Any plans to support TSNE with PCA init?

Question

Any plans to support TSNE with PCA init?

liming-ai opened this issue 3 years ago · 5 comments

Are there any plans to support TSNE with PCA initiation?

Answer 1 · 2021-08-11T04:21:40.000Z

There are no plans to support this natively right now. Instead, it's best to use any PCA library to compute the PCA transform of the vectors, then pass the resulting transformed vectors to TSNE-CUDA.

Answer 2 · 2021-08-11T04:25:33.000Z

There are no plans to support this natively right now. Instead, it's best to use any PCA library to compute the PCA transform of the vectors, then pass the resulting transformed vectors to TSNE-CUDA.

Got it , thanks a lot!

Answer 3 · 2021-08-11T06:50:17.000Z

Hi @DavidMChan , could you please provide some advice about the PCA library with GPU accelerate?

Answer 4 · 2021-08-11T23:19:11.000Z

This is provided by many libraries -
pytorch: https://pytorch.org/docs/stable/generated/torch.pca_lowrank.html
tensorflow: https://www.tensorflow.org/tfx/transform/api_docs/python/tft/pca
skcuda: https://scikit-cuda.readthedocs.io/en/latest/generated/skcuda.linalg.PCA.html

It is also a relatively fast CPU operation with support for multithreading in scikit learn.

Answer 5 · 2021-08-13T06:05:07.000Z

This is provided by many libraries -
pytorch: https://pytorch.org/docs/stable/generated/torch.pca_lowrank.html
tensorflow: https://www.tensorflow.org/tfx/transform/api_docs/python/tft/pca
skcuda: https://scikit-cuda.readthedocs.io/en/latest/generated/skcuda.linalg.PCA.html

It is also a relatively fast CPU operation with support for multithreading in scikit learn.

Thanks again!