Accuracy Goes Down And Loss Increases After 19th Epoch

[xuzhang5788]

After 19th epoch, I got loss: 0.3925 - acc: 0.8709 - val_loss: 0.3650 - val_acc: 0.8786

but after 20th epoch, loss increases a lot loss: 0.8816 - acc: 0.7350 - val_loss: 0.7405 - val_acc: 0.7701

then at 21th epoch loss: 0.7743 - acc: 0.7613 - val_loss: 0.6743 - val_acc: 0.7912

Is it normal? Did you change the learning rate after 20 epochs? What is the final loss and accuracy after 100th epoch when you trained your model?

After 100 epochs training, I got loss: 0.5779 - acc: 0.8223 - val_loss: 0.5234 - val_acc: 0.8355

Is it acceptable or I need to train more epochs?

Thanks

[yangxiufengsia]

Hi @xuzhang5788, If you want to obtain fast convergence, please reduce the hidden-dimensions of GRU(e.g. reduce 512 to 256).