\u672c\u5730\u7535\u8111\u7684\u914d\u7f6e\u8fbe\u4e0d\u5230\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u8bad\u7ec3\u7684\u8981\u6c42\uff0cGPU\u4e91\u670d\u52a1\u5668\u6309\u4f7f\u7528\u65f6\u95f4\u8ba1\u8d39\uff0c\u914d\u7f6e\u7075\u6d3b\uff0c\u4e3a\u6211\u4eec\u8bad\u7ec3\u590d\u6742\u6a21\u578b\u63d0\u4f9b\u4e86\u4e00\u4e2a\u89e3\u51b3\u529e\u6cd5\u3002<\/p>\n\n\n\n
\u4e0b\u9762\u4ee5\u300aDive into Deep Learning\u300b\uff08\u52a8\u624b\u5b66\u6df1\u5ea6\u5b66\u4e60\uff09\u7b2c2\u7248 \u7b2c7.1\u5c0f\u8282 \u300a\u6df1\u5ea6\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08AlexNet\uff09\u300b \u4e3a\u4f8b\uff0c\u4ecb\u7ecd\u5982\u4f55\u5728GPU\u4e91\u670d\u52a1\u5668\u4e0a\u8bad\u7ec3\u6a21\u578b\uff0c\u5e76\u53c2\u8003\u7b2c5.5\u5c0f\u8282\u300a\u8bfb\u5199\u6587\u4ef6\u300b\uff0c\u5c06\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u53c2\u6570\u4fdd\u5b58\u5230\u6587\u4ef6\uff0c\u4f9b\u4ee5\u540e\u4f7f\u7528\u3002<\/p>\n\n\n\n
\u2022 3. \u65b0\u5efapython\u6587\u4ef6\u5e76\u6267\u884c
\u65b0\u5efa\u4e00\u4e2a\u76ee\u5f55\u5e76\u8fdb\u5165
mkdir pytorch_train && cd $_
vim alexnet.py\uff0c\u5185\u5bb9\u5982\u4e0b\uff1a
<\/p>\n\n\n\n
# 7.1 AlexNet
import matplotlib
# Patch the missing private method to match the standard public .get method
if not hasattr(matplotlib.rcParams, '_get'):
matplotlib.rcParams._get = matplotlib.rcParams.get
import torch
from torch import nn
from d2l import torch as d2l
import matplotlib.pyplot as plt
net = nn.Sequential(
# \u8fd9\u91cc\u4f7f\u7528\u4e00\u4e2a11*11\u7684\u66f4\u5927\u7a97\u53e3\u6765\u6355\u6349\u5bf9\u8c61\u3002
# \u540c\u65f6\uff0c\u6b65\u5e45\u4e3a4\uff0c\u4ee5\u51cf\u5c11\u8f93\u51fa\u7684\u9ad8\u5ea6\u548c\u5bbd\u5ea6\u3002
# \u53e6\u5916\uff0c\u8f93\u51fa\u901a\u9053\u7684\u6570\u76ee\u8fdc\u5927\u4e8eLeNet
nn.Conv2d(1, 96, kernel_size=11, stride=4, padding=1), nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2),
# \u51cf\u5c0f\u5377\u79ef\u7a97\u53e3\uff0c\u4f7f\u7528\u586b\u5145\u4e3a2\u6765\u4f7f\u5f97\u8f93\u5165\u4e0e\u8f93\u51fa\u7684\u9ad8\u548c\u5bbd\u4e00\u81f4\uff0c\u4e14\u589e\u5927\u8f93\u51fa\u901a\u9053\u6570
nn.Conv2d(96, 256, kernel_size=5, padding=2), nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2),
# \u4f7f\u7528\u4e09\u4e2a\u8fde\u7eed\u7684\u5377\u79ef\u5c42\u548c\u8f83\u5c0f\u7684\u5377\u79ef\u7a97\u53e3\u3002
# \u9664\u4e86\u6700\u540e\u7684\u5377\u79ef\u5c42\uff0c\u8f93\u51fa\u901a\u9053\u7684\u6570\u91cf\u8fdb\u4e00\u6b65\u589e\u52a0\u3002
# \u5728\u524d\u4e24\u4e2a\u5377\u79ef\u5c42\u4e4b\u540e\uff0c\u6c47\u805a\u5c42\u4e0d\u7528\u4e8e\u51cf\u5c11\u8f93\u5165\u7684\u9ad8\u5ea6\u548c\u5bbd\u5ea6
nn.Conv2d(256, 384, kernel_size=3, padding=1), nn.ReLU(),
nn.Conv2d(384, 384, kernel_size=3, padding=1), nn.ReLU(),
nn.Conv2d(384, 256, kernel_size=3, padding=1), nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2),
nn.Flatten(),
# \u8fd9\u91cc\uff0c\u5168\u8fde\u63a5\u5c42\u7684\u8f93\u51fa\u6570\u91cf\u662fLeNet\u4e2d\u7684\u597d\u51e0\u500d\u3002\u4f7f\u7528dropout\u5c42\u6765\u51cf\u8f7b\u8fc7\u62df\u5408
nn.Linear(6400, 4096), nn.ReLU(),
nn.Dropout(p=0.5),
nn.Linear(4096, 4096), nn.ReLU(),
nn.Dropout(p=0.5),
# \u6700\u540e\u662f\u8f93\u51fa\u5c42\u3002\u7531\u4e8e\u8fd9\u91cc\u4f7f\u7528Fashion-MNIST\uff0c\u6240\u4ee5\u7528\u7c7b\u522b\u6570\u4e3a10\uff0c\u800c\u975e\u8bba\u6587\u4e2d\u76841000
nn.Linear(4096, 10))
X = torch.randn(1, 1, 224, 224)
for layer in net:
X=layer(X)
print(layer.__class__.__name__,'output shape:\\t',X.shape)
batch_size = 128
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
lr, num_epochs = 0.01, 10
d2l.train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())
plt.show()
plt.savefig(\"alexnet_train.png\")
# save model params
torch.save(net.state_dict(), \"alexnet.params\")
\u6267\u884c\u7a0b\u5e8f\uff0c\u6807\u51c6\u8f93\u51fa\u548c\u9519\u8bef\u8f93\u51fa\u90fd\u5b9a\u5411\u5230\u6587\u4ef6\u5e76\u5728\u540e\u53f0\u6267\u884c\uff08\u9632\u6b62ssh\u8fde\u63a5\u610f\u5916\u65ad\u5f00\u65f6\u7a0b\u5e8f\u672a\u6267\u884c\u5b8c\u5c31\u9000\u51fa\uff09
nohup python alexnet.py > output.log\u00a0 2>&1\u00a0 &
\u67e5\u770b\u6267\u884c\u60c5\u51b5\uff1a
tail -f output.log
\u6267\u884c\u5b8c\u6bd5\uff0c\u53ef\u4ee5\u770b\u5230\u751f\u6210\u7684\u635f\u5931\u548c\u6b63\u786e\u7387\u7684\u7ed8\u56fe\u6587\u4ef6alexnet_train.png\u548c\u6a21\u578b\u53c2\u6570\u6587\u4ef6alexnet.params\u3002
<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"\u672c\u5730\u7535\u8111\u7684\u914d\u7f6e\u8fbe\u4e0d\u5230\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u8bad\u7ec3\u7684\u8981\u6c42\uff0cGPU\u4e91\u670d\u52a1\u5668\u6309\u4f7f\u7528\u65f6\u95f4\u8ba1\u8d39\uff0c\u914d\u7f6e\u7075 … \u7ee7\u7eed\u9605\u8bfb