dssg/cincinnati2015-public

Name: cincinnati2015-public

Owner: Data Science for Social Good

Description: Predicting blight in Cincinnati

Created: 2015-08-27 21:18:39.0

Updated: 2017-06-27 15:11:12.0

Pushed: 2017-04-20 19:24:59.0

Homepage: null

Size: 342

Language: Python

GitHub Committers

User	Most Recent Commit	# Commits

Other Committers

User	Email	Most Recent Commit	# Commits

README

Note: This repository reflects the work completed during the summer of 2015. To see the continuation of the project, please visit the current repo.

About

First settled in 1788, Cincinnati is one of the oldest American cities west of the original colonies. Today, the city struggles with aging home stock, stifling economic redevelopment in some neighborhoods.

DSSG is working with the City of Cincinnati to identify properties at risk of code violations or abandonment. We hope that early intervention strategies can prevent further damage and stimulate neighborhood revitalization. Read more about our project here.

Getting started

Get the code

git clone https://github.com/dssg/cincinnati2015-public.git
cd cincinnati2015-public

Install all pre-requisites

conda create -n "cincinnati" --yes --file requirements.conda python=2.7
source activate cincinnati

Configure database

cp dbconfig.sample dbconfig.py
update database configuration in dbconfig.py

Load data into postgres

… see the etl directory

Create features from the data

… see the blight_risk_prediction directory

Run the modeling pipeline

Create output directories

mkdir results
mkdir predictions

Configure the model

edit default.yaml (options are documented in default.yaml)

Run the model

python -m blight_risk_prediction/model

Output

Each model run produces a pickle file which contains:

• the full list of parcels predicted to have violations
• the configuration file used to generate that model
• feature importances

These output files include a timestamp in their filename such that they will not be accidentally overwritten. These files can be used with the evaluation web application in evaluation.

Repository layout

• blight_risk_prediction - our modeling pipeline
• docs - some additional documentation
• etl - scripts for loading the Cincinnati datasets into a postgres database
• evaluation - webapp we use for comparing different models
• postprocess - add details (e.g. address) about properties to predictions
• targeting_priority - re-rank predictions according to some targeting priority
• test - unit tests