Hey, thanks!

1. No particular reason for the time steps. I just had to pick a number that captures enough history from the data.
2. Let’s say we have daily Time Series data (1 data point for each day). If we select 24 as a number of time steps — we’ll use the previous 24 days to predict today, we’ll use the previous 23 days and today to predict tomorrow and so on.

I’ll suggest you dive a bit deeper into the/your dataset right before passing it as an input to the LSTM. Explore the shapes and take a couple of sequences to figure out what is happening.

I hope that helps!