tf.Print

该op已经deprecated,目前已经不建议使用
以前大部分使用tf.Print 打印变量类似一下代码

1
2
3
4
some_op = ...
some_op = tf.Print(some_op, [tf.shape(some_op), "some_op:"])
...
sess.run(some_op)

这个API已经在新的里面被移除了,使用小写tf.print代替,但是操作方式有点差异

tf.print

tf.print
没有是一个print operation

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
import tensorflow as tf
import sys

print_ops = []

def print_op(*inputs , **kwargs):
    print_ops.append(tf.print(*inputs, **kwargs))

x = tf.range(10)

print_op("x", x)
print_op("x_no_summarize", x, summarize = -1)
print_op("x_output_stream", x, output_stream = sys.stdout)

# tf1.13开始支持 文件 file:// 开头
print_op("x_file", x, output_stream = "file:///tmp/debug.log")

with tf.Session() as sess:
    with tf.control_dependencies(print_ops):
        b = x * 3
    print(sess.run(b))

x [0 1 2 … 7 8 9]
x_no_summarize [0 1 2 3 4 5 6 7 8 9]
x_output_stream [0 1 2 … 7 8 9]
[ 0 3 6 9 12 15 18 21 24 27]

1
2
cat /tmp/debug.log                                                              
x_file [0 1 2 ... 7 8 9]

tf.print eager 模式

  1. 单个tensor

    1
    2
    3
    tf.compat.v1.enable_eager_execution()
    tensor = tf.range(10)
    tf.print(tensor, output_stream=sys.stderr)

  2. 多个tensor

    1
    2
    3
    tf.compat.v1.enable_eager_execution()
    tensor = tf.range(10)
    tf.print("tensors:", tensor, {2: tensor * 2}, output_stream=sys.stdout)

  3. 函数内

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    import tensorflow as tf
    import sys
    tf.enable_eager_execution()

    @tf.contrib.eager.defun
    def f():
        tensor = tf.range(10)
        tf.print("debug:", tensor, output_stream=sys.stderr)
        return tensor

    range_tensor = f()