malmgrek/political-projections

:telescope: Dimensionality reductions of political surveys' results

Dimensionality reduction demo application

Summary

This project implements some well-known dimensionality reduction methods

• Principal component analysis (PCA),
• Independent component analysis (ICA)
• Factor analysis (FA)

for the Finnish national broadcasting company survey for parliamentary electoral candidates, as well as the "expert level dataset" in the 2019 Chapel Hill expert survey. In principle, analogous methodology is applicable to any standardizable survey datasets. The main aim of this project to demo how an interactive Plotly Dash service works for exploring the results. Nevertheless, closer examination of the results may reveal some interesting facts about present day politics such as clustering of typical questions regarding nationalism, multiculturalism, liberalism, left/right and so on.

A version of the app is running at https://political-projections.herokuapp.com.

Table of Contents

• User guide
  • Quick start
  • For developers
    • Install development environment
    • Unit testing
    • Running the Dash service locally

User guide

Quick start

Build and run application locally using Docker:

make up

Stop the running container:

make stop

When the application started succesfully, there should be a Dash application running at http://localhost:8050.

For developers

Install development environment

Install dependencies and the package using Conda:

conda env create --force --name dimred-demo --file environment.yml
source /path/to/conda/bin/activate dimred-demo
pip install -e .

Optionally, install manually useful development tools such as Jupyter.

Unit testing

pytest -v          # Run all tests except webtests
pytest --webtest   # Run all tests including webtests

Running the Dash service locally

The Plotly Dash service can be launched with

python app.py

By default, the service runs at http://localhost:8050.