[feature request] Kernel pca

[magehex]

Would be great if you could implement a kernel pca running on cuda

[danielhanchen]

Hello @magehex! Thanks for your feature request and thanks for using cuML! Kernel PCA might be in cuML in a later release :)

By the way if you are using Kernel PCA for data visualization purposes, have you maybe tried cuML's UMAP or TSNE?

[magehex]

Hi danielhanchen,

Thanks for your answer. How many time (approx.) will it takes until the later release?

I'm not using Kernel PCA for data visualizaton purposes :)

[cjnolet]

Indeed, this would be a great addition to cuML. In fact, we have already done some preparation for this by adding primitives for efficiently computing kernel grammian matrices during the creation of SVM [1].

This is not on our near-term roadmap currently, and it would be hard for us to give an exact release, though having feature requests like this definitely helps to get things like this on the list.

[1] https://github.com/rapidsai/cuml/blob/branch-0.11/cpp/src_prims/matrix/grammatrix.h

[codercody]

New contributor here, I'll give it a shot since I have some research experience with kpca.

[oyilmaz-nvidia]

I think this can be added fairly easily. Some of the nonlinear functions like sigmoid is already included in ml-prims. After the nonlinear mapping, existing pca can be used.

Don’t know when this can be added to roadmap though.

Onur Yilmaz | NVIDIA Solutions Architect cell: (201) 455 9226 email: oyilmaz@nvidia.com

---

From: Cody Johnson notifications@github.com Sent: Sunday, November 3, 2019 2:15:01 PM To: rapidsai/cuml cuml@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Subject: Re: [rapidsai/cuml] [feature request] Kernel pca (#1317)

New contributor here, I'll give it my shot since I have some research experience with kpca.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/rapidsai/cuml/issues/1317?email_source=notifications&email_token=AINLU4NS2LBQ5EMYM4ZBGFDQR4IKLA5CNFSM4JHH5OQKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEC52OFI#issuecomment-549168917, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AINLU4KZX6TEVDO3QVAZBRLQR4IKLANCNFSM4JHH5OQA.

---

This email message is for the sole use of the intended recipient(s) and may contain confidential information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message.

[cjnolet]

@codercody, we certainly love contributions!

Feel free to open a Pull request as soon as you feel ready. You can also open a work in progress pull request if you would like to get early feedback.

Take a look at the contributor guide and development guides in the repository for general contribution info. Also feel free to reach out with questions.

[codercody]

@cjnolet Actually I just realized I’m not allowed to contribute to open source by company policy. :( On the bright side, my friend @KathrynMO who contributed to Dask with me said she’s on it!

[teju85]

@codercody and @KathrynMO did any of you folks start on this one? If yes, any updates, please?

[shub2k]

2020 is going to end and still they didnt add Kernel PCA !