Skip to content

Latest commit

 

History

History
37 lines (26 loc) · 1.28 KB

TITAN.md

File metadata and controls

37 lines (26 loc) · 1.28 KB

Running TITAN

We provide a config file and a pretrained model to reproduce the experiment on TITAN reported in our paper.

Steps to reproduce the experiment

  1. Be sure you installed TITAN's dependencies
pip install extern/titan
pip install extern/pytoda
  1. Download the trained model at https://ibm.box.com/shared/static/g5gfmghuwpg8osr4mljk8zp8uvar9o9e.pt
  2. To use our config file out-of-the-box, place the file at the path
<PATH_TO_REPO>/data/titan_model/smi_full_fold1_published_code/weights/best_ROC-AUC_bimodal_mca.pt

  1. In the provided config file at <PATH_TO_REPO>/sample_configs/titan.json, substitute all the occurences of <PATH_TO_REPO> to the actual path where you cloned this repository.

  2. In the provided config file at <PATH_TO_REPO>/sample_configs/titan.json, change the root entry to a folder where you want to store the results.

  3. (optional) If you installed DECODE in a conda environment, activate the environment, e.g. (if your environment is called decode)

conda activate decode
  1. Run DECODE
run_interpretability_pipeline --config <PATH_TO_REPO>/sample_configs/titan.json --user_directory <PATH_TO_REPO>/example/titan