Active-Learning-Playground

This program compares the performance of active learning and no active learning with Gaussian Process Regressor. The data it takes are kmer protein sequences(in protein alphabet) and their corresponding protein affinity values(continuous value) in a csv file. The protein alphabet is first tranformed to binary vectors using one hot encoding so that each letter becomes a feature. It then trains Gaussian Process Regressor using active learning with query strategy picking samples with highest standard deviation according to the regressor prediction. Note: currently, the active learning performs worse than no active learning. ;(