1 Tensor运算特点
我们先来看看Tensor的运算特点吧
import torch
from time import time
print(torch.__version__)
1.10.0+cu113
a = torch.ones(1000)
b = torch.ones(1000)
将这两个向量按元素逐一做标量加法:
start = time()
c = torch.zeros(1000)
for i in range(1000):
c[i] = a[i] + b[i]
print(time() - start)
0.020173072814941406
将这两个向量直接做矢量加法:
start = time()
d = a + b
print(time() - start)
8.20159912109375e-05
结果很明显，后者比前者更省时。因此，我们应该尽可能采用矢量计算，以提升计算效率。
广播机制例子🌰：
a = torch.ones(3)
b = 10
print(a + b)
tensor([11., 11., 11.])
2 线性回归的从零开始实现
%matplotlib inline
import torch
from IPython import display
from matplotlib import pyplot as plt
import numpy as np
import random
print(torch.__version__)
1.10.0+cu113
生成数据集
num_inputs = 2
num_examples = 1000
true_w = [2, -3.4]
true_b = 4.2
features = torch.randn(num_examples, num_inputs,
dtype=torch.float32)
labels = true_w[0] * features[:, 0] + true_w[1] * features[:, 1] + true_b
labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()),
dtype=torch.float32)
print(features[0], labels[0])
tensor([0.1706, 1.0724]) tensor(0.8927)
def use_svg_display():
# 用矢量图显示
display.set_matplotlib_formats('svg')
def set_figsize(figsize=(3.5, 2.5)):
use_svg_display()
# 设置图的尺寸
plt.rcParams['figure.figsize'] = figsize
# # 在../d2lzh_pytorch里面添加上面两个函数后就可以这样导入
# import sys
# sys.path.append("..")
# from d2lzh_pytorch import *
set_figsize()
plt.scatter(features[:, 1].numpy(), labels.numpy(), 1);
# 本函数已保存在d2lzh包中方便以后使用
def data_iter(batch_size, features, labels):
num_examples = len(features)
indices = list(range(num_examples))
random.shuffle(indices) # 样本的读取顺序是随机的
for i in range(0, num_examples, batch_size):
j = torch.LongTensor(indices[i: min(i + batch_size, num_examples)]) # 最后一次可能不足一个batch
yield features.index_select(0, j), labels.index_select(0, j)
batch_size = 10
for X, y in data_iter(batch_size, features, labels):
print(X, '\n', y)
break
tensor([[-1.5561, 1.5366],
[ 0.7773, -0.4141],
[-0.7296, 0.2837],
[-0.1572, -0.1902],
[-1.1338, 0.0436],
[ 1.5135, -2.5492],
[ 0.5583, 0.2310],
[-1.3505, -1.5909],
[-0.2826, 1.3457],
[ 0.2002, -2.1443]])
tensor([-4.1393, 7.1701, 1.7999, 4.5387, 1.7729, 15.8816, 4.5391, 6.9128,
-0.9564, 11.9046])
初始化模型参数
w = torch.tensor(np.random.normal(0, 0.01, (num_inputs, 1)), dtype=torch.float32)
b = torch.zeros(1, dtype=torch.float32)
w.requires_grad_(requires_grad=True)
b.requires_grad_(requires_grad=True)
tensor([0.], requires_grad=True)
定义模型
def linreg(X, w, b): # 本函数已保存在d2lzh包中方便以后使用
return torch.mm(X, w) + b
定义损失函数
def squared_loss(y_hat, y): # 本函数已保存在pytorch_d2lzh包中方便以后使用
return (y_hat - y.view(y_hat.size())) ** 2 / 2
定义优化算法
def sgd(params, lr, batch_size): # 本函数已保存在d2lzh_pytorch包中方便以后使用
for param in params:
param.data -= lr * param.grad / batch_size # 注意这里更改param时用的param.data
训练模型
lr = 0.03
num_epochs = 3
net = linreg
loss = squared_loss
for epoch in range(num_epochs): # 训练模型一共需要num_epochs个迭代周期
# 在每一个迭代周期中，会使用训练数据集中所有样本一次（假设样本数能够被批量大小整除）。X
# 和y分别是小批量样本的特征和标签
for X, y in data_iter(batch_size, features, labels):
l = squared_loss(net(X, w, b), y).sum() # l是有关小批量X和y的损失
l.backward() # 小批量的损失对模型参数求梯度
sgd([w, b], lr, batch_size) # 使用小批量随机梯度下降迭代模型参数
# 不要忘了梯度清零
w.grad.data.zero_()
b.grad.data.zero_()
train_l = loss(net(features, w, b), labels)
print('epoch %d, loss %f' % (epoch + 1, train_l.mean().item()))
epoch 1, loss 0.031926 epoch 2, loss 0.000115 epoch 3, loss 0.000054
print(true_w, '\n', w)
print(true_b, '\n', b)
[2, -3.4]
tensor([[ 2.0000],
[-3.3996]], requires_grad=True)
4.2
tensor([4.1990], requires_grad=True) 3 线性回归的简洁实现
