I re-trained the model on patterns that included +/- 1 degree peak shifts. However, this led to ~8% reduction in accuracy on the validation and test sets (simulated data). It also did not solve the variation in predictions on the experimental patterns with large peak shifts.
I re-trained the model on patterns that included +/- 1 degree peak shifts. However, this led to ~8% reduction in accuracy on the validation and test sets (simulated data). It also did not solve the variation in predictions on the experimental patterns with large peak shifts.