Quan Tran

A bug in generating lag features

1

There is a bug in week 4 programming assignment notebook where you generate lag features 'After creating a grid, we can calculate some features. We will use lags from [1,...

https://github.com/anhquan0412/basic_model_scratch/blob/fd83305ae69460cf7f2ea0f4c2bbbf633eed652c/model/neural_network.py#L77 inp is (wx + b), activation is act(inp) or act(wx+b)

https://github.com/anhquan0412/basic_model_scratch/blob/fd83305ae69460cf7f2ea0f4c2bbbf633eed652c/model/neural_network.py#L82

Gradient unable to backprop if we use argmax or torch.where

3

Hi, Correct me if I am wrong but in the code snippet to calculate D_l for the dependency loss (https://github.com/Ugenteraan/Deep_Hierarchical_Classification/blob/e4f20ae51a2daabfc1c01f6fdab778ef31cc7617/model/hierarchical_loss.py#L65), argmax is non-differentiable, thus the gradient wrt to dloss won't...