Tensorflow2 math常用操作

张量的二元操作可以进行广播，与numpy的广播规则一样

一、tf.math

1.abs

返回张量的绝对值，输出类型、形状与输入一致

tf.math.abs(
    x,
    name=None
)

a = tf.constant([[1, 2, -1], [4, -3, 2]], dtype=tf.float32)
tf.abs(a)
>>>
<tf.Tensor: id=4, shape=(2, 3), dtype=float32, numpy=
array([[1., 2., 1.],
       [4., 3., 2.]], dtype=float32)>

2.accumulate_n

同add_n()，对多个张量对应位置求和

a = tf.constant([[1, 2], [3, 4]])
b = tf.constant([[5, 0], [0, 6]])
tf.math.accumulate_n([a, b, a])  # [[7, 4], [6, 14]]

# Explicitly pass shape and type
tf.math.accumulate_n([a, b, a], shape=[2, 2], tensor_dtype=tf.int32) # [[7,  4],[6, 14]]

3.add

x和y对应元素相加，相当于+

tf.math.add(
    x,
    y,
    name=None
)

4.add_n

同accumulate_n()

5.argmax

返回张量沿某一轴的最大值的下标

tf.math.argmax(
    input,
    axis=None,
    output_type=tf.dtypes.int64,
    name=None
)

A=tf.constant([2,20,30,3,6]) # Constant 1-D Tensor
tf.math.argmax(A) # output 2 as index 2 (A[2]) is maximum in tensor A
B=tf.constant([[2,20,30,3,6],[3,11,16,1,8],[14,45,23,5,27]])
tf.math.argmax(B,0) # [2, 2, 0, 2, 2]
tf.math.argmax(B,1) # [2, 2, 1]

axis原则

设axis=i ,则numpy沿着第i个下标变化的方向进行操作；所以，axis=0时，按列操作

6.argmin

与argmax()类似

7.bincount

8.ceil

对每个元素向上取整

输入类型必须是float

tf.math.ceil(
    x,
    name=None
)

9.floor和round

floor: 向下取整

round: 最近取整

10.confusion_matrix

计算标签和输出的混淆矩阵

tf.math.confusion_matrix(
    labels,
    predictions,
    num_classes=None,
    weights=None,
    dtype=tf.dtypes.int32,
    name=None
)

# num_class自动计算为4+1，标签自动假定为[0, 1, 2, 3, 4]
tf.math.confusion_matrix([1, 2, 4], [2, 2, 4]) ==>
      [[0 0 0 0 0]
       [0 0 1 0 0]
       [0 0 1 0 0]
       [0 0 0 0 0]
       [0 0 0 0 1]]

11. equal

x = tf.constant([2, 4]) 
y = tf.constant([2, 4]) 
tf.math.equal(x, y) 
>>>
<tf.Tensor: id=96, shape=(2,), dtype=bool, numpy=array([ True,  True])>

12. 逻辑操作

logical_and、logical_or、logical_not、logical_xor

x = tf.constant([False, False, True, True], dtype = tf.bool)
y = tf.constant([False, True, False, True], dtype = tf.bool)
z = tf.logical_xor(x, y, name="LogicalXor")
#  here z = [False  True  True False]

13. maximum和minimum

返回元素级 x > y ? x : y 结果

tf.math.maximum(
    x,
    y,
    name=None
)

x = tf.constant([[1, 3], [3, 2]], dtype=tf.float32)
y = tf.constant([[2, 3], [0, 4]], dtype=tf.float32)
tf.maximum(x, y)
>>>
<tf.Tensor: id=106, shape=(2, 2), dtype=float32, numpy=
array([[2., 3.],
       [3., 4.]], dtype=float32)>

14.pow

指数运算，相当于 x**y

tf.math.pow(
    x,
    y,
    name=None
)

x = tf.constant([[2, 2], [3, 3]])
y = tf.constant([[8, 16], [2, 3]])
tf.pow(x, y)  # [[256, 65536], [9, 27]]

15.reduce_max和reduce_min

张量沿某一轴的最大或最小值

设置keepdims=True，保持维度

tf.math.reduce_max(
    input_tensor,
    axis=None,
    keepdims=False,
    name=None
)

x = tf.constant([[1, 3], [5, 2]], dtype=tf.float32)
tf.reduce_max(x, axis=0)
>>>
<tf.Tensor: id=137, shape=(2,), dtype=float32, numpy=array([5., 3.], dtype=float32)>

tf.reduce_max(x, axis=0, keepdims=True)
>>>
<tf.Tensor: id=143, shape=(1, 2), dtype=float32, numpy=array([[5., 3.]], dtype=float32)>

16.reduce_mean和reduce_sum

沿张量某一轴求平均值/和

17.top_k

返回最后一维上的前K大的值和下标

tf.math.top_k(
    input,
    k=1,
    sorted=True,
    name=None
)

x = tf.constant([[1, 3], [5, 2]], dtype=tf.float32)
v, i =tf.math.top_k(x, 1)
print(v)
print(i)
>>>
tf.Tensor(
[[3.]
 [5.]], shape=(2, 1), dtype=float32) tf.Tensor(
[[1]
 [0]], shape=(2, 1), dtype=int32)