tensorflow 2.0 基础操作 之 维度变换

3.4 维度变换

• tf.shape, tf.ndim
• tf.reshape
• tf.transpose
• tf.expand_dim
• tf.squeeze

tf.shape, tf.ndim

a = tf.random.normal([4, 28, 28, 3])
a.shape   # TensorShape([4, 28, 28, 3])
a.ndim   # 4 (int)

tf.reshape

a = tf.random.normal([4, 28, 28, 3])
a.shape   # TensorShape([4, 28, 28, 3])
a.ndim   # 4

tf.reshape(a, [4, 28*28, 3]).shape   #  TensorShape([4, 784, 3])
tf.reshape(a, [4, -1, 3]).shape   # TensorShape([4, 784, 3])	

tf.reshape(a, [4, 784*3]).shape   # TensorShape([4, 2352])
tf.reshape(a, [4, -1]).shape   # TensorShape([4, 2352])

# Reshape is flexible
tf.reshape(tf.reshape(a, [4, -1]),[4, 28, 28, 3]).shape   # TensorShape([4, 28, 28, 3])
tf.reshape(tf.reshape(a, [4, -1]),[4, -1, 3]).shape   # TensorShape([4, 784, 3])	

Reshape could lead to potential bugs!

images: [4, 28, 28, 3]
[b, h, w, 3]
reshape to: [4, 784, 3]
[b, pixel, 3]
[4, 784, 3] ℎ???ℎ?: 28, ????ℎ: 28 [4, 28, 28, 3]
[4, 784, 3] ℎ???ℎ?: 14, ????ℎ: 56 [4, 14, 56, 3]
[4, 784, 3] ????ℎ: 28, ℎ???ℎ?: 28 [4, 28, 28, 3]
 

tf.transpose

a = tf.random.normal([4, 3, 2, 1])
a.shape   # TensorShape([4, 3, 2, 1])
tf.transpose(a).shape   # TensorShape([1, 2, 3, 4])
a = tf.random.normal([4, 28, 28, 3]) 
a.shape   # TensorShape([4, 28, 28, 3])
tf.transpose(a, perm=[0,3,1,2]).shape    # TensorShape([4, 3, 28, 28])
# tensorflow tensor to pytorch tensor

tf.expand_dim

• a: [classes, students, classes]
  • [4, 35, 8]
• add school dim
• [1, 4, 35, 8] + [1, 4, 35, 8]
• [2, 4, 35, 8]

a = tf.ones([4,35,10])
a.shape   # TensorShape([4, 35, 10])
tf.expand_dims(a, axis=0).shape  # front   TensorShape([1, 4, 35, 10])
tf.expand_dims(a, axis=-1).shape  # behind   TensorShape([4, 35, 10, 1])
tf.expand_dims(a, axis=-2).shape  # behind   TensorShape([4, 35, 1, 10])

tf.squeeze

• Only squeeze for shape=1 dim

• [4, 35, 8, 1]

• [1, 4, 35, 8]

• [1, 4, 35, 1]

a = tf.ones([1,4,1,35,1,10])
tf.squeeze(a).shape   # # TensorShape([4, 35, 10])