al-jshen / compute Goto Github PK

Are there any plans on developing the methods for fitting the given values into the distribution and then find the parameters for that distribution.

I think it'd be very valuable. Fitting a distribution is computationally intensive so being able to do it in rust instead of python or R, would be great.

Switch from autodiff to reverse.

Most of the optimization functions are going to be Rn -> R. Reverse-mode will make things a lot faster because it only requires a single pass to get all the gradients for the inputs wrt to the loss, whereas forward-mode takes n passes.

Could you provide some examples about Levenberg-Marquardt ，i don't know how to use it .Thanks!

binom_coeff panics when k > n

Panics attempt to multiply with overflow if n is 334 and k is 11

+= etc. for vector-vector and vector-scalar. benchmark. if faster, test reimplementing + by cloning self, then applying +=.

use wyrand from https://github.com/Absolucy/nanorand-rs which should be faster, the latest 0.6 release has a method for generating floats