Linear Classifiers — Perceptron, Logistic Regression, Linear SVM, SGD

Problem. Build fast, interpretable linear models for classification and compare training dynamics, regularization, and decision boundaries.

Data. Tabular features with a binary or multiclass label (dataset/Movies Dataset.zip).

Approach.

• Preprocessing: imputation (if needed) + StandardScaler.
• Models: Perceptron, LogisticRegression, LinearSVC, SGDClassifier.
• Regularization: L2 (C / alpha sweep), early stopping where applicable.
• Evaluation artifacts: decision boundary plots, confusion matrix, learning curves.
• Options: class_weight='balanced' for imbalance; feature importance via coefficients.

What I Learned.

• How scaling affects margin-based linear models.
• Trade-offs between Logistic (probabilities) vs Linear SVM (margins).
• Regularization strength (C/alpha) vs generalization.
• When SGD is preferable for larger datasets.

Quick Start

# clone
git clone https://github.com/Joe-Naz01/linear_classifiers.git
cd linear_classifiers

python -m venv .venv
# Windows: .venv\Scripts\activate
source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook