3.2 线性回归的从零开始
这是我的第一个代码,也算是属于自己的hello world了,特此纪念,希望继续努力。
代码中引入了3.1中的计时模块,用来对比训练时间。
import random
import torch
from d2l import torch as d2l
import sys
sys.path.append("..")
from timer import Timer
# 定时器计时
timer = Timer()
# 生成数据集
def synt_data(w, b, num):
X = torch.normal(0, 1, (num, len(w)))
y = torch.matmul(X, w) + b
y += torch.normal(0, 0.01, y.shape)
return X, y.reshape((-1, 1))
true_w = torch.tensor([2, -3.4])
true_b = 4.2
features, labels = synt_data(true_w, true_b, 1000)
# d2l.set_figsize()
# d2l.plt.scatter(features[: , 0].detach().numpy(), labels.detach().numpy(), 1)
# d2l.plt.show()
# 读取数据集
def data_iter(batch_size, features, labels):
num = len(features)
# 打乱下标
indices = list(range(num))
random.shuffle(indices)
for i in range(0, num, batch_size):
# 每次获取10个数据作为一个batch
batch_indices = torch.tensor(indices[i: min(i + batch_size, num)])
# 获取数据
yield features[batch_indices], labels[batch_indices]
batch_size = 10
# for X, y in data_iter(batch_size, features, labels):
# print(X, '\n', y)
# break
# 初始化模型参数
w = torch.normal(0, 0.01, (2,1), requires_grad=True)
b = torch.zeros(1, requires_grad=True)
# 定义模型
def linreg(X, w, b):
return torch.matmul(X, w) + b
#损失函数
def squ_loss(y_hat, y):
return (y_hat - y.reshape(y_hat.shape))**2 / 2
# 定义优化算法,简单梯度下降
def sgd(params, lr, batch_size):
with torch.no_grad():
for param in params:
param -= lr * param.grad /batch_size
param.grad.zero_()
# 训练
lr = 0.03
num_epochs = 3
net = linreg
loss = squ_loss
for epoch in range(num_epochs):
for X, y in data_iter(batch_size, features, labels):
l = loss(net(X, w ,b), y)
l.sum().backward()
sgd([w,b], lr, batch_size)
with torch.no_grad():
train_l = loss(net(features, w, b), labels)
print(f'epoch {epoch + 1 }, loss {float(train_l.mean()):f}')
print(f'time {timer.stop(): .5f} sec')
3.3 线性回归的简单实现
这段代码敲的时候有几个有趣的发现:
- 手动写的梯度下降函数里面,有自动的梯度清零和参数更新,但是torch实现的SGD应该是没有的,所以才需要使用
trainer.zero_grad()
和trainer.step()
。 yeild
的使用可以实现迭代器一样的效果
import numpy as np
import torch
from torch.utils import data
from torch import nn
from d2l import torch as d2l
import sys
sys.path.append("..")
from timer import Timer
# 定时器计时
timer = Timer()
# 生成数据集
true_w = torch.tensor([2, -3.4])
true_b = 4.2
features, labels = d2l.synthetic_data(true_w, true_b, 1000)
# 读取数据集
def load_array(data_arrays, batch_size, is_train = True):
# 构建一个迭代器
dataset = data.TensorDataset(*data_arrays)
return data.DataLoader(dataset, batch_size, shuffle=is_train)
batch_size = 10
data_iter = load_array((features, labels), batch_size)
# 定义模型
net = nn.Sequential(nn.Linear(2, 1))
# 定义参数
net[0].weight.data.normal_(0, 0.01)
net[0].bias.data.fill_(0)
# 定义损失函数
loss = nn.MSELoss()
# 定义优化算法
trainer = torch.optim.SGD(net.parameters(), lr = 0.03)
# 训练
num_epoch = 3
for epoch in range(num_epoch):
for X, y in data_iter:
l = loss(net(X), y)
trainer.zero_grad()
l.backward()
trainer.step()
l = loss(net(features), labels)
print(f'epoch {epoch + 1}, loss {l:f}')
print(f'time {timer.stop(): .5f} sec')
对比
两个代码的运行结果分别为:
(d2l) z**@e****:~/deeplearning/linear_regression$ ***/miniconda3/envs/d2l/bin/python ***/deeplearning/linear_regression/model_simple.py
epoch 1, loss 0.000301
epoch 2, loss 0.000114
epoch 3, loss 0.000114
time 0.16474 sec
(d2l) z**@e****:~/deeplearning/linear_regression$ ***/miniconda3/envs/d2l/bin/python ***/deeplearning/linear_regression/model.py
epoch 1, loss 0.028095
epoch 2, loss 0.000099
epoch 3, loss 0.000052
time 0.13666 sec
可以看到,针对于简单的线性回归而言,手动写的代码无论是最终的精度还是时间上都是更优的,搜索到的可能的原因是:使用现有的机器学习框架可能会带来一些开销,例如框架本身的启动时间、内存占用等,手动编写的代码可以避免这些开销。
以上就是全部的内容了,由于作者刚开始学习,能力浅薄,待我学成归来,也许会有更深的了解。
标签:loss,torch,batch,学习,epoch,3.2,3.3,data,size From: https://www.cnblogs.com/zcry/p/18048424