search

juntang-zhuang / explain_invertible

repo for "Decision explanation and feature importance for invertible networks"
13 stars 6 forks source link

ABIDE data #2

Open shengaoya opened 2 years ago

shengaoya commented 2 years ago

hi， Very interesting work, but there are some problems when I reproduce, I don't know where to put ABIDE data, could you please tell me?

juntang-zhuang commented 2 years ago

This repository does not include dataloader files for ABIDE data, you need to write your own. For example, preprocess ABIDE and calculate connectome in nilearn, save data in file (e.g. npz format), then write a dataloader in PyTorch and replace the toy dataloader. Remember to flatten the connectome to 1D and change input dimension accordingly in network definition.

15265792229 commented 5 months ago

Hello, I saw your paper Invisible Network for Classification and Biomarker Selection for ASD on PaperWithcode. Following the data processing method in your paper, I have generated a matrix in 1d form of ABIDE cc200 (as mentioned in your paper). However, I am unsure of the next steps for training and which parameters need to be changed in order to use your reversible network for training