合併與分割

目錄

• Merge and split
• concat
• Along distinct dim/axis
• stack: create new dim
• Dim mismatch
• Unstack
• Split

Merge and split

• tf.concat
• tf.split
• tf.stack
• tf.unstack

concat

• Statistics ablout scores
  • [class1-4,students,scores]
  • [class5-6,students,scores]

import tensorflow as tf

# 6個班級的學生分數狀況
a = tf.ones([4, 35, 8])
b = tf.ones([2, 35, 8])

c = tf.concat([a, b], axis=0)
c.shape

TensorShape([6, 35, 8])

# 3個學生學生補考
a = tf.ones([4, 32, 8])
b = tf.ones([4, 3, 8])

tf.concat([a, b], axis=1).shape

TensorShape([4, 35, 8])

Along distinct dim/axis

stack: create new dim

• Statistics about scores
  • School1:[classes,students,scores]
  • School2:[classes,students,scores]
  • [schools,calsses,students,scores]

a = tf.ones([4, 35, 8])
b = tf.ones([4, 35, 8])

a.shape

TensorShape([4, 35, 8])

b.shape

TensorShape([4, 35, 8])

tf.concat([a, b], axis=-1).shape

TensorShape([4, 35, 16])

tf.stack([a, b], axis=0).shape

TensorShape([2, 4, 35, 8])

tf.stack([a, b], axis=3).shape

TensorShape([4, 35, 8, 2])

Dim mismatch

a = tf.ones([4, 35, 8])
b = tf.ones([3, 33, 8])

try:
    tf.concat([a, b], axis=0).shape
except Exception as e:
    print(e)

ConcatOp : Dimensions of inputs should match: shape[0] = [4,35,8] vs. shape[1] = [3,33,8] [Op:ConcatV2] name: concat

# concat保證只有一個維度不相等
b = tf.ones([2, 35, 8])
c = tf.concat([a, b], axis=0)
c.shape

TensorShape([6, 35, 8])

# stack保證全部維度相等
try:
    tf.stack([a, b], axis=0)
except Exception as e:
    print(e)

Shapes of all inputs must match: values[0].shape = [4,35,8] != values[1].shape = [2,35,8] [Op:Pack] name: stack

Unstack

a.shape

TensorShape([4, 35, 8])

b = tf.ones([4, 35, 8])

c = tf.stack([a, b])

c.shape

TensorShape([2, 4, 35, 8])

aa, bb = tf.unstack(c, axis=0)

aa.shape, bb.shape

(TensorShape([4, 35, 8]), TensorShape([4, 35, 8]))

# [2,4,35,8]
res = tf.unstack(c, axis=3)

# 8個[2, 4, 35]的Tensor
res[0].shape, res[1].shape, res[7].shape

(TensorShape([2, 4, 35]), TensorShape([2, 4, 35]), TensorShape([2, 4, 35]))

# [2,4,35,8]
res = tf.unstack(c, axis=2)

# 35個[2, 4, 8]的Tensor
res[0].shape, res[1].shape, res[34].shape

(TensorShape([2, 4, 8]), TensorShape([2, 4, 8]), TensorShape([2, 4, 8]))

Split

• 相比較unstack靈活性更強

# 8個Tensor，全爲1
res = tf.unstack(c, axis=3)
len(res)

8

# 2個Tensor，一個六、一個2
res = tf.split(c, axis=3, num_or_size_splits=2)
len(res)

2

res[0].shape

TensorShape([2, 4, 35, 4])

res = tf.split(c, axis=3, num_or_size_splits=[2, 2, 4])

res[0].shape, res[1].shape, res[2].shape

(TensorShape([2, 4, 35, 2]),
 TensorShape([2, 4, 35, 2]),
 TensorShape([2, 4, 35, 4]))