Tensorflow_2 Tutorials TF2 入门-Tensor

张量及其操作

Tensor的创建

1. 从numpy创建

   import tensorflow as tf
   import numpy as np
   
   
   if __name__ == '__main__':
       x = np.ones([1, 2], dtype=np.float)
       X = tf.convert_to_tensor(x, dtype=tf.float32)
       print(X) # tf.Tensor([1. 1.], shape=(2,), dtype=float32)
   

2. 从python内置类型创建

import tensorflow as tf
import numpy as np


if __name__ == '__main__':
    x = [5, 5]
    X = tf.convert_to_tensor(x, dtype=tf.float32)
    print(X) # tf.Tensor([5. 5.], shape=(2,), dtype=float32)

3. 使用tensorflow创建

   import tensorflow as tf
   import numpy as np
   
   
   if __name__ == '__main__':
       m, n = 3, 5
       a = tf.constant([0.] * 5, dtype=tf.float32) # 创建自定义常量
       b = tf.zeros(5, dtype=tf.float32) # 创建全0常量， 全一常量ones(...)， 单位矩阵eye...
       c = tf.Variable([0.] * 5, dtype=tf.float32)  # 创建变量 可求导！
       print(a, b , c, sep='\n')
   	"""
   		tf.Tensor([0. 0. 0. 0. 0.], shape=(5,), dtype=float32)
   		tf.Tensor([0. 0. 0. 0. 0.], shape=(5,), dtype=float32)
   		<tf.Variable 'Variable:0' shape=(5,) dtype=float32, numpy=array([0., 0., 0., 0., 0.], dtype=float32)>
   	"""
   

4. 类型转换
   你还在为创建随机的变量而苦恼吗？你还在为类型不匹配而焦虑吗？赶快拨打电话12315投诉吧(打了你就输了)

a = tf.constant([0.] * 5, dtype=tf.float32)
b = tf.Variable(a)
print(tf.is_tensor(b), type(b), b.dtype)
'''
True <class 'tensorflow.python.ops.resource_variable_ops.ResourceVariable'> <dtype: 'float32'>
#1 Variable 也是tensorflow的张量类型
#2 tf.Variable 可以将常量转化为 ResourceVariable
#3 dtype 不重新指定与constant的类型一致
'''
b = tf.cast(b, tf.float64)
print(b.dtype)
'''
<dtype: 'float64'>
#1 类型必须是tf的类型
#2 cast的返回值是变化后的向量，传入的原始向量不会被改变！
'''

Tensor的操作

数学运算

import tensorflow as tf
import numpy as np

if __name__ == '__main__':
	# 运算
    a = tf.ones([5, 5])
    q = tf.matmul(a, a) # 矩阵乘法 @
    w = tf.multiply(a, a) # 按位乘法 *
    e = tf.add(a, a) # 按位加法 +
    r = tf.subtract(a, a) # 按位减法 -
    t = tf.divide(a, a) # 按位除法 /
    ## a // a # 取整除法
    print(q, w, e, r, t)
    # 自动广播
    a = tf.ones([5, 5])
    b = tf.ones([5, 1]) * 2 # [[2.],[2.],[2.],[2.],[2.]]
    q = tf.matmul(a, b)
    w = tf.multiply(a, b)
    e = tf.add(a, b)
    r = tf.subtract(a, b)
    t = tf.divide(a, b)
    print(q, w, e, r, t)
    # 函数运算
    a = tf.ones([1, 1])
    q = tf.exp(a)
    w = tf.math.log(a) # log_e
    e = tf.cos(a)
    r = tf.sinh(a)
    t = tf.asin(a)
    print(q, w, e, r, t)

张量运算

import tensorflow as tf
import numpy as np

if __name__ == '__main__':
    from tensorflow.keras.datasets.mnist import load_data

    (x_train, y_train), (x_test, y_test) = load_data()

    X = tf.constant(x_train) # 60000 * 28 * 28

    print(X.shape, X.ndim) # 输出形状和维度
    # (60000, 28, 28) 3
    # 分割数据
    X1 = tf.split(X, 2) # 将数据分割为k块
    X1[0] = tf.expand_dims(X1[0], axis=-1) # 在最后面加一个维度
    X1[1] = tf.expand_dims(X1[1], axis=-1) # 在最后面加一个维度
    Y = tf.concat(X1, 3) # 按第三维度合并
    print(Y.shape) # TensorShape([30000, 28, 28, 2])

    # 改变形状
    print(tf.reshape(Y, [30000, 784, 2]).shape)
    print(tf.reshape(Y, [30000, -1, 2]).shape)
    # (30000, 784, 2)
    # (30000, 784, 2) # -1 只能有一个！表示余下所有
    print(tf.reshape(Y, [30000, -1, 28]).shape)
    # (30000, 56, 28)
    # 还原形状
    print(tf.equal(tf.reshape(tf.reshape(Y, [30000, -1, 28]), [30000, 28, 28, 2]), Y))
    # 压缩维度为1的维度
    print(X1[0].shape)
    print(tf.squeeze(X1[0]).shape)
    # (30000, 28, 28, 1)
    # (30000, 28, 28)

broadcast&tile&expand_dims

broadcast 广播，把低位数据广播到高维
expand_dims 扩展维度，在指定位置增加一个维度
tile 复制某些维度