R语言实现单隐藏层MLP循环训练代码

很高兴回答您的问题！以下是一个简单的 R 代码实现单隐藏层 MLP 循环训练的示例：

# 加载数据
data(iris)
x <- iris[, 1:4]
y <- iris[, 5]
# 将类别变量转换为数字
y <- as.numeric(y)
y <- y - 1
# 将输出变量转换为矩阵形式
y <- matrix(y, nrow = length(y), ncol = 1)
# 分割数据集
set.seed(123)
train_index <- sample(1:length(y), size = 0.7 * length(y))
test_index <- setdiff(1:length(y), train_index)
train_x <- x[train_index, ]
train_y <- y[train_index, ]
test_x <- x[test_index, ]
test_y <- y[test_index, ]
# 定义模型参数
input_size <- ncol(train_x)
hidden_size <- 10
output_size <- ncol(train_y)
learning_rate <- 0.01
epochs <- 100
# 初始化权重和偏置
w1 <- matrix(rnorm(input_size * hidden_size), nrow = input_size, ncol = hidden_size)
b1 <- matrix(rnorm(hidden_size), nrow = 1, ncol = hidden_size)
w2 <- matrix(rnorm(hidden_size * output_size), nrow = hidden_size, ncol = output_size)
b2 <- matrix(rnorm(output_size), nrow = 1, ncol = output_size)
# 定义 sigmoid 函数
sigmoid <- function(x) {
  1 / (1 + exp(-x))
}
# 训练模型
for (i in 1:epochs) {
  # 前向传播
  hidden_layer <- sigmoid(train_x %*% w1 + b1)
  output_layer <- sigmoid(hidden_layer %*% w2 + b2)
  # 计算损失函数
  loss <- sum((train_y - output_layer) ^ 2)
  # 反向传播
  output_error <- (train_y - output_layer) * output_layer * (1 - output_layer)
  hidden_error <- output_error %*% t(w2) * hidden_layer * (1 - hidden_layer)
  # 更新权重和偏置
  w2 <- w2 + t(hidden_layer) %*% output_error * learning_rate
  b2 <- b2 + colSums(output_error) * learning_rate
  w1 <- w1 + t(train_x) %*% hidden_error * learning_rate
  b1 <- b1 + colSums(hidden_error) * learning_rate
}
# 测试模型
test_hidden_layer <- sigmoid(test_x %*% w1 + b1)
test_output_layer <- sigmoid(test_hidden_layer %*% w2 + b2)
test_prediction <- ifelse(test_output_layer > 0.5, 1, 0)
accuracy <- sum(test_prediction == test_y) / length(test_y)
print(paste0("测试集准确率为：", accuracy))

希望这个代码示例能够帮助您！如果您有任何其他问题，欢迎随时问我。