5.pytorch学习：非线性激活——Non-linear Activations

目录

ReLU()

Sigmoid()

加入数据集查看效果

---

ReLU()

官网：

ReLU — PyTorch 1.10 documentationhttps://pytorch.org/docs/1.10/generated/torch.nn.ReLU.html#torch.nn.ReLU

用relu激活函数处理一个张量。

import torch
import torch.nn as nn

input = torch.randn(1, 3, 5, 5)


class ZiDingYi(nn.Module):
    def __init__(self):
        super(ZiDingYi, self).__init__()
        self.relu1 = nn.ReLU(inplace=False)

    def forward(self, x):
        output = self.relu1(input)
        return output


zidingyi = ZiDingYi()
output = zidingyi(input)
print(input)
print(output)

结果：根据relu函数的原理，对照两个tensor

tensor([[[[ 0.4775,  0.5198, -0.9001,  2.0372,  1.1024],
          [ 0.4123,  0.4447, -2.2582, -0.9545,  1.5849],
          [ 0.3350, -1.4441,  1.0662, -0.3530, -0.6260],
          [ 1.5861, -0.1249,  0.4190,  1.4067,  1.4868],
          [-2.0234, -1.1176, -2.6588,  0.1786,  0.8254]],

         [[ 1.5229,  0.4838, -1.0335,  0.0907, -0.0862],
          [-1.1347,  1.0359,  0.8005,  0.0480,  0.3533],
          [-1.6377, -0.8304,  1.1285, -1.6594,  2.0554],
          [-0.1789,  1.2895, -0.2744, -1.3726,  0.8518],
          [-0.8479,  0.4450,  0.1576, -1.0171,  0.1124]],

         [[ 0.1418,  0.1456, -0.4504,  0.6623, -1.0393],
          [-0.8508,  0.6686,  1.1152, -0.2147,  0.9810],
          [ 0.4100, -0.9736,  0.2510,  0.6155,  0.2945],
          [-0.2406,  0.8804,  0.8310, -0.3539,  0.2641],
          [-1.3649,  0.7333, -0.1349, -0.2516, -1.5076]]]])
tensor([[[[0.4775, 0.5198, 0.0000, 2.0372, 1.1024],
          [0.4123, 0.4447, 0.0000, 0.0000, 1.5849],
          [0.3350, 0.0000, 1.0662, 0.0000, 0.0000],
          [1.5861, 0.0000, 0.4190, 1.4067, 1.4868],
          [0.0000, 0.0000, 0.0000, 0.1786, 0.8254]],

         [[1.5229, 0.4838, 0.0000, 0.0907, 0.0000],
          [0.0000, 1.0359, 0.8005, 0.0480, 0.3533],
          [0.0000, 0.0000, 1.1285, 0.0000, 2.0554],
          [0.0000, 1.2895, 0.0000, 0.0000, 0.8518],
          [0.0000, 0.4450, 0.1576, 0.0000, 0.1124]],

         [[0.1418, 0.1456, 0.0000, 0.6623, 0.0000],
          [0.0000, 0.6686, 1.1152, 0.0000, 0.9810],
          [0.4100, 0.0000, 0.2510, 0.6155, 0.2945],
          [0.0000, 0.8804, 0.8310, 0.0000, 0.2641],
          [0.0000, 0.7333, 0.0000, 0.0000, 0.0000]]]])

Process finished with exit code 0

Sigmoid()

官网：

Sigmoid — PyTorch 1.10 documentationhttps://pytorch.org/docs/1.10/generated/torch.nn.Sigmoid.html#torch.nn.Sigmoid

import torch
import torch.nn as nn

input = torch.randn(1, 3, 5, 5)


class ZiDingYi(nn.Module):
    def __init__(self):
        super(ZiDingYi, self).__init__()
        # self.relu1 = nn.ReLU(inplace=False)
        self.sigmoid1 = nn.Sigmoid()

    def forward(self, x):
        # output = self.relu1(input)
        output = self.sigmoid1(input)
        return output


zidingyi = ZiDingYi()
output = zidingyi(input)
print(input)
print(output)

