Open LouisEchard opened 6 years ago
Time to improve the architecture
resnet part TODO
regularization with L1 and L2 norms are not good for forecasting extreme events. Our event follows a poison/expo distribution. Drop normalization.
Time to improve the architecture