Mamy Ratsimbazafy

Results 601 comments of Mamy Ratsimbazafy

I've started working on this and finished porting `types.nim` to fully use `Matrix[M, N, T]` syntax: https://github.com/mratsim/linear-algebra/blob/float-parametric/linalg/private/types.nim ~It passes all the tests~. Edit: I was actually loading the original linalg....

Would love to discuss on Nim numerical ecosystem as well. Maybe on gitter/irc? As @andreaferretti knows, I had tough challenges to use `linalg`'s `DVector` and `DMatrix` types in my autograd...

In France, as well so same timezone. However this is a side project for me so I'm only free during lunch time or in the evening.

I've updated PR https://github.com/unicredit/linear-algebra/pull/14 Base type in linalg can now generic over floats with the original DVector32 ... being provided for compatibility/syntactic sugar: ```Nim type Vector*[N: static[int], T: SomeReal] =...

Just to make sure, the type definition currently is: ```Nim type Vector*[N: static[int], T: SomeReal] = ref array[N, T] Matrix*[M, N: static[int], T: SomeReal] = object order: OrderType data: ref...

Bad news, I realized that when launching tests I was importing the working linalg library instead of my modifications. It seemed to good to be true to pass all the...

I've reset the branch to rework from scratch while making sure I load the WIP version with the test suite. I'll keep the `static` keyword on procs to avoid introducing...

Regarding the solve issue, I've reproduced a minimum test case. It seems like compiler bug https://github.com/nim-lang/Nim/issues/4554: ```nim import linalg let a = matrix(@[ @[3.0, 1.0], @[1.0, -2.0] ], 2, 2)...

For now I'm stuck as all types of workarounds for the `solve` proc do not work: Reproducing with a minimal function actual show that I will have more problems down...

Opened https://github.com/nim-lang/Nim/issues/5641, https://github.com/nim-lang/Nim/issues/5643, https://github.com/nim-lang/Nim/issues/5644 Edit https://github.com/nim-lang/Nim/issues/5645 as well