MLP神经网络的训练
知识点回顾：
- PyTorch和cuda的安装
- 查看显卡信息的命令行命令（cmd中使用）
- cuda的检查
- 简单神经网络的流程
- 数据预处理（归一化、转换成张量）
- 模型的定义
- 继承nn.Module类
- 定义每一个层
- 定义前向传播流程
- 定义损失函数和优化器
- 定义训练流程
- 可视化loss过程
python
import torch
import torch.nn as nn
import torch.optim as optim
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler
import numpy as np
import matplotlib.pyplot as plt
# 检查CUDA可用性并输出相关信息
if torch.cuda.is_available():
print("CUDA可用!")
device_count = torch.cuda.device_count()
print(f"可用的CUDA设备数量:{device_count}")
current_device = torch.cuda.current_device()
print(f"当前使用的CUDA设备索引:{current_device}")
device_name = torch.cuda.get_device_name(current_device)
print(f"当前CUDA设备的名称:{device_name}")
cuda_version = torch.version.cuda
print(f"CUDA版本:{cuda_version}")
else:
print("CUDA不可用。")
# 加载并准备Iris数据集
iris = load_iris()
X = iris.data
y = iris.target
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
print(X_train.shape)
print(y_train.shape)
print(X_test.shape)
print(y_test.shape)
# 数据标准化处理
scaler = MinMaxScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)
# 转换为PyTorch张量
X_train = torch.FloatTensor(X_train)
y_train = torch.LongTensor(y_train)
X_test = torch.FloatTensor(X_test)
y_test = torch.LongTensor(y_test)
# 定义多层感知机模型
class MLP(nn.Module):
def __init__(self):
super(MLP, self).__init__()
self.fc1 = nn.Linear(4, 10)
self.relu = nn.ReLU()
self.fc2 = nn.Linear(10, 3)
def forward(self, x):
out = self.fc1(x)
out = self.relu(out)
out = self.fc2(out)
return out
# 初始化模型、损失函数和优化器
model = MLP()
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01) # 添加了学习率参数
# 训练模型
num_epochs = 20000
losses = []
for epoch in range(num_epochs):
outputs = model.forward(X_train)
loss = criterion(outputs, y_train)
optimizer.zero_grad()
loss.backward()
optimizer.step()
losses.append(loss.item())
if (epoch + 1) % 100 == 0:
print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {loss.item():.4f}')
# 可视化训练损失
plt.plot(range(num_epochs), losses)
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.title('Training Loss over Epochs')
plt.show()输出结果：
python
(120, 4)
(120,)
(30, 4)
(30,)
Epoch [100/20000], Loss: 1.0980
Epoch [200/20000], Loss: 1.0692
Epoch [300/20000], Loss: 1.0398
Epoch [400/20000], Loss: 1.0060
Epoch [500/20000], Loss: 0.9665
Epoch [600/20000], Loss: 0.9214
Epoch [700/20000], Loss: 0.8731
Epoch [800/20000], Loss: 0.8241
Epoch [900/20000], Loss: 0.7777
Epoch [1000/20000], Loss: 0.7360
Epoch [1100/20000], Loss: 0.6991
Epoch [1200/20000], Loss: 0.6666
Epoch [1300/20000], Loss: 0.6380
Epoch [1400/20000], Loss: 0.6127
Epoch [1500/20000], Loss: 0.5903
Epoch [1600/20000], Loss: 0.5703
Epoch [1700/20000], Loss: 0.5523
Epoch [1800/20000], Loss: 0.5360
Epoch [1900/20000], Loss: 0.5211
Epoch [2000/20000], Loss: 0.5074
Epoch [2100/20000], Loss: 0.4948
Epoch [2200/20000], Loss: 0.4830
Epoch [2300/20000], Loss: 0.4720
Epoch [2400/20000], Loss: 0.4616
Epoch [2500/20000], Loss: 0.4518
Epoch [2600/20000], Loss: 0.4425
Epoch [2700/20000], Loss: 0.4336
Epoch [2800/20000], Loss: 0.4251
Epoch [2900/20000], Loss: 0.4168
