Assignment 2: Logistic Regression

Assignment Goal

The goal of this project is to write a class that performs logistic regression and apply it to the NSCLC dataset to predict whether a patient has small cell or non-small cell lung cancer (NSCLC) using demographic, medication, lab values, and procedures performed prior to their diagnosis features.

Assignment Overview

In class, we learned how to derive an Ordinary Least Squares (OLS) estimator in a linear regression model, used to identify the best fitting line (see workshop 2).

For this assignment, we will be implementing a logistic regression model using the same framework.

Logistic regression is useful for classification because the function outputs a value between 0 and 1, corresponding to a categorical classification

For this project, you will be given a set of simulated medical record data (reference) from patients with small cell or non-small cell lung cancer. Write a logistic regression to predict whether a person belongs in one class or another.

Starting the Assignment

Steps:

1. Fork this Repo and clone the fork

2. Create a conda environment from the requirements.txt file 
    - This will ensure you have the minimum requirements to run this repo
    - If you are unsure of how to do this, please consult this conda cheatsheet 

(Conda Cheatsheet)

    - Activate the Conda Environment
    - Make sure you correctly select the interpreter path for VSCode
    
3. Read the rest Assignment Context below
    
4. Complete TODOs in logreg.py 
    - calculate_gradient
    - loss_function
    - make_prediction

5. Install Pytest & Make Local package
    - Use flit: Same process as Assignment 1
    - Note: Required for unit testing

6. Unit Testing
    - Check if fit appropriately trains model & weights get updated
    - Check loss approaches 0 
    - Check predict works as intended

7. Push to GitHub
    - If you need a refresher, please consult this Git cheatsheet 

(Git Cheatsheet)

Assignment Context

Logistic Regression

To allow for binary classification using logistic regression, we use a sigmoid function to model the data. We will define a loss function to keep track of how the model performs. Instead of using Rsquared, like you've done before in workshops 2 & 3, we will be implementing a log loss (binary cross-entropy) function. This function will minimize the error when the predicted y is close to an expected value of 1 or 0.

Resources to help you get started:

• Sigmoid Function (click here)
• Understanding Binary Cross-Entropy and log loss (click here)
• Binary Cross-Entropy mathematical implementation (click here)

Dataset

Class labels are encoded in the NSCLC column of the dataset. This is what you are trying to predict.

• 1 = NSCLC
• 0 = Small cell

A set of features has been pre-selected for you to use in main.py, but feel free to use other features.

• Gender
• Penicillin V Potassium 250 MG
• Penicillin V Potassium 500 M
• Computed tomography of chest and abdomen
• Plain chest X-ray (procedure)', 'Diastolic Blood Pressure
• Body Mass Index
• Body Weight
• Body Height
• Systolic Blood Pressure
• LDL Cholesterol
• HDL Cholesterol
• Triglycerides
• etc...

A full list of features is provided in utils.py. Note: You do not need to modify anything in utils.py.

Grading (20 points total)

Code (9 points)

• Correct implementation of calculate_gradient function (3)
• Correct implementation of loss_function (3)
• Correct implementation of make_prediction (3)

Tests (8 points)

• Test cases for gradient (2)
• Test cases for loss (2)
• Test cases for outputs (2)
• Test cases for accuracy (2)

Documentation & GitHub (3 points)

9. Write comments documenting your code (2)
10. Push your finished assignment to GitHub (1)