Tensorflow(十一) —— Tensor的填充与复制
1. 主要方法
1、pad
2、tile
3、broadcast_to
2. pad方法
import numpy as np
import pandas as pd
import tensorflow as tf
"""
tf.pad(a,[[a11,a12],[a21,a22],...)
指定每个轴的前面填充a11条数据,后面填充a12条数据,默认填充为0
"""
a = tf.reshape(tf.range(9),[3,3])
print("a:",a.numpy())
a1 = tf.pad(a,[[0,0],[0,0]])
print("a1:",a1.numpy())
a2 = tf.pad(a,[[1,0],[0,0]]) # 行前面pad一行0
print("a2:",a2.numpy())
a3 = tf.pad(a,[[1,1],[0,0]],constant_values = 1)
print("a3:",a3.numpy())
a4 = tf.pad(a,[[1,1],[1,0]],constant_values=100)
print("a4:",a4.numpy())
a5 = tf.pad(a,[[1,1],[1,1]],constant_values=1000)
print("a5:",a5)
# image padding
b = tf.random.normal([4,28,28,3])
b1 = tf.pad(b,[[0,0],[2,2],[2,2],[0,0]]) # 行和列的前面各pad两条数据
print("b1:",b1.shape)
3. tile方法
# ************************ tile
"""
repeat data along dim n times
"""
a = tf.reshape(tf.range(9),[3,3])
print("a:",a.numpy())
a1 = tf.tile(a,[1,2])
print("a1:",a1.numpy(),a1.shape)
a2 = tf.tile(a,[2,1])
print("a2:",a2.numpy(),a2.shape)
a3 = tf.tile(a,[2,2]) # 先复制axis较小的轴
print("a3:",a3.numpy(),a3.shape)
4. tilt 与 broadcast_to的区别
"""
将[3,3] => [2,3,3]
"""
a = tf.reshape(tf.range(9),[3,3])
print("a:",a.numpy(),a.shape)
a1 = tf.expand_dims(a,axis = 0)
print("a1:",a1.numpy(),a1.shape)
b = tf.tile(a1,[2,1,1])
print("b:",b.numpy())
c = tf.broadcast_to(a,[2,3,3])
print("c:",c.numpy())
本文为参考龙龙老师的“深度学习与TensorFlow 2入门实战“课程书写的学习笔记
by CyrusMay 2022 04 16