Multiple features input in model

I have been recently working with tensorflow and neural works in general, and had a few concept questions.

Lets say I have a table where for every customer, there are X products columns, X associated dates for those products, X scores for each product, etc. From my understanding, this means that we have three features.

So customer 1 has [[A, 2022, 0.55], [B, 2021, 0.11], [B, 2022, 0.82]].

If I have a sample of 1m customers, how will that be used as an input? I have to first pad the everything to be equal size, so customer 1 will have the following sequence if we pad to 5 for example: [[A, 2022, 0.55], [B, 2021, 0.11], [B, 2022, 0.82], [0, 0, 0], [0, 0, 0]]

I’m using LSTM, so I have to reshape the input into [batch, timesteps, features] where batch is the sample size, timesteps is 5 (following the above example), and features is 3. Is that correct?

Also, if I encode the products into integers, it should be fine as such, without having to one-hot encode it, right?

One last thing; what is the difference in having the sequences as different features against having every feature as a separate input?

Hello @Iden_Crisler

Thank you for using TensorFlow
In the above problem statement, the input shape would be [batch_size, timesteps, features], in this case [1m, 5, 3],
We can encode the products into integers if the relationship between products is not critical, and need not to be learned by model, one-hot encoding can be used if the model needs to understand that products are distinct categories of themselves,
In sequences as different features preserve order of product interactions for each customer, by reading patterns overtime,
When having every feature as separate input we can only analyze non-sequential relationship between features, i.e we lose the sequential modeling capability that LSTMs provide.