GSoC'24 Progress: Enhancing Discriminant Analysis and Exploring Generalized Additive Models

Hello Everyone,
I'm thrilled to share the latest updates on my project. Here's what I've been working on recently:

Achievements

Implementation of the ClassificationDiscriminant Class: I have successfully implemented the class definition for ClassificationDiscriminant. This class includes essential methods such as predict, loss, margin, and crossval, enhancing its functionality for discriminant analysis. This addition significantly improves our model's capability to classify data by finding the linear combination of features that best separates different classes. 'ficdiscr' function for training the discriminant analysis model provides the necessary parameters for classification tasks.

Modification of the ClassificationPartitionedModel Class: To accommodate the new discriminant class, I have modified the ClassificationPartitionedModel class. This ensures that the model can handle cross-validation and other partitioned operations seamlessly with the new discriminant class.

Current Focus

Improving the MinGamma Property:  While the ClassificationDiscriminant class is mostly complete, the MinGamma property still needs some improvement. This property is crucial for regularization in linear discriminant analysis, and refining it will enhance the model's performance and stability. 

Development of the ClassificationGAM Class: I have started working on the ClassificationGAM class, which will bring generalized additive models (GAMs) into our toolkit. GAMs are powerful for capturing non-linear relationships between features and the target variable, and their inclusion will broaden the scope and applicability of our models. 

Thank you for following along! 😀

Comments

Post a Comment

Popular posts from this blog

GSoC'24 Project: Adding GAM and Discriminant Classification Classes and Implementing Missing Methods

I am Ruchika Sonagote from India, and I am currently studying Computer Science at the Indian Institute of Technology Gandhinagar. I am thrilled to share that I will contribute to the GNU Octave Statistics Package this summer. I am thankful to Andreas Bertsatos for mentoring me throughout this project. I am enthusiastic about the opportunity to contribute to the GNU Octave Statistics Package and am committed to making meaningful contributions to its development. This project aims to significantly extend the classification capabilities of the GNU Octave statistics package by implementing additional methods for the existing `ClassificationKNN` class and introducing new classification classes, namely `ClassificationDiscriminant` and `ClassificationGAM`. These enhancements will bridge the gap between GNU Octave and MATLAB in terms of advanced data analysis and machine learning functionalities, fostering a richer environment for scientific computing. Proposed Timeline ...
Read more

GSoC'24 Week 1 & 2 Progress: Building a Cross-Validation Framework in Octave

Image
In the past two weeks, I have been working on enhancing the statistics package for GNU Octave by implementing a cross-validation framework for classification models. The focus has been on implementing a 'crossval' method in the 'ClassificationKNN' class and creating a ‘ClassificationPartitionedModel’ class and a ‘kfoldPredict’ method. Together, these provide a solid foundation for cross-validation in Octave. This blog will walk you through the progress, challenges, and solutions implemented.  Cross-validation is a method used to evaluate and improve the performance of machine learning models. It is essential in classification models, where the goal is to categorize data into predefined classes. Cross-validation helps assess how well a model will generalize to an independent dataset, which is critical for ensuring its robustness and reliability. The process involves partitioning the dataset into a set of folds, training the model on some folds, and then validating it on ...
Read more

GSoC'24 Progress: Implementing 'loss' function for 'ClassificationKNN' class

I am excited to share the latest progress on my project. Since my previous blog post, I have been working diligently on implementing a 'loss' function for the 'ClassificationKNN' model. This journey has been challenging and rewarding, pushing me to overcome several obstacles. Achievements Implementation   of the 'loss' Function: I successfully implemented the 'loss' function for the 'ClassificationKNN' model. This function is crucial as it measures error between the predicted and actual class labels, guiding the model towards better performance.  Overcoming Challenges: One of the main difficulties I encountered was normalizing the weights of observations. I discovered a bug in the 'Prior' property of the 'ClassificationKNN' object, which was supposed to help normalize the weights. This bug is a significant roadblock.  Next Steps Fixing the Bug: I will delve into the 'ClassificationKNN' function to locate and correct the er...
Read more