Occasionally, distances in 3D view mode doesn't reflect cosine score in the top 100 hit list.

mwang87 / SMART_NMR

Repository to help develop SMART

https://smart.ucsd.edu/

MIT License

0 stars 0 forks source link

Occasionally, distances in 3D view mode doesn't reflect cosine score in the top 100 hit list. #131

Open beowulf-zc opened 4 years ago

beowulf-zc commented 4 years ago

So what I got from the discussion this morning with Ming, Raphael, Hyunwoo and Gary was that we have the primary top 100 hits, the cosine score list in the embedding projector, and the 3D cluster map. All three may be inconsistent with each other.

mwang87 commented 4 years ago

what is "3D cluster map"?

beowulf-zc commented 4 years ago

The 3D data point view in the embedding project.

mwang87 commented 4 years ago

So this is the same thing as "cosine score list in the embedding projector"?

beowulf-zc commented 4 years ago

not sure it is based on the same cosine score as the list in the right panel. But can be a 3D distance measure.

mwang87 commented 4 years ago

Where is the number coming from? Can you provide a screenshot of what number you're talking about?

beowulf-zc commented 4 years ago

Hi Ming, I am talking about these numbers.

mwang87 commented 4 years ago

The data passed to projector is confirmed to be consistent with the results page. Differences will be because of different similarity function calculated by projector.