this depends also on the loop. e.g. if you have `for i in range(60, 800):
X_train.append(training_set_scaled[i-60:i, 0])` then the first iteration uses the data 0:60 and the second the data 1:61. So, in that case the step is indeed 1 (time lag).