李梦圆
Results
1
comments of
李梦圆
I modified the loss function, and added initialization of the last layer in the model. the best valid diff is 3.65 with rankl2loss and 3.84 with rankl1loss after training 80...