tensorflow: Shapes and Shaping 探究

定义

Tensor Transformations - Shapes and Shaping: TensorFlow provides several operations that you can use to determine the shape of a tensor and change the shape of a tensor.

tensorflow提供了一些操作，让用户可以定义和修改tensor的形状

---

常用API

tf.shape

  以tensor形式，返回tensor形状。

tf.shape(input, name=None, out_type=tf.int32)

import tensorflow as tft = tf.constant(value=[[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]], dtype=tf.int32)with tf.Session() as sess:    print (sess.run(tf.shape(t)))

[2 2 3]

  另一种方法也可以的到类似答案：

import tensorflow as tft = tf.constant(value=[[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]], dtype=tf.int32)print (t.shape)

(2, 2, 3)

tf.size

  以tensor形式，返回tensor元素总数。

tf.size(input, name=None, out_type=tf.int32)

import tensorflow as tft = tf.ones(shape=[2, 5, 10], dtype=tf.int32)with tf.Session() as sess:    print (sess.run(tf.size(t)))

100

tf.rank

  以tensor形式，返回tensor阶数。

tf.rank(input, name=None)

import tensorflow as tft = tf.constant(value=[[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]], dtype=tf.int32)with tf.Session() as sess:    print (sess.run(tf.rank(t)))

3

tf.reshape

  以tensor形式，返回重新被塑形的tensor。

tf.reshape(tensor, shape, name=None)

import tensorflow as tft = tf.constant(value=[1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=tf.int32)with tf.Session() as sess:    print (sess.run(t))    print    print (sess.run(tf.reshape(t, [3, 3])))

[1 2 3 4 5 6 7 8 9][[1 2 3] [4 5 6] [7 8 9]]

tf.squeeze

  以tensor形式，返回移除指定维后的tensor。

tf.squeeze(input, axis=None, name=None, squeeze_dims=None)

• axis=None 时：
  Removes dimensions of size 1 from the shape of a tensor.
  将tensor中 维度为1 的 所有维 全部移除
• axis=[2, 4] 时：
  将tensor中 维度为1 的 第2维 和 第4维 移除

import tensorflow as tft = tf.ones(shape=[1, 2, 1, 3, 1, 1], dtype=tf.int32)with tf.Session() as sess:    print (sess.run(tf.shape(t)))    print    print (sess.run(tf.shape(tf.squeeze(t))))    print    print (sess.run(tf.shape(tf.squeeze(t, axis=[2, 4]))))

[1 2 1 3 1 1][2 3][1 2 3 1]

tf.expand_dims

  以tensor形式，返回插入指定维后的tensor。

tf.expand_dims(input, axis=None, name=None, dim=None)

import tensorflow as tft = tf.ones(shape=[2, 3, 5], dtype=tf.int32)with tf.Session() as sess:    print (sess.run(tf.shape(t)))    print    print (sess.run(tf.shape(tf.expand_dims(t, 0))))    print    print (sess.run(tf.shape(tf.expand_dims(t, 1))))    print    print (sess.run(tf.shape(tf.expand_dims(t, 2))))    print    print (sess.run(tf.shape(tf.expand_dims(t, 3))))    print    print (sess.run(tf.shape(tf.expand_dims(t, -1))))

[2 3 5][1 2 3 5][2 1 3 5][2 3 1 5][2 3 5 1][2 3 5 1]

---

---