Closed cramirezs closed 1 day ago
NCA is one supervised method and we need labeled data for using that, where having labels are a hard process in biological data. Maybe, it is better to move on semi-supervised metric learning algorithms
Hmm. Has anybody ever used NCA for single-cell data? Not that I know. As a general principle, I don't think we should be adding methods that are not in actual use in the community... Or what do people think? Otherwise we can have hundreds of methods easily.
I think part of the idea behind openproblems is that it will open up single-cell data to the greater ML community. If we don't add it, others probably will. You could ofc still focus your efforts on more widely used methods, but i wouldn't preclude a method.
This issue has been automatically closed because it has not had recent activity.
What is the method? Neighborhood Components Analysis (NCA) tries to find a feature space such that a stochastic nearest neighbour algorithm will give the best accuracy. I feel like this is sort of cheating but might be worth checking how it performs. Here is where it was proposed.
Where is the code located? https://scikit-learn.org/stable/auto_examples/neighbors/plot_nca_dim_reduction.html
Which task(s) could it be used for? Dimensionality reduction.