I have done several ad hoc models for specific elections and intend to:

• make code accessible as modules (code was in Jupyter notebooks)
• want to pull together my work into a more generalized form
• systematically track different methodologies
• keep "data wrangling" code separate from model code when possible

This project will use the pipenv (great tool) from: [http://github.com/hemmendinger/analysistools]

Uses Pandas.Dataframe as its main datastructure.

Data wrangling operations that are "editorial" in nature are included. By editorial I mean, an operation that reflects some subjective decision or assumption that could affect the model's performance.

• Make it easy to add, remove, and change model features
• Make models easily reproducible
• Adequate performance for tens of thousands of polls

Watch for bugs due to ordering of dates, currently assuming ascending, but will attempt to avoid this problem.