Epoch [3000/20000], Loss: 0.4088
Epoch [3100/20000], Loss: 0.4010
Epoch [3200/20000], Loss: 0.3935
Epoch [3300/20000], Loss: 0.3860
Epoch [3400/20000], Loss: 0.3787
Epoch [3500/20000], Loss: 0.3716
Epoch [3600/20000], Loss: 0.3645
Epoch [3700/20000], Loss: 0.3576
Epoch [3800/20000], Loss: 0.3508
Epoch [3900/20000], Loss: 0.3440
Epoch [4000/20000], Loss: 0.3374
Epoch [4100/20000], Loss: 0.3308
Epoch [4200/20000], Loss: 0.3243
Epoch [4300/20000], Loss: 0.3179
Epoch [4400/20000], Loss: 0.3115
Epoch [4500/20000], Loss: 0.3053
Epoch [4600/20000], Loss: 0.2991
Epoch [4700/20000], Loss: 0.2930
Epoch [4800/20000], Loss: 0.2870
Epoch [4900/20000], Loss: 0.2810
Epoch [5000/20000], Loss: 0.2752
Epoch [5100/20000], Loss: 0.2694
Epoch [5200/20000], Loss: 0.2638
Epoch [5300/20000], Loss: 0.2583
Epoch [5400/20000], Loss: 0.2530
Epoch [5500/20000], Loss: 0.2478
Epoch [5600/20000], Loss: 0.2428
Epoch [5700/20000], Loss: 0.2378
Epoch [5800/20000], Loss: 0.2330
Epoch [5900/20000], Loss: 0.2284
Epoch [6000/20000], Loss: 0.2238
Epoch [6100/20000], Loss: 0.2193
Epoch [6200/20000], Loss: 0.2150
Epoch [6300/20000], Loss: 0.2108
Epoch [6400/20000], Loss: 0.2067
Epoch [6500/20000], Loss: 0.2027
Epoch [6600/20000], Loss: 0.1989
Epoch [6700/20000], Loss: 0.1951
Epoch [6800/20000], Loss: 0.1914
Epoch [6900/20000], Loss: 0.1878
Epoch [7000/20000], Loss: 0.1844
Epoch [7100/20000], Loss: 0.1810
Epoch [7200/20000], Loss: 0.1778
Epoch [7300/20000], Loss: 0.1746
Epoch [7400/20000], Loss: 0.1716
Epoch [7500/20000], Loss: 0.1686
Epoch [7600/20000], Loss: 0.1658
Epoch [7700/20000], Loss: 0.1630
Epoch [7800/20000], Loss: 0.1604
Epoch [7900/20000], Loss: 0.1578
Epoch [8000/20000], Loss: 0.1553
Epoch [8100/20000], Loss: 0.1528
Epoch [8200/20000], Loss: 0.1505
Epoch [8300/20000], Loss: 0.1483
Epoch [8400/20000], Loss: 0.1461
Epoch [8500/20000], Loss: 0.1440
Epoch [8600/20000], Loss: 0.1420
Epoch [8700/20000], Loss: 0.1400
Epoch [8800/20000], Loss: 0.1381
Epoch [8900/20000], Loss: 0.1363
Epoch [9000/20000], Loss: 0.1345
Epoch [9100/20000], Loss: 0.1328
Epoch [9200/20000], Loss: 0.1312
Epoch [9300/20000], Loss: 0.1295
Epoch [9400/20000], Loss: 0.1280
Epoch [9500/20000], Loss: 0.1265
Epoch [9600/20000], Loss: 0.1250
Epoch [9700/20000], Loss: 0.1236
Epoch [9800/20000], Loss: 0.1222
Epoch [9900/20000], Loss: 0.1208
Epoch [10000/20000], Loss: 0.1195
Epoch [10100/20000], Loss: 0.1183
Epoch [10200/20000], Loss: 0.1170
Epoch [10300/20000], Loss: 0.1159
Epoch [10400/20000], Loss: 0.1147
Epoch [10500/20000], Loss: 0.1136
Epoch [10600/20000], Loss: 0.1125
Epoch [10700/20000], Loss: 0.1114
Epoch [10800/20000], Loss: 0.1104
Epoch [10900/20000], Loss: 0.1094
Epoch [11000/20000], Loss: 0.1084
Epoch [11100/20000], Loss: 0.1075
Epoch [11200/20000], Loss: 0.1065
Epoch [11300/20000], Loss: 0.1056
Epoch [11400/20000], Loss: 0.1048
Epoch [11500/20000], Loss: 0.1039
Epoch [11600/20000], Loss: 0.1031
Epoch [11700/20000], Loss: 0.1023
Epoch [11800/20000], Loss: 0.1015
Epoch [11900/20000], Loss: 0.1007
Epoch [12000/20000], Loss: 0.0999
