Note that if certain files cannot be found, you may need to manually add the paths in your script using:

import sys
sys.path.append("file_absolute_path")

This document provides instructions on how to replicate our "Attention Accuracy" analysis from Section 4.3 of our paper.

Run the command in ./script/mi_vs_learned.sh and you will get figures like that:

Run the command in ./script/gsu_performance.sh to analyze the retrieval performance. Then run the following command for visualization:

python gsu_performance_draw.py

This will generate a figure like:

gsu_performance_draw.py records the learned correlation scores for each test sample. To perform a case study, you can run the following command:

python case_study.py

This will generate figures like: