Hi, Riniki,

In the post, Joonatan Samuel described very well. In general, samples are the number of chars/words/sentences; Time steps are the length of each sample (in this case each letter can be a time step); and Features are an amount of numbers that represent each time step (you define how many you want).

Taking your samples in one array, we can exemplify:

[‘a’, ‘home’, ‘How are you?’]

So, we have to fill the items size, that everyone has the same length (to matrix calculations):

[‘a___________’, ‘home________’, ‘How are you?’]

Now, the shape (samples, time_steps, features) is: (3, 12, X).
Why ‘X’? Because, you define how many features you want to represent each char (32, 64, 128, 512…)

This is a general view, but I hope the explanation has become clear.