svmlite 0.3.0

A lightweight SVM implementation from scratch

SVMLite

Work in Progress

This project is part of CS6375 Machine learning course at University of Texas at Dallas.

A lightweight Python library implementing Support Vector Machines from scratch for educational and experimental use.

Installation

pip install svmlite

Quick Start

from svmlite.svm import SVCLite
from svmlite.utils import StandardScalerLite
from svmlite.metrics import accuracy_score
import numpy as np

# prepare data
X = np.array([[1, 2], [2, 3], [3, 3], [6, 5], [7, 8], [8, 7]])
y = np.array([-1, -1, -1, 1, 1, 1])

# scale features
scaler = StandardScalerLite()
X_scaled = scaler.fit_transform(X)

# train SVM
model = SVCLite(C=1.0)
model.fit(X_scaled, y, learning_rate=0.01, n_iters=1000)

# predict
predictions = model.predict(X_scaled)
print("Predictions:", predictions)

# evaluate
acc = accuracy_score(y, predictions)
print("Accuracy:", acc)

Features

• Implemenation of primal form (hard margin and soft margin) of SVM Classification using Stochastic Gradient Descent (SGD).
• QP (Quadratic Programming) based SVM implementation using cvxopt
• Kernel Support: Linear, Polynomial, RBF kernel, Sigmoid kernel and Custom Kernel support
• SMO (Sequential Minimal Optimization) algorithm for optimization (simplified heuristic for selecting alpha pairs)
• Multiclass classification using One-vs-One (OvO) and One-vs-All (OvA) strategies

Modules Implemented from Scratch

• SVM Classifier
• Kernel Functions: Linear, Polynomial, RBF, Sigmoid
• Standard Scaler
• Metric functions: Accuracy