I am experimenting with various loss functions. I tried mse_loss and L1_loss and the val_loss is around 10K. HAs anybody tried a different network or a loss function that has given a lower val loss?
I made my own loss function and that was nothing but the square root over mse_loss (and that made it rmse_loss, ). Doing this reduced the large value of val_loss.
I haven’t tried L1_loss but I’ll check it out.
Thanks @ayush-mishra-ald for replying. I did try RMSE as well. Not much traction for my customized set.
I tried SmoothHuberloss in pytorch. Seems much better. LEt me know how it works for you!