tensorflow: session开始对前tensor做一些处理

tf.boolean_mask

这个操作可以用于留下指定的元素，类似于numpy的操作。

import numpy as np
tensor = tf.range(4)
mask = np.array([True, False, True, False])
bool_mask = tf.boolean_mask(tensor, mask)
print sess.run(bool_mask)
[0 2]

tf.greater

首先张量x和张量y的尺寸要相同，输出的tf.greater(x, y)也是一个和x，y尺寸相同的张量。如果x的某个元素比y中对应位置的元素大，则tf.greater(x, y)对应位置返回True，否则返回False。

import tensorflow as tf

x = tf.Variable([[1,2,3], [6,7,8], [11,12,13]])
y = tf.Variable([[0,1,2], [5,6,7], [10,11,12]])

x1 = tf.Variable([[1,2,3], [6,7,8], [11,12,13]])
y1 = tf.Variable([[10,1,2], [15,6,7], [10,21,12]])

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(tf.greater(x, y)))
    print(sess.run(tf.greater(x1, y1)))



[[ True  True  True]
 [ True  True  True]
 [ True  True  True]]
[[False  True  True]
 [False  True  True]
 [ True False  True]]

tf.py_func

py_func(
    func,
    inp,
    Tout,
    stateful=True,
    name=None
)

参数:
func: 一个 Python 函数, 它接受 NumPy 数组作为输入和输出，并且数组的类型和大小必须和输入和输出用来衔接的 Tensor 大小和数据类型相匹配.
inp: 输入的 Tensor 列表.
Tout: 输出 Tensor 数据类型的列表或元祖.
stateful: 状态，布尔值.
name: 节点 OP 的名称.

i = tf.constant([[0,1,2,3,4],
                [9,8,0,3,0]])
a  = tf.cast(i,tf.bool)
b = tf.gather(i,1)
c = tf.not_equal(b,0)
neg_c = tf.logical_not(c)
indices = tf.where(c)
neg_indices = tf.where(neg_c)
def choose(x):
    return np.random.choice(np.ravel(x))
d = tf.py_func(choose,[indices],tf.int64)
with tf.Session() as sess:
    print(sess.run(a))
    print(sess.run(b))
    print(sess.run(c))
    print("neg_c:",sess.run(neg_c))
    print("indices:",sess.run(indices))
    print("neg_indices:",sess.run(neg_indices))
    print("....",sess.run(d))



[[False  True  True  True  True]
 [ True  True False  True False]]
[9 8 0 3 0]
[ True  True False  True False]
neg_c: [False False  True False  True]
indices: [[0]
 [1]
 [3]]
neg_indices: [[2]
 [4]]
.... 1

tf.cond

tf.cond(pred, true_fn=None, false_fn=None, strict=False, name=None, fn1=None, fn2=None)

Return true_fn() if the predicate pred is true else false_fn()

import tensorflow as tf

a = tf.placeholder(tf.bool)  #placeholder for a single boolean value
b = tf.cond(tf.equal(a, tf.constant(True)), lambda: tf.constant(10), lambda: tf.constant(0))
sess = tf.InteractiveSession()
res = sess.run(b, feed_dict = {a: True})
sess.close()
print(res)

10

tf.while_loop

tf.while_loop(
cond,
body,
loop_vars,
shape_invariants=None,
parallel_iterations=10,
back_prop=True,
swap_memory=False,
name=None,
maximum_iterations=None,
return_same_structure=False
)

作用：Repeat body while the condition cond is true

注意的是：loop_vars 是一个传递进去cond与body的 tuple, namedtuple or list of tensors . cond 与 body同时接受 both与 loop_vars一样多的参数。

例子：

def body(x):
    a = tf.random_uniform(shape=[2, 2], dtype=tf.int32, maxval=100)
    b = tf.constant(np.array([[1, 2], [3, 4]]), dtype=tf.int32)
    c = a + b
    return tf.nn.relu(x + c)
def condition(x):
    return tf.reduce_sum(x) < 100x = tf.Variable(tf.constant(0, shape=[2, 2]))with tf.Session():
    tf.initialize_all_variables().run()
    result = tf.while_loop(condition, body, [x])
    print(result.eval())