nicklogin / distractorselector

Supervised generation of distractors for language testing

Distractor Selector

1. Installation

Clone this repository and install requirements

git clone https://github.com/nicklogin/DistractorSelector
python -m pip install -r requirements.txt
python -m nltk.downloader punkt

2. Installation via Docker

You can install DistractorSelector image from Dockerhub:

docker pull niklogin/disselector:latest

You can also build a Docker image from source after cloning this repository:

docker build . -t disselector:latest

3. Usage via terminal/command line

DistractorSelector accepts a CSV file with sentences as an input and outputs a CSV file with distractors.

The input file must contian the following fields (example - gold_standard/gold_standard_input.csv):

Masked sentence - The sentence where the target word is replaces with [MASK] token

Right answer - The target word

Run this command to get output of DistractorSelector:

python -m distractor_generator --

DistractorSelector accepts the following set of CLI arguments:

Argument	Description	Default value
--filename	Path to input file	gold_standard/gold_standard_input.csv
--output_filename	Path to output file	data/gold_standard_output.csv
--sep	Field delimiter in a CSV file	;
--index_col	The name of the index column in a CSV file	None
--n	Number of distractors on the classifier input	20
--no-clf	Do not use classifier	-
--clf_path	Path to file with the saved classifier model	XGBAllFeats/clf.pkl
--cols_path	Path to file with feature names to be used with the classifier	XGBAllFeats/cols.json

4. Usage via a WebAPI

Execute this command in command line/terminal to run the API of DistractorSelector:

python -m api

The documentation will be available at http://localhost:5000/docs