Skip to content

Latest commit

 

History

History
34 lines (30 loc) · 2.6 KB

README.md

File metadata and controls

34 lines (30 loc) · 2.6 KB
Raw

HIConcept

This is the code repository for paper 'Explaining Language Models’ Predictions with High-Impact Concepts'

The motivation is built on the following causal graph: causal_graph

Running instructions: inside src/, use the following running scripts:

  1. For text experiments:
    • baselines: run_text_baselines.sh
    • CausalConcept with ablation studies: run_text_causal.sh
  2. For toy experiments: run_toy.sh
  3. For human evaluation: human_eval.sh
  4. For hyperparameter analysis with number of concepts: num_of_concepts.sh

File structure:

  1. src/: source code
    • text_main.py: main running scripts that loads data, classification model, and topic model with evaluation.
    • conceptshp.py: our implementation of concept models, we looked at original conceptSHAP repo: https://github.com/chihkuanyeh/concept_exp; and Berkeley's version: https://github.com/arnav-gudibande/conceptSHAP as references.
    • helper.py: our implementation of self-constructed CNN-based transformer classification models
    • toy_helper.py: helper functions for toy dataset, such as creating images, datasets, etc.
    • text_helper_v2.py: helper functions for text experiments, such as loading datasets, loading text classification models, etc.
    • visualize.py: functions to visualize topics, including gradcam and BERT visualization
    • bcvae.py & elbo_decomposition.py & lib/: implementation of Beta-TCVAE models from https://github.com/rtqichen/beta-tcvae
    • BERT_explainability/ & BERT_rationale_benchmark: folder that includes helper functions to achieve BERT transformer visualization, taken from https://github.com/hila-chefer/Transformer-Explainability
    • code used to analyze the results
      • analyze_csv.py: script used to analyze human evaluation results, including agreement calculations, etc.
      • wordcloud.ipynb: script used to generate wordcloud images in the appendix.
      • produce_html_examples.py: generates human evaluation examples
  2. src/models/: directory for storing classification models and topic models, after training, cls model will be stored in directory: models/{DATASET}/{CLS_MODEL}/cls_model.pkl, topic model will be stored in directory:models/{DATASET}/{CLS_MODEL}/{TOPIC_MODEL}/topic_model_{TOPIC_MODEL}.pkl. Both will be accompanied by a training graph with loss and accuracy.
  3. src/human_eval_examples/: directory to save generated human evaluation examples.
  4. environment.yml: file to create the conda environment. Run ' conda env create -f environment.yml'
  5. t5_env.yml: please use this environment to fine-tune t5.