用numpy实现RNN

import numpy as np

# 定义RNN的参数
input_size = 10  # 输入大小
hidden_size = 20  # 隐藏层大小
output_size = 5  # 输出大小

# 定义RNN的权重
Wxh = np.random.randn(hidden_size, input_size) * 0.01  # 输入到隐藏层的权重
Whh = np.random.randn(hidden_size, hidden_size) * 0.01  # 隐藏层到隐藏层的权重
Why = np.random.randn(output_size, hidden_size) * 0.01  # 隐藏层到输出层的权重
bh = np.zeros((hidden_size, 1))  # 隐藏层的偏置
by = np.zeros((output_size, 1))  # 输出层的偏置

# 定义RNN的前向传播函数
def rnn_forward(inputs):
    h = np.zeros((hidden_size, 1))  # 初始化隐藏层状态
    outputs = []  # 存储每个时间步的输出
    
    for t in range(len(inputs)):
        x = inputs[t]  # 当前时间步的输入
        h = np.tanh(np.dot(Wxh, x) + np.dot(Whh, h) + bh)  # 计算隐藏层状态
        y = np.dot(Why, h) + by  # 计算输出层
        
        outputs.append(y)  # 存储当前时间步的输出
    
    return outputs, h

# 测试RNN的前向传播函数
inputs = [np.random.randn(input_size, 1) for _ in range(5)]  # 生成5个随机输入
outputs, _ = rnn_forward(inputs)
print(outputs)