predictive-modeling-py

Predictive modeling in Python.

This code in this repository attempts to build a predictive model on a subset of data collected by Steinmetz et al. for their 2019 publication. Final project for STA 141 @ UC Davis.

Directory

The contents of this repository are described below.

Folders

• scripts/ contains the files used to load in the .rds session files into Python and store them.
  • scripts/open_rds.py contains a script used to open the .rds files and store them as ListVector objects. See the module docstring for details.
  • scripts/mouse.py creates a class to contain the information on a given session. Data are transformed to have uniform data types across sessions (mainly numpy.ndarray and pandas.DataFrame objects). This class also transforms some of these raw data into features that will be used for the predictive models. These features are not present in the orginal .rds files and are used extensively in the report.
• data/ contains the original .rds files.
• test_data/ contains the test .rds files used to evaluate the model.
• figures/ contains all of the plots used in the report. The report does not pull from this folder; the plots are created dynamically. This folder serves as a separate repository of images.

Notebooks

• data_structure.ipynb contains scripts to test the pipelines between R and Python. It serves as validation effort to ensure the open_rds.py file is working well and that the data transformation applied by mouse.py are giving accurate results when compared to the raw data.
• exploratory_data_analysis.ipynb contains the basic data analysis conducted on the data for the purpose of this report.
• model_training.ipynb conducts the actual traning of models and the selection of a "best" model.
• model_test.ipynb contains the scripts that test the full model on the test sets in test_data/ and an analysis on the effectiveness of that model.

The Model

• random_forest_model.pkl is the final model selected from model_training.ipynb. Use joblib.load('random_forest_model.pkl') to load the model.

The Report

• report.ipynb contains the written report.
• report.html contains the HTML compiled report, generated via jupyter nbconvert report.ipynb --to html --no-input to omit the code cells. This is the file submitted for the assignment.