MLP 多层感知机

import torch
import torch.nn as nn

# 定义两层 MLP
class SimpleMLP(nn.Module):
    def __init__(self):
        super().__init__()
        self.layer1 = nn.Linear(2, 3)   # 输入2维 → 隐藏3维
        self.relu = nn.ReLU()
        self.layer2 = nn.Linear(3, 2)   # 隐藏3维 → 输出2维

    def forward(self, x):
        h = self.relu(self.layer1(x))     # 隐藏层 + ReLU
        y = self.layer2(h)                 # 输出层
        return y

# 创建模型并手动设置权重
model = SimpleMLP()

# 设置 W₁, b₁
model.layer1.weight.data = torch.tensor([
    [0.5, 0.2],    # → h₁
    [-0.3, 0.7],   # → h₂
    [0.8, -0.1]    # → h₃
])
model.layer1.bias.data = torch.tensor([0.1, -0.2, 0.05])

# 设置 W₂, b₂
model.layer2.weight.data = torch.tensor([
    [0.4, -0.1, 0.3],   # → o₁
    [-0.2, 0.6, 0.5]    # → o₂
])
model.layer2.bias.data = torch.tensor([0.1, -0.05])

# 前向传播
x = torch.tensor([1.0, 0.5])
output = model(x)
probs = torch.softmax(output, dim=0)

print(f"输出: {output}")
print(f"概率: {probs}")
# 输出: tensor([0.6200, 0.2100])
# 概率: tensor([0.6011, 0.3989])