前向传播:
y=torch.mul(w,x) #等价于w*x
y=torch.mm(w,x)
z=torch.add(y,b) #等价于y+b
#定义输入张量x
x = torch.tensor(2.0, requires_grad=True)
# 初始化权重参数W,偏移量b、并设置require_grad属性为True,自动跟踪历史导数
w=torch.randn(1,requires_grad=True) # torch.rand(2,2,requires_grad=True)
b=torch.randn(1,requires_grad=True)
# 实现前向传播
y=torch.mul(w,x) #等价于w*x
z=torch.add(y,b) #等价于y+b
#查看x,w,b页子节点的requite_grad属性
# requires_grad(自动获取梯度)设置为True
print("x,w,b的require_grad属性分别为:{},{},{}".format(x.requires_grad,w.requires_grad,b.requires_grad))
print(x)
print("w",w)
print('b',b)
print('y',y)
print('z',z)
反向传播:
y3.backward()
手动实现机器学习:
import torch
from IPython import display
from matplotlib import pyplot as plt
import numpy as np
import random
%matplotlib inline
构建神经网络:
定义模型 类创建,CLASS Net(父类:nn.moudle)
初始化函数: 参数-n_feature 特征个数
继承:super(Net,self).__init—(), 初始化父类
初始化自己:self.linear=nn.linear(输入特征维度,输出维度,偏置=True默认)
遍历 for item in self.linear.params:
参数出来了
初始化参数: nn.init.normul(item,m=,)
定义前向传播
y = self.linear(x)
return y
实例化对象
import numpy as np
import torch
from torch import nn
初始化模型参数
损失函数:
优化算法:
训练模型
class Net(nn.Module):
"""
使用sequential构建网络,Sequential()函数的功能是将网络的层组合到一起
"""
#indim 输入维度. n_hidden 隐藏层个数
def __init__(self, in_dim, n_hidden_1, n_hidden_2, out_dim):
#继承父类属性
super(Net, self).__init__()
#定义自己每一层,Sequential,拼接每一层方法(第一步线性连接nn.linear(输入,输出),第二归一化处理nn.BatchNorm1d(n_hidden_1))
self.layer1 = nn.Sequential(nn.Linear(in_dim, n_hidden_1),nn.BatchNorm1d(n_hidden_1))
self.layer2 = nn.Sequential(nn.Linear(n_hidden_1, n_hidden_2),nn.BatchNorm1d(n_hidden_2))
self.layer3 = nn.Sequential(nn.Linear(n_hidden_2, out_dim))
def forward(self, z):
#
z = F.relu(self.layer1(z))
z = F.relu(self.layer2(z))
z = self.layer3(z)
return z
model = Net(28 * 28, 300, 100, 10)
字体识别
import numpy as np
import torch
# 导入 pytorch 内置的 mnist 数据
from torchvision.datasets import mnist
#导入预处理模块
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
#导入nn及优化器
import torch.nn.functional as F
import torch.optim as optim
from torch import nn
import matplotlib.pyplot as plt
%matplotlib inlin
说明
①transforms.Compose可以把一些转换函数组合在一起;
②Normalize([0.5], [0.5])对张量进行归一化,这里两个0.5分别表示对张量进行归一化的全局平均值和方差。因图像是灰色的只有一个通道,如果有多个通道,需要有多个数字,如三个通道,应该是Normalize([m1,m2,m3], [n1,n2,n3])
③download参数控制是否需要下载,如果./data目录下已有MNIST,可选择False。
④用DataLoader得到生成器,这可节省内存。
train_batch_size = 64
test_batch_size = 128
learning_rate = 0.01
num_epoches = 20
lr = 0.01
momentum = 0.5
#定义预处理函数,这些预处理依次放在Compose函数中。
transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize([0.5], [0.5])])
#下载数据,并对数据进行预处理
train_dataset = mnist.MNIST('./data', train=True, transform=transform, download=True)
test_dataset = mnist.MNIST('./data', train=False, transform=transform)
#dataloader是一个可迭代对象,可以使用迭代器一样使用。
