125 lines
3.8 KiB
Python
125 lines
3.8 KiB
Python
|
import torch
|
|||
|
|
import torch.nn as nn
|
|||
|
|
import numpy as np
|
|||
|
|
|
|||
|
|
#这个实验的目的是比较RNN和GRU在相同任务上的性能，即学习序列中两个随机位置数值的和。
|
|||
|
|
# 数据生成
|
|||
|
|
def generate_add_data(seq_len=30):
|
|||
|
|
data = torch.zeros(seq_len, 2) # (seq_len, 2)
|
|||
|
|
idx1, idx2 = np.random.choice(seq_len, 2, replace=False)
|
|||
|
|
val1, val2 = np.random.rand()*0.5, np.random.rand()*0.5
|
|||
|
|
data[idx1, 0] = val1
|
|||
|
|
data[idx2, 0] = val2
|
|||
|
|
target = torch.tensor([val1 + val2]).view(1,1)
|
|||
|
|
return data.unsqueeze(0), target # (1, seq_len, 2)
|
|||
|
|
|
|||
|
|
# 模型定义
|
|||
|
|
class AdditionRNN(nn.Module):
|
|||
|
|
def __init__(self):
|
|||
|
|
super().__init__()
|
|||
|
|
self.rnn = nn.RNN(2, 16, batch_first=True)
|
|||
|
|
self.fc = nn.Linear(16, 1)
|
|||
|
|
|
|||
|
|
def forward(self, x):
|
|||
|
|
out, _ = self.rnn(x)
|
|||
|
|
return self.fc(out[:, -1, :])
|
|||
|
|
|
|||
|
|
class AdditionGRU(nn.Module):
|
|||
|
|
def __init__(self):
|
|||
|
|
super().__init__()
|
|||
|
|
self.gru = nn.GRU(2, 16, batch_first=True)
|
|||
|
|
self.fc = nn.Linear(16, 1)
|
|||
|
|
|
|||
|
|
def forward(self, x):
|
|||
|
|
out, _ = self.gru(x)
|
|||
|
|
return self.fc(out[:, -1, :])
|
|||
|
|
|
|||
|
|
# 训练对比
|
|||
|
|
import matplotlib.pyplot as plt
|
|||
|
|
|
|||
|
|
# 修改后的训练函数，记录损失变化
|
|||
|
|
def train_addition():
|
|||
|
|
rnn = AdditionRNN()
|
|||
|
|
gru = AdditionGRU()
|
|||
|
|
criterion = nn.MSELoss()
|
|||
|
|
optim_rnn = torch.optim.Adam(rnn.parameters(), lr=0.01)
|
|||
|
|
optim_gru = torch.optim.Adam(gru.parameters(), lr=0.01)
|
|||
|
|
|
|||
|
|
# 记录训练过程
|
|||
|
|
losses = {'RNN': [], 'GRU': []}
|
|||
|
|
|
|||
|
|
for step in range(1000):
|
|||
|
|
inputs, target = generate_add_data(seq_len=30)
|
|||
|
|
|
|||
|
|
# RNN训练
|
|||
|
|
optim_rnn.zero_grad()
|
|||
|
|
rnn_pred = rnn(inputs)
|
|||
|
|
loss_rnn = criterion(rnn_pred, target)
|
|||
|
|
loss_rnn.backward()
|
|||
|
|
optim_rnn.step()
|
|||
|
|
|
|||
|
|
# GRU训练
|
|||
|
|
optim_gru.zero_grad()
|
|||
|
|
gru_pred = gru(inputs)
|
|||
|
|
loss_gru = criterion(gru_pred, target)
|
|||
|
|
loss_gru.backward()
|
|||
|
|
optim_gru.step()
|
|||
|
|
|
|||
|
|
# 记录损失
|
|||
|
|
losses['RNN'].append(loss_rnn.item())
|
|||
|
|
losses['GRU'].append(loss_gru.item())
|
|||
|
|
|
|||
|
|
if step % 200 == 0:
|
|||
|
|
print(f"Step {step:03d} | RNN Loss: {loss_rnn.item():.4f} | GRU Loss: {loss_gru.item():.4f}")
|
|||
|
|
|
|||
|
|
# 绘制损失曲线
|
|||
|
|
plt.figure(figsize=(10, 5))
|
|||
|
|
plt.plot(losses['RNN'], label='RNN', alpha=0.7)
|
|||
|
|
plt.plot(losses['GRU'], label='GRU', alpha=0.7)
|
|||
|
|
plt.xlabel('Training Steps')
|
|||
|
|
plt.ylabel('MSE Loss')
|
|||
|
|
plt.title('Training Comparison: RNN vs GRU')
|
|||
|
|
plt.legend()
|
|||
|
|
plt.grid(True)
|
|||
|
|
plt.show()
|
|||
|
|
|
|||
|
|
return rnn, gru
|
|||
|
|
|
|||
|
|
# 执行训练
|
|||
|
|
rnn_model, gru_model = train_addition()
|
|||
|
|
def show_test_cases(model, model_name, num_cases=5):
|
|||
|
|
print(f"\n{model_name} 测试样例:")
|
|||
|
|
criterion = nn.MSELoss()
|
|||
|
|
total_error = 0
|
|||
|
|
|
|||
|
|
for case_idx in range(num_cases):
|
|||
|
|
# 生成测试数据
|
|||
|
|
inputs, target = generate_add_data()
|
|||
|
|
seq_len = inputs.shape[1]
|
|||
|
|
|
|||
|
|
# 解析输入数据
|
|||
|
|
non_zero_indices = torch.nonzero(inputs[0, :, 0])
|
|||
|
|
pos1, pos2 = non_zero_indices[0].item(), non_zero_indices[1].item()
|
|||
|
|
val1 = inputs[0, pos1, 0].item()
|
|||
|
|
val2 = inputs[0, pos2, 0].item()
|
|||
|
|
|
|||
|
|
# 模型预测
|
|||
|
|
with torch.no_grad():
|
|||
|
|
pred = model(inputs)
|
|||
|
|
loss = criterion(pred, target)
|
|||
|
|
|
|||
|
|
# 格式输出
|
|||
|
|
print(f"案例 {case_idx+1}:")
|
|||
|
|
print(f"输入序列长度: {seq_len}")
|
|||
|
|
print(f"数值位置: [{pos1:2d}]={val1:.4f}, [{pos2:2d}]={val2:.4f}")
|
|||
|
|
print(f"真实值: {target.item():.4f}")
|
|||
|
|
print(f"预测值: {pred.item():.4f}")
|
|||
|
|
print(f"绝对误差: {abs(pred.item()-target.item()):.4f}")
|
|||
|
|
print("-" * 40)
|
|||
|
|
total_error += abs(pred.item()-target.item())
|
|||
|
|
|
|||
|
|
print(f"平均绝对误差: {total_error/num_cases:.4f}\n")
|
|||
|
|
|
|||
|
|
show_test_cases(rnn_model, "RNN")
|
|||
|
|
show_test_cases(gru_model, "GRU")
|