Skip to content

murbany/dnp_capstone

Repository files navigation

Capstone for Harvard's AC297r: Large Language Model Assisted Survey Webapp

In collaboration with the Data Nutrition Project (DNP), we created a platform for ML assisted surveys to investigate improvemets in response quality for the existing survey used to generate the DNPs Data Nutrition Label (DNL)

Installation

Using Python 3.9+

Can install most of the env with conda, but need to run the pip install the torch txt file bc it doesn't play nice with the conda env u:admin pw:secret at login

Then create a new application (or embed to the existing one). See example_app.py as for example. Create config similar to example_config.py. It must contain QUESTIONNAIRE_DIR which must point to a directory with questionnaire files and QUESTIONNAIRE_SUBMISSIONS_DIR where submissions will be written. Don't forget to udpate QUESTIONNAIRE_BASIC_AUTH tuple. It is used for authentication to the questionnaire list and results pages.

Usage

Site root provides a list of existing questionnaires with links to forms and results.

Look at questionnaires/texteditor.json for a questionnaire example. Each questionnaire is an object with the following keys:

extends
title
comment
template
submit
questions

With extends you can specify a different json file as a base, it should be prefixed with _ so it will not be considered a stand-alone questionnaire. With template you can specify alternative template. If not specified questionnaire.html is used. questions is a list of objects which represent (surprisingly) questions. Each such object must be one of the following types: string (text), text (textarea), radio and checkbox. radio, checkbox questions must have options. To mark option as "other" field (a field with a text input) prefix it with "+" sign.

You can override templates by creating templates with the same names in your application's template directory or you can specify template option for desired questionnaires.

About

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published