Using Partial Labelling (Semi-Supervised UMAP)

[yuan596]

![Uploading ~S]4WTV4A4J7D]6V)RB2RXO.png…]() hello, I only have some of my data labelled when I use umap. How to glabelset the predicted labels of the unknown label data points？

[lmcinnes]

If you create a vector where labels are given a positive value and unlabelled points are given a value of -1 then that will be interpreted appropriately when passed as the y vector for fit.

[yuan596]

hello, I have already done this step.But I want to know which label of unlabelled points belong to in the result.It is best to output the labels predicted by UMAP 

------------------ 原始邮件 ------------------ 发件人: "lmcinnes/umap" @.>; 发送时间: 2021年10月13日(星期三) 凌晨5:39 @.>; @.@.>; 主题: Re: [lmcinnes/umap] Using Partial Labelling (Semi-Supervised UMAP) (#781)

If you create a vector where labels are given a positive value and unlabelled points are given a value of -1 then that will be interpreted appropriately when passed as the y vector for fit.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[lmcinnes]

UMAP does not predict labels, it merely provides a low dimensional representation. If you want to predict labels you'll need to use a classifier on the low dimensional representation.

[yuan596]

if I want to use alignedUMAP ,what kind of data sets are needed for this method?

------------------ 原始邮件 ------------------ 发件人: "lmcinnes/umap" @.>; 发送时间: 2021年10月13日(星期三) 晚上11:11 @.>; @.@.>; 主题: Re: [lmcinnes/umap] Using Partial Labelling (Semi-Supervised UMAP) (#781)

UMAP does not predict labels, it merely provides a low dimensional representation. If you want to predict labels you'll need to use a classifier on the low dimensional representation.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[yuan596]

when I use labels to separate classes (Supervised UMAP),the points of the same class are basically clustered together.Can I get the weights/contributions of this variables in the dimensionality reduction? I mean,which variables can separate classes cleanly?

[lmcinnes]

It is almost certainly the labels that are contributing the clean class separation. Moreover since UMAP works fundamentally with distances and inter-relationships rather than coordinates it is not really possible to extract any meaningful feature weights or contributions. The best approach, if that's what you really need, would be to try using ParametricUMAP (possibly in semi-supervised; see some of the ParametricUMAP notebooks for more details) and then analyse the resulting learned network weights which are an explicit mapping from input coordinates to the resulting space.

[yuan596]

Thank you very much for your answer.

[yuan596]

if I want to use alignedUMAP ,what kind of data sets are needed for this method?

[jc-healy]

AlignedUMAP attempts to embed two data sets into a joint space through regularization. You need two datasets with some subset (maybe all) of your points present in both datasets. Then you pass those datasets into AlignedUMAP along with a dictionary telling UMAP which points in your data match. It returns you a pair of embedding where it tried to keep the common points at the same co-ordinates.

https://umap-learn.readthedocs.io/en/latest/aligned_umap_basic_usage.html

On Mon, Oct 18, 2021 at 2:51 AM yuan @.***> wrote:

if I want to use alignedUMAP ,what kind of data sets are needed for this method?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/lmcinnes/umap/issues/781#issuecomment-945416863, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC3IUWXKW7FHOSWDISEJ25DUHO7WNANCNFSM5F2Q6O3Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.