R Keras-具有拟合误差的超简单LSTM示例

Question

我正在尝试构建一个玩具模型，以演示LSTM如何预测序列的下几次迭代。我的代码运行到最后一行没有任何错误。

# Simulating dummy data
seq <- data.frame(x_train = (seq(0,8, 1)/10), y_train = (seq(0,8, 1)/10))
seq$x_train <- seq$x_train + 0.1

# Reshaping
x_train <- array_reshape(seq$x_train, dim = c(9,1,1))
y_train <- array_reshape(seq$y_train, dim = c(9,1))

# Checking dimensions
dim(x_train); dim(y_train)

# Building the model
m <- keras_model_sequential()
m %>% 
  layer_lstm(units = 10, input_shape =c(9,1), batch_size = 9, return_sequences = T, stateful = T) %>%
  layer_dense(units = 1)

summary(m)

# Compiling
m %>% compile(loss = "mse", optimizer = "adam")

这就是问题所在-

for (i in 1:9) {
  m %>% fit(object = x_train, y_train, batch_size = 1, shuffle = FALSE)
  m %>% reset_states()
}

我收到以下错误，但我不确定为什么：

Error: $ operator is invalid for atomic vectors

有人知道我在做什么错吗？