This is the repo for a time-series classification model based on the transformer architecture. It uses a number of different encoder blocks to perform classification on a timeseries dataset. Linear attention works but windowed attention is not properly implemented.