Why R session aborted when performing 1D-CNN? - r

Question

I'm trying to perform 1D-CNN on a small dataset, which has only 3000+ samples. Lookback parameter is set to 7 for the generator. R session aborted after few seconds of training.
However, when the lookback parameter is set to any value other than 7 (such as 5, 9 or 10), the 1D-CNN model can work! I also try GRU, LSTM, and fully-connected layers with the lookback parameter set to 7, everything works fine.
My environment are as fallows: R 3.6.1, Rstudio 1.5.5001, Tensorflow 2.0.0, Keras 2.2.5.0.
The codes is shown below:
lookback <- 7
step <- 1
delay <- 1
batch_size <- 32
nbfeature <- 31
generator_WeeklyTotal <- function(data, lookback, delay, min_index, max_index,
shuffle = FALSE, batch_size, step = 1) {
if (is.null(max_index)) max_index <- nrow(data) - delay - 1
i <- min_index + lookback
function() {
if (shuffle) {
rows <- sample(c((min_index+lookback):max_index), size = batch_size)
} else {
if (i + batch_size >= max_index)
i <<- min_index + lookback
rows <- c(i:min(i+batch_size, max_index))
i <<- i + length(rows)
}
samples <- array(0, dim = c(length(rows),
lookback / step,
nbfeature))
targets <- array(0, dim = c(length(rows)))
for (j in 1:length(rows)) {
indices <- seq(rows[[j]] - lookback, rows[[j]],
length.out = dim(samples)[[2]])
samples[j,,] <- data[indices,60:90]
targets[[j]] <- data[rows[[j]] + delay,6]
}
list(samples, targets)
}
}
train_gen <- generator_WeeklyTotal(
data = DLtrain,
lookback = lookback,
delay = delay,
min_index = 1+lookback,
max_index = 2922+lookback,
shuffle = FALSE,
step = step,
batch_size = batch_size
)
val_gen = generator_WeeklyTotal(
data = DLtrain,
lookback = lookback,
delay = delay,
min_index = 2923+lookback,
max_index = 3287+lookback,
step = step,
batch_size = batch_size
)
test_gen <- generator_WeeklyTotal(
data = DLtrain,
lookback = lookback,
delay = delay,
min_index = 3288+lookback,
max_index = 3652+lookback,
step = step,
batch_size = batch_size
)
val_steps <- (3287+lookback - 2923+lookback - lookback) / batch_size
test_steps <- (3652+lookback - 3288+lookback - lookback) / batch_size
model_conv1d <- keras_model_sequential() %>%
layer_conv_1d(filters = 64, kernel_size = 5, activation = "relu", padding = "same",
input_shape = list(NULL, nbfeature)) %>%
layer_max_pooling_1d(pool_size = 3) %>%
layer_conv_1d(filters = 64, kernel_size = 5, activation = "relu",padding = "same") %>%
layer_max_pooling_1d(pool_size = 3) %>%
layer_conv_1d(filters = 64, kernel_size = 5, activation = "relu",padding = "same") %>%
layer_global_max_pooling_1d() %>%
layer_dense(units = 1)]
I wonder why R session aborted when I set the lookback parameter to be 7 but works with other value? Is there any way to solve this problem because it is necessary to set lookback to be 7 for the purpose of predicting by using weekly data.