DSLR - Data Science and Logistic Regression

Harry Potter project: implementation of a magical sorting hat using logistic regression.

Project Structure

dslr/
├── describe.py              # Data analysis
├── histogram.py             # Histogram visualization
├── scatter_plot.py          # Scatter plot visualization
├── pair_plot.py             # Pair plot visualization
├── logreg_train.py          # Model training
├── logreg_predict.py        # Prediction
├── evaluate.py              # Complete evaluation
├── config.yaml              # Configuration file
├── Makefile                 # Build automation
└── dslr/                    # Main package
    ├── core/                # Business logic
    │   ├── model.py         # Logistic regression model
    │   ├── preprocessing.py # Data preprocessing
    │   └── statistics.py    # Statistical functions
    ├── optimizers/          # Optimization algorithms
    │   └── gradient_descents.py
    └── visualization/       # Visualization tools
        └── plots.py

Usage

Step 1: Data Analysis

python describe.py datasets/dataset_train.csv

Step 2: Visualization

python histogram.py datasets/dataset_train.csv
python scatter_plot.py datasets/dataset_train.csv
python pair_plot.py datasets/dataset_train.csv

Step 3: Machine Learning

# Training
python logreg_train.py datasets/dataset_train.csv

# Prediction  
python logreg_predict.py datasets/dataset_test.csv

# Complete evaluation
python evaluate.py

Model Overview

• Features used: 17 (after feature engineering)
• Supported optimizers: Gradient Descent, SGD, Mini-batch

Feature Engineering

The model automatically uses:

• All original numerical features
• Best Hand encoded (Left=0, Right=1)
• Age calculated from Birthday
• Birth_Month and Birth_Season

Configuration

Modify config.yaml to adjust:

• Learning rate (default: 0.1)
• Number of epochs (default: 10000)
• Optimizer (sgd/batch/mini-batch/compare)
• Batch size (default: 32)
• Selected features (default: all)