import torch
from torch import nn
import numpy as np
torch.manual_seed(1)
print(torch.__version__)
torch.set_default_tensor_type('torch.FloatTensor')
生成数据集
num_inputs = 2
num_examples = 1000
true_w = [2, -3.4]
true_b = 4.2
features = torch.tensor(np.random.normal(0, 1, (num_examples, num_inputs)), dtype=torch.float)
labels = true_w[0] * features[:, 0] + true_w[1] * features[:, 1] + true_b
labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()), dtype=torch.float)
读取数据
import torch.utils.data as Data
batch_size = 10
# 将训练数据的特征和标签组合
dataset = Data.TensorDataset(features, labels)
# 把 dataset 放入 DataLoader
data_iter = Data.DataLoader(
dataset=dataset, # torch TensorDataset format
batch_size=batch_size, # mini batch size
shuffle=True, # 要不要打乱数据 (打乱比较好)
num_workers=2, # 多线程来读数据
)
for X, y in data_iter:
print(X, '\n', y)
break
tensor([[-0.0163, -1.0072],
[-0.3554, -0.1807],
[-1.2406, -2.3683],
[ 1.3847, 1.9209],
[-0.7570, -0.3135],
[ 0.3181, -0.8122],
[-0.3864, 0.0382],
[ 1.0939, -0.1225],
[ 0.7272, 0.4801],
[ 0.6706, -0.7972]])
tensor([7.6005, 4.1017, 9.7864, 0.4568, 3.7355, 7.5675, 3.2881, 6.7967, 4.0404,
8.2513])
定义模型
class LinearNet(nn.Module):
def __init__(self, n_feature):
super(LinearNet, self).__init__()
self.linear = nn.Linear(n_feature, 1)
def forward(self, x):
y = self.linear(x)
return y
net = LinearNet(num_inputs)
print(net) # 使用print可以打印出网络的结构
LinearNet( (linear): Linear(in_features=2, out_features=1, bias=True) )
# 写法一
net = nn.Sequential(
nn.Linear(num_inputs, 1)
# 此处还可以传入其他层
)
# 写法二
net = nn.Sequential()
net.add_module('linear', nn.Linear(num_inputs, 1))
# net.add_module ......
# 写法三
from collections import OrderedDict
net = nn.Sequential(OrderedDict([
('linear', nn.Linear(num_inputs, 1))
# ......
]))
print(net)
print(net[0])
Sequential( (linear): Linear(in_features=2, out_features=1, bias=True) ) Linear(in_features=2, out_features=1, bias=True)
for param in net.parameters():
print(param)
Parameter containing: tensor([[0.5347, 0.7057]], requires_grad=True) Parameter containing: tensor([0.6873], requires_grad=True)
初始化模型参数
from torch.nn import init
init.normal_(net[0].weight, mean=0.0, std=0.01)
init.constant_(net[0].bias, val=0.0) # 也可以直接修改bias的data: net[0].bias.data.fill_(0)
Parameter containing: tensor([0.], requires_grad=True)
for param in net.parameters():
print(param)
Parameter containing: tensor([[-0.0142, -0.0161]], requires_grad=True) Parameter containing: tensor([0.], requires_grad=True)
定义损失函数
loss = nn.MSELoss()
定义优化算法
import torch.optim as optim
optimizer = optim.SGD(net.parameters(), lr=0.03)
print(optimizer)
SGD (
Parameter Group 0
dampening: 0
lr: 0.03
momentum: 0
nesterov: False
weight_decay: 0
)
# 为不同子网络设置不同的学习率
# optimizer =optim.SGD([
# # 如果对某个参数不指定学习率，就使用最外层的默认学习率
# {'params': net.subnet1.parameters()}, # lr=0.03
# {'params': net.subnet2.parameters(), 'lr': 0.01}
# ], lr=0.03)
# # 调整学习率
# for param_group in optimizer.param_groups:
# param_group['lr'] *= 0.1 # 学习率为之前的0.1倍
训练模型
num_epochs = 3
for epoch in range(1, num_epochs + 1):
for X, y in data_iter:
output = net(X)
l = loss(output, y.view(-1, 1))
optimizer.zero_grad() # 梯度清零，等价于net.zero_grad()
l.backward()
optimizer.step()
print('epoch %d, loss: %f' % (epoch, l.item()))
epoch 1, loss: 0.000457 epoch 2, loss: 0.000081 epoch 3, loss: 0.000198
dense = net[0]
print(true_w, dense.weight.data)
print(true_b, dense.bias.data)
[2, -3.4] tensor([[ 1.9999, -3.4005]]) 4.2 tensor([4.2011])
Comments NOTHING