Fell free to add more content my friend. This is an open community, your contributions will be more than welcome ;)

For sure, always after digging into the Deep Learning, you must perform as much as Exploratory Data Analysis as you can. You need to know very well how your data look like because, perhaps, you won't need a neurla network to solve your problem. For time-series you have ARIMA, even Bayesian Filters and other well-knonw techniques. Other considerations, is your data labeled, I mean, do you know the threshold?

The sampling period is something you have to come up with... there is no a recipe, every problem is different, You might do some research in the literature. The LSTM can accept an input with the following shape: (batch_size, num_samples, features)