Statistical Test for Generated Hypotheses by Diffusion Model

This pacakge is the implementation of the paper "Statistical Test for Generated Hypotheses by Diffusion Model" for experiments.

Installation & Requirements

This pacakage has the following dependencies:

• Python (version 3.10 or higher, we use 3.10.13)
• sicore (we use version 1.0.0)
• tensorflow (we use version 2.15.0)
• scipy (we use version 1.11.4)
• numpy (we use version 1.26.2)
• omegaconf (we use version 2.3.0)
• tqdm

Please install these dependencies by pip.

pip install sicore
pip install tensorflow
pip install scipy
pip install numpy
pip install omegaconf
pip install tqdm
pip install matplotlib

Reproducibility

Since we have already got the results in advance, you can reproduce the figures by running following code by running plot.ipynb. The results will be saved in "/results/fig" folder.

To reproduce the results, please see the following instructions after installation step. The results will be saved in "./results" folder as pickle file.

For indipendence (type I error rate experiment). Before execute the command, please set data/category to iid in config.yml

sh fpr1.sh
sh fpr2.sh
sh fpr3.sh
sh fpr4.sh

For indipendence (power experiment). Before execute the command, please set data/category to iid in config.yml

sh power1.sh
sh power2.sh
sh power3.sh
sh power4.sh

For correlation (type I error rate experiment). Before execute the command, please set data/category to corr in config.yml

sh fpr1.sh
sh fpr2.sh
sh fpr3.sh
sh fpr4.sh

For correlation (power experiment). Before execute the command, please set data/category to corr in config.yml

sh power1.sh
sh power2.sh
sh power3.sh
sh power4.sh