How can I improve the val_acc ?

[liuguochao]

I did this experiment with your code, but the validation accuracy is only 0.74, how can I improve the val_acc ?

[nikhiltitus]

Were you able to get 74% with the code. The authors mentioned 67% here: http://deepsound.io/music_genre_recognition.html

[nikhiltitus]

I have a model with higher accuracy implemented as per: http://benanne.github.io/2014/08/05/spotify-cnns.html

[liuguochao]

Yes, I replaced RMSprop with Adam, and 0.78 is get.

[nikhiltitus]

I am able to get around 83% with my model. I will be uploading this soon.

[nikhiltitus]

Frankly speaking I think the data set is too small.

[nikhiltitus]

You can check the code here: https://github.com/nikhiltitus/genre-recognition There is a parameter in train_model.py called model choice. You will have to input 2 for the second model.
ie:

python train_model.py -c 2