mxtsai / elnet Goto Github PK

when performing self.feature_dp, it assigns according to self.feature_dropout method which never got initialized. There seems to be code missing or inheriting from a parent class that is not shared in this repo.

Thank you for your work. I tried to reproduce the results, but the results were not good. I only did simple data enhancement during training, and did not make the number of positive and negative samples consistent. What I want to know is whether data enhancement has a big impact on the experimental results. My training code mainly refers to this link.. Thanks in advance.

Thanks for your code, it really helps me a lot. In paper Making Convolutional Networks Shift-Invariant Again, maxpool is divided into two steps: densely max-pooled and blurpool. However, in your implementation of blurpool, I find that the first one(max-pooled) is left out. I wonder if this implementation can improve the accuracy and why.