tf.where
tf.where(condition, x=None, y=None, name=None)
如果x，y均为空，则返回满足条件的索引indices
1
2
| a = tf.Variable([[1,2,3,4],[3,4,2,1]])
tf.where(a>3)
|
out:
1
2
3
4
| <tf.Tensor: shape=(2, 2), dtype=int64, numpy=
array([[0, 3],
[1, 1]])>
|
如果x，y均不为空，则 满足条件位置的值为x相应位置的值，其余为y相应位置的值。(非常实用)
1
2
3
| a = tf.Variable([[1,2,3,4],[3,4,2,1]])
b = tf.zeros_like(a)
tf.where(a>3, b, a)
|
out:
1
2
3
| <tf.Tensor: shape=(2, 4), dtype=int32, numpy=
array([[1, 2, 3, 0],
[3, 0, 2, 1]], dtype=int32)>
|
tf.stack
tf.stack(values, axis=0, name=”stack”)
向量堆叠函数
values.shape = (A, B, C)
if axis == 0 then the output tensor will have the shape (N, A, B, C).
if axis == 1 then the output tensor will have the shape (A, N, B, C).
example:
1
2
3
4
5
6
7
| a = tf.Variable([[1,2,0],[1,2,1]])
b = tf.Variable([[4,5,0],[4,5,1]])
c = tf.Variable([[7,8,0],[7,8,1]])
d = tf.Variable([[10,11,0],[10,11,1]])
stack_tensors_0 = tf.stack([a,b,c,d],axis=0)
stack_tensors_1 = tf.stack([a,b,c,d],axis=1)
|
out:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
| stack_tensors_0
<tf.Tensor: shape=(4, 2, 3), dtype=int32, numpy=
array([[[ 1, 2, 0],
[ 1, 2, 1]],
[[ 4, 5, 0],
[ 4, 5, 1]],
[[ 7, 8, 0],
[ 7, 8, 1]],
[[10, 11, 0],
[10, 11, 1]]], dtype=int32)>
stack_tensors_1
<tf.Tensor: shape=(2, 4, 3), dtype=int32, numpy=
array([[[ 1, 2, 0],
[ 4, 5, 0],
[ 7, 8, 0],
[10, 11, 0]],
[[ 1, 2, 1],
[ 4, 5, 1],
[ 7, 8, 1],
[10, 11, 1]]], dtype=int32)>
|
tf.gather
gather_v2(params,indices,validate_indices=None,axis=None,batch_dims=0,name=None)
根据索引，根据axis进行向量提取。只能根据一维度进行提取。
example:
1
2
| a = tf.Variable([[1,1,1],[2,2,2],[3,3,3]])
gather_tensor = tf.gather(a,[2,1,1,0])
|
out:
1
2
3
4
5
| <tf.Tensor: shape=(4, 3), dtype=int32, numpy=
array([[3, 3, 3],
[2, 2, 2],
[2, 2, 2],
[1, 1, 1]], dtype=int32)>
|
tf.gather_nd
gather_nd_v2(params, indices, batch_dims=0, name=None)
根据多维度进行提取
example:
假设text 有2个sequence，每个sequence有2个单词，经过Embedding 后的维度为4维。
根据每个sequence的序列进行抽取。
第0个sequence抽取序列为[1,1,0,0]
第1个sequence抽取序列为[0,1,0,1]
1
2
3
| text = tf.Variable([[[1,1,1,1],[2,2,2,2]],[[3,3,3,3],[4,4,4,4]]])
index=tf.Variable([[[0,1],[0,1],[0,0],[0,0]],[[1,0],[1,1],[1,0],[1,1]]])
gather_nd_tensor = tf.gather_nd(text,index)
|
out:
1
2
3
4
5
6
7
8
9
10
| gather_nd_tensor
<tf.Tensor: shape=(2, 4, 4), dtype=int32, numpy=
array([[[2, 2, 2, 2],
[2, 2, 2, 2],
[1, 1, 1, 1],
[1, 1, 1, 1]],
[[3, 3, 3, 3],
[4, 4, 4, 4],
[3, 3, 3, 3],
[4, 4, 4, 4]]], dtype=int32)>
|
tf.slice
slice(input_, begin, size, name=None)
tf切片函数
example:
1
2
3
4
| t = tf.constant([[[1, 1, 1], [2, 2, 2]],
[[3, 3, 3], [4, 4, 4]],
[[5, 5, 5], [6, 6, 6]]])
tf.slice(t, [1, 0, 0], [1, 1, 3])
|
out:
1
| <tf.Tensor: shape=(1, 1, 3), dtype=int32, numpy=array([[[3, 3, 3]]], dtype=int32)>
|
tf.cond
tf.cond(pred, true_fn=None, false_fn=None, name=None)
tf条件函数
可以利用tf.cond来动态选择层。在TF2.x中 可以直接用if条件进行替代。
1
2
3
| a=10
b=20
cond_layer = tf.cond(a>b,lambda :tf.keras.layers.Dense(a),lambda :tf.keras.layers.Dense(b))
|
tf.cumsum
tf.cumsum(x, axis=0, exclusive=False, reverse=False, name=None)
按轴累加器
1
2
3
| tf.cumsum([a,b,c]) = [a,a+b,a+b+c]
tf.cumsum([a,b,c],reverse=False) = [a+b+c,a+b,a]
tf.cumsum([a,b,c],exclusive=True) = [0,a,a+b]
|
example:
1
2
3
4
| a = tf.ones((4,3))
cumsum_1 = tf.cumsum(a)
cumsum_2 = tf.cumsum(a, axis=1)
cumsum_3 = tf.cumsum(a,exclusive=True)
|
out:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
| cumsum_1
<tf.Tensor: shape=(4, 3), dtype=float32, numpy=
array([[1., 1., 1.],
[2., 2., 2.],
[3., 3., 3.],
[4., 4., 4.]], dtype=float32)>
cumsum_2
<tf.Tensor: shape=(4, 3), dtype=float32, numpy=
array([[1., 2., 3.],
[1., 2., 3.],
[1., 2., 3.],
[1., 2., 3.]], dtype=float32)>
cumsum_3
<tf.Tensor: shape=(4, 3), dtype=float32, numpy=
array([[0., 0., 0.],
[1., 1., 1.],
[2., 2., 2.],
[3., 3., 3.]], dtype=float32)>
|
tf.clip_by_value
clip_by_value(t, clip_value_min, clip_value_max,name=None)
tf剪支函数
将t中 小于clip_value_min的替换成clip_value_min，大于 clip_value_max 替换成 clip_value_max
example：
1
2
| t = tf.Variable([[1,2,3,4],[5,6,7,8]])
clip_tensor= tf.clip_by_value(t,3,5)
|
out：
1
2
3
4
| clip_tensor
<tf.Tensor: shape=(2, 4), dtype=int32, numpy=
array([[3, 3, 3, 4],
[5, 5, 5, 5]], dtype=int32)>
|