Skip to content

Latest commit

 

History

History
20 lines (16 loc) · 761 Bytes

README.md

File metadata and controls

20 lines (16 loc) · 761 Bytes

balanced-TCP

This project is an experimental implementation of the paper "Balance Calibration for Identifying Incorrect Classifications with Distinguishable Uncertainty" (under review in AISTATS 2022).

avatar

dataset extension

Override new subclass of torch.utils.data.Dataset
Implement init(), getitem() and len()
Save your code into balanced-TCP/helpers/MyDataset.py

network extension

Setup a new .py file into balanced-TCP/models/
Inherit torch.nn.Module
Implement init() and forward()
Follow the examples given in balanced-TCP/models/

experiment extension

Setup a new .yaml file into balanced-TCP/conf/
Follow the examples given in balanced-TCP/conf/
Revise balanced-TCP/RUNME.sh for detailed settings