DSLR – Data Science × Logistic Regression

This project focuses on building a logistic regression classifier to predict Hogwarts Houses from student data. Implemented as a team project, it demonstrates proficiency in data analysis, visualization, preprocessing, feature selection, and machine learning workflows, while highlighting collaboration, code design, and problem-solving skills.

---

Installation

To set up the project and install all dependencies, run:

$ make install

You should see:

Dependencies installed successfully!
To activate the virtual environment, run:
source config/.venv_dslr/bin/activate

Activate the environment:

$ source config/.venv_dslr/bin/activate

(.venv_dslr) $

---

Running the Programs

All programs are located in the src/ folder. You can run them with Python 3:

python3 src/describe.py

For detailed usage information:

python3 src/describe.py --help

---

Programs Overview

Data Analysis

• describe.py [file] [--advanced]
  • Arguments:
    • file (optional): Path to the CSV file (default: datasets/dataset_train.csv)
  • Options:
    • -h, --help: Show help message
    • -a, --advanced: Include advanced statistics (missing values, unique counts, IQR)
  • Purpose: Generate a comprehensive statistical summary of numerical features in the dataset.

Data Visualization

• histogram.py [file]

  • Purpose: Creates histograms for each numerical feature grouped by Hogwarts House, showing grade distributions.

• scatter_plot.py [file]

  • Purpose: Creates scatter plots to visualize correlations between selected features (e.g., Defense Against the Dark Arts vs Astronomy).

• pair_plot.py [file]

  • Purpose: Generates a pair plot matrix to visualize relationships between features, with points colored by House.

Machine Learning

• logreg_train.py [file]

  • Purpose: Trains a one-versus-all logistic regression model to classify students into Hogwarts Houses.
  • Features used: Potions, Charms, Transfiguration, Astronomy, Divination, Ancient Runes, Muggle Studies
  • Output: Saves trained model weights to weights.txt.

• logreg_predict.py [file]

  • Purpose: Uses the trained model to predict House assignments for new students.
  • Output: Saves predictions to houses.csv and prints results in the terminal.

---

Example Usage

# Explore dataset statistics
python3 src/describe.py datasets/dataset_train.csv

# Visualize grade distributions
python3 src/histogram.py datasets/dataset_train.csv

# Explore feature correlations
python3 src/scatter_plot.py datasets/dataset_train.csv

# Generate pair plots for multiple features
python3 src/pair_plot.py datasets/dataset_train.csv

# Train logistic regression model
python3 src/logreg_train.py datasets/dataset_train.csv

# Predict Hogwarts Houses for test dataset
python3 src/logreg_predict.py datasets/dataset_test.csv

---

Model Details

The logistic regression model uses one-versus-all classification:

• 4 binary classifiers (one per House)
• Each classifier predicts if a student belongs to that House or not
• Features standardized using z-score normalization
• Sigmoid activation function with binary cross-entropy loss
• Gradient descent optimization with up to 10,000 iterations and early stopping

---

Output Files

• weights.txt: Contains trained model weights for each House
• houses.csv: Predictions for the test dataset in the required format