Open jolespin opened 7 years ago
HI @jolespin
thanks for you interest in the fast tSNE implementation.
You can, however, not call the ptsne.cpp
directly, but have to compile it into a binary first.
This is why the VizBin.jar
integrates these and locally installs a binary matching the respective OS.
This is also what @piotr-gawron referred to, i.e., you must have a binary compiled for the respective OS.
We provide more info on this under https://github.com/claczny/VizBin/wiki/Minimal-Build-Environment. For more info on the cross-compilation, you should also have a look at the respective Makefile @ https://github.com/claczny/VizBin/blob/master/src/backend/bh_tsne/Makefile.
Cheers,
Cedric
Thanks @claczny, I know you're busy with PhD stuff (from one of the VizBin threads) so I really appreciate the response. If VizBin is already compiled on a local machine, can the binary that it generates can be used for this? Or would it need to be generated separately.
Yes it can. For instance if you use dist jar file: https://github.com/claczny/VizBin/blob/master/VizBin-dist.jar you can extract it and go to /tsne dir. In this directory you can find precompiled tsne for Windows, Linux and MacOS.
@jolespin
Regarding
I know you're busy with PhD stuff (from one of the VizBin threads)
that must be an older post ;) Currently finishing my first PostDoc. That does not mean that I am less busy now, though XD.
I want to write a Python wrapper that calls the very fast implementation in https://github.com/claczny/VizBin/blob/master/src/backend/bh_tsne/ptsne.cpp in the backend and then load the embeddings into
pandas
ornumpy
.My main question is how to call the ptsne.cpp script in an OSX terminal?
It looks in the input data just needs to be parameters on each line and then the PCA components for each row separated by spaces between columns. Apologies if this is out-of-scope but I can post my code (if it works) on here if anyone would find it useful.
Suggested new issue via https://github.com/claczny/VizBin/issues/15