I noticed that even when parameters are explicitly filled in in some of the dml constructors (e.g. NCA, LMNN), when I try to look at those values, I keep getting that the values are set to None and not the default values according to the documentation.
My first thought was that since many of the arguments shadowed your default arguments, that could be why they didn't show up, but the Nones persisted even when I changed those parameters. (But then I remembered that that makes no sense, so that couldn't be the issue.)
Also, side question: I know that algorithms of a nearest-neighbor variety are inherently expensive even when you're using the Euclidean R^N metric and not also trying to learn a metric, but when I tried to fit LMNN on a dataset of size: 6460 x 143 feels like it shouldn't be taking upwards of 10 hours. Is this extremely long runtime normal?
Finally, I just wanted to say that the survey paper that led me to your repo was wonderfully written.
Cheers,
Aaron
I noticed that even when parameters are explicitly filled in in some of the dml constructors (e.g. NCA, LMNN), when I try to look at those values, I keep getting that the values are set to
Noneand not the default values according to the documentation.My first thought was that since many of the arguments shadowed your default arguments, that could be why they didn't show up, but the
Nones persisted even when I changed those parameters. (But then I remembered that that makes no sense, so that couldn't be the issue.)Also, side question: I know that algorithms of a nearest-neighbor variety are inherently expensive even when you're using the Euclidean R^N metric and not also trying to learn a metric, but when I tried to fit
LMNNon a dataset of size: 6460 x 143 feels like it shouldn't be taking upwards of 10 hours. Is this extremely long runtime normal?Finally, I just wanted to say that the survey paper that led me to your repo was wonderfully written.
Cheers,
Aaron