train_loader = DataLoader(train_dataset, batch_size=train_batch_size, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=test_batch_size, shuffle=False)
examples = enumerate(test_loader)
batch_idx, (example_data, example_targets) = next(examples)
fig = plt.figure()
for i in range(6):
plt.subplot(2,3,i+1)
plt.tight_layout()
# example_data[i][0] 这个操作是拿出
plt.imshow(example_data[i][0], cmap='summer', interpolation='none')
plt.title("Ground Truth: {}".format(example_targets[i]))
创建神经网络预测:
class Net(nn.Module):
"""
使用sequential构建网络,Sequential()函数的功能是将网络的层组合到一起
"""
def __init__(self, in_dim, n_hidden_1, n_hidden_2, out_dim):
super(Net, self).__init__()
self.layer1 = nn.Sequential(nn.Linear(in_dim, n_hidden_1),nn.BatchNorm1d(n_hidden_1))
self.layer2 = nn.Sequential(nn.Linear(n_hidden_1, n_hidden_2),nn.BatchNorm1d(n_hidden_2))
self.layer3 = nn.Sequential(nn.Linear(n_hidden_2, out_dim))
def forward(self, z):
z = F.relu(self.layer1(z))
z = F.relu(self.layer2(z))
z = self.layer3(z)
return z
#检测是否有可用的GPU,否则使用CPU
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#实例化网络,定义隐藏层个数
model = Net(28 * 28, 300, 100, 10)
# model.to(device)
# 定义损失函数和优化器
# 定义交叉熵损失
criterion = nn.CrossEntropyLoss()
# 梯度下降model.parameters() 所有参数,矩阵
optimizer = optim.SGD(model.parameters(), lr=lr, momentum=momentum)
训练模型
# 开始训练
losses = [] #训练
acces = []
eval_losses = [] #评估
eval_acces = []
for epoch in range(num_epoches):
train_loss = 0
train_acc = 0
# 训练模式
model.train()
#动态修改参数学习率
if epoch%5==0:
optimizer.param_groups[0]['lr']*=0.1
# 做训练
for img, label in train_loader:
# img=img.to(device)
# label = label.to(device)
# n行一列
img = img.view(img.size(0), -1)
# 前向传播
out = model(img)
# 计算损失
loss = criterion(out, label)
# 反向传播
optimizer.zero_grad()
loss.backward()
# 参数更新
optimizer.step()
# 记录误差
train_loss += loss.item() #python
# 计算分类的准确率
_, pred = out.max(1)
num_correct = (pred == label).sum().item()
acc = num_correct / img.shape[0]
train_acc += acc
losses.append(train_loss / len(train_loader))
acces.append(train_acc / len(train_loader))
# 在测试集上检验效果
eval_loss = 0
eval_acc = 0
# 将模型改为预测模式
model.eval()
for img, label in test_loader:
img=img.to(device)
label = label.to(device)
img = img.view(img.size(0), -1)
out = model(img)
loss = criterion(out, label)
# 记录误差
# .item() 转化为python可以识别的类型
eval_loss += loss.item()
# 记录准确率
_, pred = out.max(1)
#判断布尔索引是否相等,求和
num_correct = (pred == label).sum().item()
acc = num_correct / img.shape[0]
# 每一层循环的损失都累加一次,
eval_acc += acc
eval_losses.append(eval_loss / len(test_loader))
eval_acces.append(eval_acc / len(test_loader))
print('epoch: {}, Train Loss: {:.4f}, Train Acc: {:.4f}, Test Loss: {:.4f}, Test Acc: {:.4f}'
.format(epoch, train_loss / len(train_loader), train_acc / len(train_loader),
eval_loss / len(test_loader), eval_acc / len(test_loader)))