Passing data to the keras model.fit() function more effficiently

I am training an LSTM wich on a time series. The input sequence has length 480, and I create my trainingsdata from a bigger timeseries array, where all the values are in the right order. I take blocks of length 480, starting at every possible position in the array and fill the container that will be passed to the model.fit() function with these blocks. But this is of course very inefficient as I need huge amounts of memory to train my model as the first sample contains values [0, 1, 2, ... , 479] and the second one values [1, 2, 3, ... , 480] and so on... When training im not able to fit all this (redundant) data into my RAM at once. Is there any way to solve this by using e.g. some kind of array that references the corresponding parts in the bigger timeseries array? Thanks for any help!

from Recent Questions - Stack Overflow https://ift.tt/3r03u78
https://ift.tt/eA8V8J