结果：根据Sigmoid函数的原理，对照两个tensor

tensor([[[[ 0.8320,  0.3261,  0.5211,  0.8005,  0.2384],
          [ 1.4506, -2.1106, -0.2674,  1.3780, -0.0732],
          [ 0.7145,  0.8867,  0.1283,  1.1789, -0.0598],
          [ 0.9552, -1.3852,  1.2036,  0.2603,  0.3854],
          [-1.5885, -2.0675, -1.3327,  0.6447,  1.9366]],

         [[-0.4702,  0.0669,  1.1686,  0.1894,  2.5503],
          [ 1.9532, -0.2793,  0.7396, -0.7552,  1.3760],
          [-0.1123,  1.0337, -0.5930,  0.0893, -0.6253],
          [ 0.6927, -0.2818,  0.5810, -0.2303, -1.0985],
          [ 0.5814, -0.6694,  0.1969, -0.0221, -0.8571]],

         [[-0.3682,  0.1004,  2.4205, -1.7189, -1.0245],
          [-0.3605, -1.0505,  0.1035,  0.6609, -0.3137],
          [ 0.0182,  1.1047,  1.5027,  0.5151,  0.2891],
          [ 0.7648,  1.3163, -0.1095, -1.2283,  0.5483],
          [ 0.6259,  0.2017,  0.9407, -0.7334, -1.2847]]]])
tensor([[[[0.6968, 0.5808, 0.6274, 0.6901, 0.5593],
          [0.8101, 0.1081, 0.4335, 0.7987, 0.4817],
          [0.6714, 0.7082, 0.5320, 0.7648, 0.4851],
          [0.7222, 0.2002, 0.7692, 0.5647, 0.5952],
          [0.1696, 0.1123, 0.2087, 0.6558, 0.8740]],

         [[0.3846, 0.5167, 0.7629, 0.5472, 0.9276],
          [0.8758, 0.4306, 0.6769, 0.3197, 0.7983],
          [0.4719, 0.7376, 0.3559, 0.5223, 0.3486],
          [0.6666, 0.4300, 0.6413, 0.4427, 0.2500],
          [0.6414, 0.3386, 0.5491, 0.4945, 0.2979]],

         [[0.4090, 0.5251, 0.9184, 0.1520, 0.2641],
          [0.4108, 0.2591, 0.5259, 0.6595, 0.4222],
          [0.5045, 0.7511, 0.8180, 0.6260, 0.5718],
          [0.6824, 0.7886, 0.4727, 0.2265, 0.6337],
          [0.6516, 0.5502, 0.7192, 0.3244, 0.2168]]]])

Process finished with exit code 0

加入数据集查看效果

import torch
import torch.nn as nn
import torchvision
from torch.utils.data import DataLoader

# input = torch.randn(1, 3, 5, 5)
from torch.utils.tensorboard import SummaryWriter


class ZiDingYi(nn.Module):
    def __init__(self):
        super(ZiDingYi, self).__init__()
        # self.relu1 = nn.ReLU(inplace=False)
        self.sigmoid1 = nn.Sigmoid()

    def forward(self, x):
        # output = self.relu1(input)
        x = self.sigmoid1(x)
        return x


train_dataset = torchvision.datasets.CIFAR10(root="./dataset", train=True, download=True,
                                             transform=torchvision.transforms.ToTensor())
train_dataloader = DataLoader(train_dataset, batch_size=64)

zidingyi = ZiDingYi()
# output = zidingyi(input)
# print(input)
# print(output)

writer = SummaryWriter("./logs_relu")
step = 0
for data in train_dataloader:
    imgs, targets = data
    writer.add_images("input", imgs, global_step=step)
    output = zidingyi(imgs)
    writer.add_images("output", output, global_step=step)
    step += 1
writer.close()

tensorboard --logdir=logs_relu

 根据sigmoid()函数曲线，观察输入图片(totensor)与输出图片(totensor)的差别