problem with Gaussian kernel

amalF / Kervolution

Kervolution implementation using TF2.0

20 stars 2 forks source link

problem with Gaussian kernel #2

Closed markqzw closed 4 years ago

markqzw commented 4 years ago

thanks for your code! but when I use the Gaussian kernel, the loss curve is always the same. do not change. can I ask why? I used my own dataset and the other kernel works well.

amalF commented 4 years ago

Hello markqzw,

The training loss of the Gaussian kernel on the MNIST dataset is decreasing (see image below). Are you able to reproduce the same result as me ?

markqzw commented 4 years ago

hello amalF, thank you very much for your reply. I use my own dataset, the input shape is (5,20,135,91) and the labels is (5,20,135,3) a regression problem. I set 100 epochs. the loss is always the same.

amalF commented 4 years ago

Hello, sorry for the delay It's going to be hard for me to debug if you are using your own dataset. Did you try different hyperparemeters ?

amalF commented 4 years ago

I will close this issue for now. please feel free to reopen thanks