lmcinnes / umap

Uniform Manifold Approximation and Projection
BSD 3-Clause "New" or "Revised" License
7.34k stars 798 forks source link

Error in intersection of two simplical manifolds (when densmap = True) - KeyError: 'graph_dists' #872

Open kunind27 opened 2 years ago

kunind27 commented 2 years ago

When I try to embed categorical and numerical data separately and try to get their intersection using the operator. That is `final_embedding = numeric_embedding categorical_embedding, when for both embeddingsdensmap=True`. I get an error - KeyError: 'graph_dists'.

This is because in the file umap/umap_.py on Line#1111 in the method simplical_set_embedding the statement dists = densmap_kwds['graph_dists'] is incorrect.

That is because, in the file umap/umap_.py on Line#2067 in the method _mul_() the statement resut._densmap_kwds=... assigns a dictionary, which does not have any key named 'graph_dists'. Could you please look into this issue?

lmcinnes commented 1 year ago

I think densmap may be incompatible with combining models I'm afraid.

On Tue, Jun 7, 2022 at 10:08 AM Kunind Sahu @.***> wrote:

When I try to embed categorical and numerical data separately and try to get their intersection using the operator. That is final_embedding = numeric_embedding categorical_embedding, when for both embeddings densmap=True. I get an error - KeyError: 'graph_dists'.

This is because in the file umap/umap_.py on Line#1111 in the method simplical_set_embedding the statement dists = densmap_kwds['graph_dists'] is incorrect.

That is because, in the file umap/umap_.py on Line#2067 in the method mul() the statement resut._densmap_kwds=... assigns a dictionary, which does not have any key named 'graph_dists'. Could you please look into this issue?

— Reply to this email directly, view it on GitHub https://github.com/lmcinnes/umap/issues/872, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC3IUBKS6FITG5QOCVOD3CLVN5JVVANCNFSM5YDCREXA . You are receiving this because you are subscribed to this thread.Message ID: @.***>