BCIAccuracy lower than that in the paper #43
Description
I first run /preprocessing/BCIIV2a.m which generate A01E.mat and A01T.mat,then I run conformer.py in 2000 epochs,this is result:
.... Epoch: 1998 Train loss: 0.066274 Test loss: 1.835294 Train accuracy 0.979167 Test accuracy is 0.791667 Epoch: 1999 Train loss: 0.001446 Test loss: 2.097695 Train accuracy 1.000000 Test accuracy is 0.770833 The average accuracy is: 0.7983940972222211 The best accuracy is: 0.8923611111111112 THE BEST ACCURACY IS 0.8923611111111112 subject 1 duration: 0:12:13.758384 Fri Oct 18 01:19:30 2024
and in paper the accurary is 88.19,but I only get 79.83%,I am wondering why is this happening and did i miss some peperprocess step?thanks so much.
The classification performance of each subject and the average results on Dataset I are presented in Table II. Conformer S01 88.19