dadaharry/logistic-regression

Logistic Regression - uWaveGestureLibrary

Methods

This project proposes an implementation of a Multi-Class Logistic Regression for any labeled datasets. These are some features fo the implementation:

• Cost function: cross entropy
• Regularization: L2 regularization
• Optimization: gradient descent
• Stopping criteria: early stopping
• Learning visualization: loss, accuracy, cross validation and confusion map

Dataset

To illustrate the results of this method, the dataset uWaveGestureLibrary is used. This dataset comprised 8 different types of time series representing gestures. A time serie is composed of positions (x,y,z) at each time step.

Run the project

The following command run the project:

python Scripts/main.py

The packages needed are numpy and matplolib. If you desire genererate synthetic blob datasets, sklearn is also needed.

Reports & Figures

The figures in the folder Figure and the slides logistic-regression-uwavegesture.pdf give an overview of the results of the Multi-Class Logistic Regression