search

DmitryUlyanov / Multicore-TSNE

Parallel t-SNE implementation with Python and Torch wrappers.
Other
1.89k stars 228 forks source link

Support for embedding in D>2 #5

Closed matteosal closed 6 years ago

matteosal commented 8 years ago

The algorithm does not seem to work properly if the target space is bigger than 2-dimensional. Is there a plan for extended functionality?

DmitryUlyanov commented 8 years ago

Some day I will give it a try

sg-s commented 7 years ago

is that even the point of t-SNE?

matteosal commented 7 years ago

It's mainly used for 2D visualization, so not really. But the real point of t-SNE is to tackle the crowding problem of "heavy" dimensionality reduction (target dimension << starting dimension), which is not tied to D=2 in any way. Hence applying the algorithm for D > 2 is perfectly meaningful and maybe could be of some use to someone.

captify-acherednychenko commented 7 years ago

Hey there. Agree, would be great to have n>2 support.

@DmitryUlyanov , If I'd just change hardcoded do_dim and other vars from 2 to say 5 everywhere, will that work?

Thanks.

DmitryUlyanov commented 7 years ago

No, it will not. This code uses quadTree data structure which can't be used with dim > 2.

captify-acherednychenko commented 7 years ago

Dmitry, thanks for replying

On Wed, Dec 21, 2016, 06:35 Dmitry Ulyanov notifications@github.com wrote:

No, it will not. This code uses quadTree data structure which can't be used with dim > 2.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/DmitryUlyanov/Multicore-TSNE/issues/5#issuecomment-268433411, or mute the thread https://github.com/notifications/unsubscribe-auth/AT-TDL9i9jIj-lVy7NEAREIfDtFTLX5iks5rKKx3gaJpZM4Kna6r .

DmitryUlyanov commented 6 years ago

Supported via #35