Binary Classifiers
This repository contains the following:
1. A method to estimate the sample complexity of binary classifiers
- Implementation of the Perceptron, Winnow, 1-Nearest Neighbor and Linear Regression algorithms
- Binary search method to improve efficiency of computation
- Uncertainty of estimates
Key results
2. Implementation from scratch of a one-versus-all Kernel Perceptron
- Implementation of Polynomial and Gaussian kernels
- Implementation of one-versus-all technique
- Augmentation of the perceptron algorithm using the above to classify digits
- Identification of images that are the hardest to classify