Open bigheiniu opened 6 years ago
Hi, it also involves other optimizations, but another thing that I have noticed recently is that py_bh_tsne
(which I derived my code from) uses squared Euclidean distance, so the hyperparameters turned out to be not directly comparable to bhtsne
lib. However I did not notice much difference between using squared or proper distance, so left squared to be the default.
Hello, would you like give a reference or short introduction about your optimizations. I'd like to imitate your improvement on Multiple maps of t-SNE, but feel a little confused in reading source code. Thanks!
Do you just implement this algorithm with the help of openMP? Have you ever make some change in the process of bhtsne?