Added tools for dimension reduction and quality assesment #17
Conversation
There was a problem hiding this comment.
this looks good. The tests took 10 min to complete on my machine, so I would consider mark the test_cli as slow, since we have one occurence of it it the fast_tests. For the "slow tests", I wonder how many combinations we need to test. After all, we are not really testing the underlying libraries but our wrapping of it and data formatting.
|
I did not see at first that the evaluation code was in the same PR. However, I was expecting umap to be more robust than that on the digit dataset, specially for value of k close to the sample value for each class (~= 180). So how to interpret this score? Also it would have been interesting to compare several projections methods and see how they perform. |
No description provided.