
[[Node: Placeholder = Placeholder[dtype=DT_INT32, shape=[2,8], _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

原因：feed 使用一个 tensor 值临时替换一个操作的输出结果. 你可以提供 feed 数据作为 run() 调用的参数. feed 只在调用它的方法内有效, 方法结束, feed 就会消失。

 

 

tensorflow 加减op实验结果

y_out = tf.constant([1.0,2.0,3.0],shape=[3])
#y_out = tf.constant([[1.0,2.0,3.0],[2.0,3.0,4.0]],shape=[2,3])
yy = tf.constant([[1.0,2.0,3.0],[4.0,5.0,6.0],[1.0,2.0,3.0]],shape=[3,3])
print sess.run(y_out+yy)

 

二维和三维矩阵相乘

tf.matmul(M1,M2)

三维＊二维

    M1 = tf.Variable(tf.random_normal([2,3,4])

    M2 = tf.Variable(tf.random_normal([5,4])

    N = tf.einsum('ijk,lk->ijl',M1,M2)

tensorflow attention

 

tensorflow.python.framework.errors_impl.InvalidArgumentError: Cannot assign a device for operation 'Adadelta/update_post_embedding/Cast': Could not satisfy explicit device specification '' because the node was colocated with a group of nodes that required incompatible device '/job:localhost/replica:0/task:0/device:GPU:0'

安装完tensorflow-gpu版本后，跑程序会报这样的错，是因为GPU不支持一些op运算。

解决办法是安装对应的tensorflow cpu版本就完美解决了。

参考：https://github.com/cysmith/neural-style-tf/issues/18

 

tensorflow-gpu 1.4.1 不支持这两个函数

#optimizer = tf.train.AdadeltaOptimizer(learning_rate=1)
#optimizer = tf.train.AdagradOptimizer(learning_rate=0.1)