机器学习 / A LEAF IN THE GARDEN
LeNet
机器学习 · 动手深度学习。来自 Daily Learning 的学习笔记。
简介
LeNet-5 是由杨立昆教授在 1998 年提出的，是最早、最经典的卷积神经网络之一。最初被设计用于解决手写数字识别问题（著名的 MNIST 数据集），在银行业的支票识别系统中得到了广泛应用。
架构
卷积编码器 + 全连接层密集块
实现
代码 4 · In [1]
import torch
from torch import nn
import torchinfo
import util已保存输出
/root/autodl-tmp/envs/daily_learning/lib/python3.11/site-packages/requests/__init__.py:92: RequestsDependencyWarning: Unable to find acceptable character detection dependency (chardet or charset_normalizer). warnings.warn(
代码 5 · In [2]
class Reshape(nn.Module):
"""将输入张量重塑为指定形状"""
def forward(self, x):
# (batch_size, 输入通道, 高, 宽)
return x.view(-1, 1, 28, 28)
net = nn.Sequential(
Reshape(), # (B, 1, 28, 28)
nn.Conv2d(1, 6, kernel_size=5, padding=2), nn.Sigmoid(), # (B, 6, 28, 28)
nn.AvgPool2d(kernel_size=2, stride=2), # (B, 6, 14, 14)
nn.Conv2d(6, 16, kernel_size=5), nn.Sigmoid(), # (B, 16, 10, 10)
nn.AvgPool2d(kernel_size=2, stride=2), # (B, 16, 5, 5)
nn.Flatten(), # (B, 16 * 5 * 5)
nn.Linear(16 * 5 * 5, 120), nn.Sigmoid(), # (B, 120)
nn.Linear(120, 84), nn.Sigmoid(), # (B, 84)
nn.Linear(84, 10) # (B, 10)
)检查模型
代码 7 · In [3]
torchinfo.summary(net, input_size=(1, 28, 28), col_names=["input_size", "output_size", "num_params"])已保存输出
=================================================================================================================== Layer (type:depth-idx) Input Shape Output Shape Param # =================================================================================================================== Sequential [1, 28, 28] [1, 10] -- ├─Reshape: 1-1 [1, 28, 28] [1, 1, 28, 28] -- ├─Conv2d: 1-2 [1, 1, 28, 28] [1, 6, 28, 28] 156 ├─Sigmoid: 1-3 [1, 6, 28, 28] [1, 6, 28, 28] -- ├─AvgPool2d: 1-4 [1, 6, 28, 28] [1, 6, 14, 14] -- ├─Conv2d: 1-5 [1, 6, 14, 14] [1, 16, 10, 10] 2,416 ├─Sigmoid: 1-6 [1, 16, 10, 10] [1, 16, 10, 10] -- ├─AvgPool2d: 1-7 [1, 16, 10, 10] [1, 16, 5, 5] -- ├─Flatten: 1-8 [1, 16, 5, 5] [1, 400] -- ├─Linear: 1-9 [1, 400] [1, 120] 48,120 ├─Sigmoid: 1-10 [1, 120] [1, 120] -- ├─Linear: 1-11 [1, 120] [1, 84] 10,164 ├─Sigmoid: 1-12 [1, 84] [1, 84] -- ├─Linear: 1-13 [1, 84] [1, 10] 850 =================================================================================================================== Total params: 61,706 Trainable params: 61,706 Non-trainable params: 0 Total mult-adds (Units.MEGABYTES): 0.42 =================================================================================================================== Input size (MB): 0.00 Forward/backward pass size (MB): 0.05 Params size (MB): 0.25 Estimated Total Size (MB): 0.30 ===================================================================================================================
用 fashion MNIST 训练
代码 9 · In [4]
batch_size = 256
train_iter, test_iter = util.data.load_data_fashion_mnist(batch_size)
训练 LeNet-5
代码 11 · In [5]
lr, num_epochs = 0.1, 10
util.train.train_ch6(net, train_iter, test_iter, num_epochs, lr, device=util.try_gpu())已保存输出
loss 1.062, train acc 0.596, test acc 0.604 106508.1 examples/sec on cuda:0
代码 12 · In [6]
torch.save(net.state_dict(), "./models/LeNet.pth")