How to reshape a time-series df for multivariate LSTM in R?

Question

I have a df in R with 6000 rows and 90 variables. I want to predict the sales volume of product A for the next 12 months based on data about product A as well as competing products (B,C,D,...). To use an LSTM, I need to reshape the df into 3D format (samples, timesteps, features) but I don't quite get how.

My df looks somewhat like this:

Date	Product	Sales	X1	X2	X3	...	X87
2017-01-01	A	0.65438	0.45438	-1.2670	0.3215	...	1.35623
2017-01-01	B	-0.55468	0.12436	-1.5677	-0.3215	...	1.35623
2017-01-01	C	0.65981	1.12345	-0.5574	0.3215	...	1.35623
2017-02-01	A	-0.12338	-1.12345	0.4543	-1.5673	...	0.42961
--------	--------	--------	---------	--------	---------	------	---------
2022-12-01	C	0.34568	1.134598	0.5678	-1.2648	...	0.34675

So far, I have split the data into train and test set and normalized. Then I ran:

# Reshape to 3-dimensional array
train_data_lstm <- train_data %>%
  as.matrix() %>% 
  array(dim = c(nrow(train_data), ncol(train_data), 1))

test_data_lstm <- test_data %>%
  as.matrix() %>% 
  array(dim = c(nrow(test_data), ncol(test_data), 1))

# Prepare sequences
lookback <- 12  # how many steps back the model should look

1. How do I continue from here?
2. Do I need to remove the Date variable? What about the product labels?

Thank you in advance!