Distance metrics are a key part of some machine learning algorithms, such as K-Nearest Neighbors KNN algorithm. Moreover, an effective distance metric can improve the performance of machine learning models, whether that's for classification tasks or clustering. In this project, we conducted experiments using the deep learning features of the (AwA2) dataset. First, we use the KNN algorithm combined with 4 simple metrics (Manhattan distance, Euclidean distance, Chebyshev distance and cosine distance) to conduct experiments and evaluate their performance. We also use preprocessing to improve efficiency(Use LDA to reduce dimension). Secondly, we tried 7 metric learning algorithms ( 4 supervised metric learning methods : LMNN, NCA, LDFA, MLKR; 3 weakly supervised metric learning methods : ITML, SDML, MMC to see different method's effect on KNN.
brianwu-s / distance_metrics Goto Github PK
View Code? Open in Web Editor NEWDistance metrics are a key part of some machine learning algorithms, such as K-Nearest Neighbors KNN algorithm. Moreover, an effective distance metric can improve the performance of machine learning models, whether that's for classification tasks or clustering. In this project, we conducted experiments using the deep learning features of the (AwA2) dataset. First, we use the KNN algorithm combined with 4 simple metrics (Manhattan distance, Euclidean distance, Chebyshev distance and cosine distance) to conduct experiments and evaluate their performance. We also use preprocessing to improve efficiency(Use LDA to reduce dimension). Secondly, we tried 7 metric learning algorithms ( 4 supervised metric learning methods : LMNN, NCA, LDFA, MLKR; 3 weakly supervised metric learning methods : ITML, SDML, MMC to see different method's effect on KNN.
License: MIT License