Hello,
I am trying to use your code for implementing a project dealing with sensor data that has repeated step functions with roughly fixed length. Your code has greatly save my implementation time. However, using your code, we got high anomaly scores at the beginning or the end of every step functions, due to the difficulty to learn and predict abrupt jumps or downs. I am wondering whether batch size and sequence length will influence the prediction quality? Or do you have suggestion for improving the prediction quality for the step-like time-series data? Thanks.
Hello, I am trying to use your code for implementing a project dealing with sensor data that has repeated step functions with roughly fixed length. Your code has greatly save my implementation time. However, using your code, we got high anomaly scores at the beginning or the end of every step functions, due to the difficulty to learn and predict abrupt jumps or downs. I am wondering whether batch size and sequence length will influence the prediction quality? Or do you have suggestion for improving the prediction quality for the step-like time-series data? Thanks.