explainable-nlp-robustness

Requirements

This project uses Python 3.8

Setup Steps

1. Clone this repository

    git clone https://github.com/ShriyaA/explainable-nlp-robustness.git
   

2. Create a virtual environment and activate it (Recommended)

    cd explainable-nlp-robustness
    python -m venv venv
    source venv/bin/activate
   

3. Install required packages

    pip install -r requirements.txt
   

4. Create output folder

    mkdir output
   

Run with default options

python src/__main__.py visualize input.txt

Generate attacks

Word Deletion:

python src/__main__.py generate data/glue-sst2-validation.csv --attack_type word_deletion

Misspelling:

python src/__main__.py generate data/glue-sst2-validation.csv --attack_type misspelling

Greedy Search

Misspelling

python src/__main__.py greedy-search data/glue-sst2-validation.csv --target_selection most --attack_type misspelling --clean_text True --combination_method sum --output_file ./output/search.csv

Plotting

python src/__main__.py plotting --scoring_method cosine

Sentence Similarity

python src/__main__.py sent-similarity

Input format

The input file should be a two-column csv file with text sample in the first column and true label in the second column. Example:

it 's a charming and often affecting journey .,1
unflinchingly bleak and desperate,0

Output format

The output is written to a file called 'output.html' in the output directory. Output is always written in append mode. The html file can be opened in the browser and shows the output in the same format created by captum's visualize_text function.