Double stochastic

• test that OOS mapping is close for same points in cdist (problem could be the zero-diagonal kernel that is considered important in [1])
• make circlefrom [1] and make as unittest
• make cdist_kwargs more flexible to clearly distinguish all normalization cases (see [1])
• generalize the bi-stochastic normalization for sparse matrices

Papers

Important

• [1] https://epubs.siam.org/doi/abs/10.1137/20M1342124 (open access!)

Others

• https://arxiv.org/pdf/1711.06711.pdf
• https://matthewhirn.files.wordpress.com/2020/06/2013-coifman-and-hirn-bi-stochastic-kernels.pdf