Satrajit Ghosh

@ghisvail - this is a path we took after careful consideration of all the issues we ran into with dataclasses. we realized we were building attrs when working with dataclasses....

agree with the sentiments here. how about: radio when "other" is not an option and number of items < 7 (magic number for now, presumably re-configurable in the future).

`d_alpha` will be exactly the same. since `L_alpha` is symmetric by requirement of the function. and if `alpha > 0` the normalization is also symmetric. so the implementation is correct...

@mfalkiewicz - ok, i was looking at the wrong place. i think you are talking about step 3 in the code: https://github.com/satra/mapalign/blob/master/mapalign/embed.py#L98 if you see the wikipedia page: https://en.wikipedia.org/wiki/Diffusion_map D_alpha...

@mfalkiewicz - should #13 still be a PR?

also tests are not passing - since the markov computation has changed. i would recommend creating a second `markov` option with your changes, unless you think the current implementation is...

given python, i don't think they can make things really private. but the module name changed from `sklearn.manifold.spectral_embedding_` to `sklearn.manifold._spectral_embedding` (https://github.com/scikit-learn/scikit-learn/blob/main/sklearn/manifold/_spectral_embedding.py). we may have to try both imports to account...

@yibeichan - i don't think we should drop the prov context - we are still based on the prov model. that's more general and the bids context should still conform...

essentially bids prov was a simplification to keep the keys readable to people and perhaps should stay. however, the openprov context should be included or referenced. here is a linkml...

the linkml one looks more comprehensive and since chris did it, i have some trust in it as well. so i would lean towards using it and that also allows...