Epoch [12100/20000], Loss: 0.0992
Epoch [12200/20000], Loss: 0.0985
Epoch [12300/20000], Loss: 0.0978
Epoch [12400/20000], Loss: 0.0971
Epoch [12500/20000], Loss: 0.0964
Epoch [12600/20000], Loss: 0.0958
Epoch [12700/20000], Loss: 0.0951
Epoch [12800/20000], Loss: 0.0945
Epoch [12900/20000], Loss: 0.0939
Epoch [13000/20000], Loss: 0.0933
Epoch [13100/20000], Loss: 0.0927
Epoch [13200/20000], Loss: 0.0922
Epoch [13300/20000], Loss: 0.0916
Epoch [13400/20000], Loss: 0.0910
Epoch [13500/20000], Loss: 0.0905
Epoch [13600/20000], Loss: 0.0900
Epoch [13700/20000], Loss: 0.0895
Epoch [13800/20000], Loss: 0.0890
Epoch [13900/20000], Loss: 0.0885
Epoch [14000/20000], Loss: 0.0880
Epoch [14100/20000], Loss: 0.0875
Epoch [14200/20000], Loss: 0.0871
Epoch [14300/20000], Loss: 0.0866
Epoch [14400/20000], Loss: 0.0862
Epoch [14500/20000], Loss: 0.0857
Epoch [14600/20000], Loss: 0.0853
Epoch [14700/20000], Loss: 0.0849
Epoch [14800/20000], Loss: 0.0845
Epoch [14900/20000], Loss: 0.0840
Epoch [15000/20000], Loss: 0.0837
Epoch [15100/20000], Loss: 0.0833
Epoch [15200/20000], Loss: 0.0829
Epoch [15300/20000], Loss: 0.0825
Epoch [15400/20000], Loss: 0.0821
Epoch [15500/20000], Loss: 0.0818
Epoch [15600/20000], Loss: 0.0814
Epoch [15700/20000], Loss: 0.0811
Epoch [15800/20000], Loss: 0.0807
Epoch [15900/20000], Loss: 0.0804
Epoch [16000/20000], Loss: 0.0800
Epoch [16100/20000], Loss: 0.0797
Epoch [16200/20000], Loss: 0.0794
Epoch [16300/20000], Loss: 0.0791
Epoch [16400/20000], Loss: 0.0788
Epoch [16500/20000], Loss: 0.0785
Epoch [16600/20000], Loss: 0.0782
Epoch [16700/20000], Loss: 0.0779
Epoch [16800/20000], Loss: 0.0776
Epoch [16900/20000], Loss: 0.0773
Epoch [17000/20000], Loss: 0.0770
Epoch [17100/20000], Loss: 0.0767
Epoch [17200/20000], Loss: 0.0765
Epoch [17300/20000], Loss: 0.0762
Epoch [17400/20000], Loss: 0.0759
Epoch [17500/20000], Loss: 0.0757
Epoch [17600/20000], Loss: 0.0754
Epoch [17700/20000], Loss: 0.0751
Epoch [17800/20000], Loss: 0.0749
Epoch [17900/20000], Loss: 0.0747
Epoch [18000/20000], Loss: 0.0744
Epoch [18100/20000], Loss: 0.0742
Epoch [18200/20000], Loss: 0.0739
Epoch [18300/20000], Loss: 0.0737
Epoch [18400/20000], Loss: 0.0735
Epoch [18500/20000], Loss: 0.0733
Epoch [18600/20000], Loss: 0.0730
Epoch [18700/20000], Loss: 0.0728
Epoch [18800/20000], Loss: 0.0726
Epoch [18900/20000], Loss: 0.0724
Epoch [19000/20000], Loss: 0.0722
Epoch [19100/20000], Loss: 0.0720
Epoch [19200/20000], Loss: 0.0718
Epoch [19300/20000], Loss: 0.0716
Epoch [19400/20000], Loss: 0.0714
Epoch [19500/20000], Loss: 0.0712
Epoch [19600/20000], Loss: 0.0710
Epoch [19700/20000], Loss: 0.0708
Epoch [19800/20000], Loss: 0.0706
Epoch [19900/20000], Loss: 0.0704
Epoch [20000/20000], Loss: 0.0702
PS E:\shucai\py>
Epoch [20000/20000], Loss: 0.0702
Epoch [20000/20000], Loss: 0.0702
Epoch [20000/20000], Loss: 0.0702
Epoch [20000/20000], Loss: 0.0702
PS E:\shucai\py>
Epoch [20000/20000], Loss: 0.0702
PS E:\shucai\py>
Epoch [20000/20000], Loss: 0.0702
Epoch [20000/20000], Loss: 0.0702
Epoch [20000/20000], Loss: 0.0702