How badly does it affect if we compare two machine learning algorithms one using Cosine Sim and other using Cosine Distance?

Question

I am working on a research project such that I need to compare several distance based classifiers - say TF-IDF based KNN and Kmeans for clustering. Suppose I use Cosine Similarity for one and Cosine Distance for the other -- how will it affect the evaluations?

If by cosine distance you mean `1-cos` or `sqrt(1-cos)` then it is directly tied with cosine similarity, [being the euclidean distance for the unit-normalized vectors](http://stats.stackexchange.com/a/36158/3277). Therefore informationally both are equivalent. But they are different form - angular measure vs linear distance. Choose what you think is more appropriate for you classification algorithm or interpretation. — ttnphns, Jul 01 '16 at 11:31

score 1 · Answer 1 · edited May 23 '17 at 12:39

1

If you meant Euclidean distance by Cosine distance,it is susceptible to entities being clustered by their L2-norm (magnitude, in the 2-dimensional case) instead of direction. i.e., vectors with quite different directions would be clustered because their distances from the origin are similar.

for the effects you can try it out by yourself referring to the answers to this question

edited May 23 '17 at 12:39

Community

1

answered Jul 01 '16 at 10:30

iamgr007

113
5

How badly does it affect if we compare two machine learning algorithms one using Cosine Sim and other using Cosine Distance?

1 Answers1