support arbitrary-precision types

[rcurtin]

Reported by rcurtin on 6 Nov 43643786 13:21 UTC
Right now everything works on double and arma::mat. Based on #300, it would be useful to support arbitrary-precision types (within the confines of what Armadillo offers). This ticket right now just proposes the ideas. There will be numerous difficult implementation details that may make this very difficult in the end; those can be detailed in the comments as they come up.

Migrated-From: http://trac.research.cc.gatech.edu/fastlab/ticket/302

[rcurtin]

Commented by rcurtin on 6 Apr 43643790 19:28 UTC
(In #300) You are right, if you use uint8_t, then it will take up about 6GB of RAM and then the method should run. The difficulty here is that mlpack does not support types other than double. The reason for this was API simplicity. I don't have a good solution for this now, or soon, so you shouldn't expect code that will work with uint8_t anytime soon (sorry!). There is some serious code restructuring that would need to happen to make uint8_t datatypes easy and I have other restructuring that right now is more important.

I've opened ticket #302 to propose the idea of arbitrary-precision support; you can follow that ticket. Like I said, I don't think it will happen anytime soon...

I will leave this ticket open as blocked by #302, so that if/when that is finished, this can be closed.

[mlpack-bot]

This issue has been automatically marked as stale because it has not had any recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions! 👍

[shrit]

Probably we will close this one before mlpack 5.0

[rcurtin]

Agreed, I think all the documentation-related refactoring and other work going on right now will end up with all mlpack algorithms accepting a matrix type soon. 👍