Port the Rationals from the PowerPack

[forki]

We should port the Rationals to FSharpx.

It seems the PowerPack development has stopped.

@mausch what do you think?

[mausch]

Interesting... I think this implies merging PowerPack into FSharpx? Or am I reading too much into it? I'm not saying it's a bad thing, just trying to figure out the implications.

[funnelweb]

See the discussion on the F# open source group,

https://groups.google.com/forum/?fromgroups=#!topic/fsharp-opensource/BRpOPujL4f0

Don Syme wrote a suggestion for which functionality to move into fsharpx. For "math" stuff he said:

FSharpx is not the natural home for a math library for F# and it's best not to confuse things and start down that direction. Instead organize that as a separate community project and keep the purpose of FSharpx clear. Much of the other math functionality has already been taken forward at http://mathprovider.codeplex.com/ and there is also Math.NET.

[funnelweb]

I think the point is - if you're going to merge this code anywhere, then it should either be a new "F# math" project or be merged into Math.NET, which is the de-facto open source math library for .NET. It has a module specifically for F# support.

[ovatsus]

The design of Math.NET is very different from the PowerPack math code. Personally, I prefer the matrix classes on Power Pack rather than the ones on Math.NET. It would probably make sense to either have a FSharpx.Math package, or create a new repository FMath or FsMath for it, as there is value there that isn't replaced by Math.NET

[ghost]

I'd really encourage FSharpx not to go down this route. Math.NET looks like a better basis for open-source math programming on with F# on .NET than anything that FSharpx is going to produce. Do not make FSharpx a kitchen-sink for random bits of math functionality that don't add up to a convincing whole - if you do it will repeat the mistake we made with the power pack itself. Indeed, I recommend you just don't get into the "math library" game at all.

Focusing on core functional programming, functional collecctions, async, networking and type providers seems like a good remit for FSharpx. I think extending to math would be a bridge to far.

For the power pack Matrix/Vector types - as I understand it, Math.NET now allows the use of storage containers that use different matrix representations and storage (including native) as long as a basic set of operations (including some linear algebra) are defined on the storage. This may be a better route for progressing the power-pack matrix types - move them into the F# wrapper in Math.NET and hook them into the rest of the Math.NET functionality. If necessary, also wrap some other Math.NET functionality in F# window dressing.

Note we use Math.NET in Try F# 3,0, e.g. see here http://preview.tryfsharp.org/Courses?ID=4.

[ovatsus]

Ok, putting math in FSharpx is probably not a good idea, but it would also be a shame if the existing work on powerpack math just dies. It would be nice to migrate it to a repository on github so it can be mantained by the community. Some things like defining pointwise operators can't be done on Math.Net because it's implemented in C# and operators can't be added on type extensions. Math.Net also doesn't have rational numbers.

[ovatsus]

Sorry, ignore my last comment, powerpack is already on github.com/fsharp/powerpack

[7sharp9]

I used the Matrix type from the Powerpack a while back and it was of better performance than one of the commercial offerings, so there is good stuff in there!

[nrolland]

Apart from having a clear cohesive separation between libraries, the math.net guys have excellent ressources. Another reason to integrate smoothly with them and not bastardize fsharpx

[funnelweb]

(FWIW "math.net" came up as a link in my email but is a spam site - the link is http://numerics.mathdotnet.com/ for those who need it)

[forki]

I opened an issue in the mathnet-numerics project (mathnet/mathnet-numerics#60) so moving the discussion over to this project.

[ovatsus]

@forki I'm trying to remove the PowerPack dependency (#165) by adding LazyList and the Async stuff, but the PowerPack is still needed by DistributionMonad because of BigRational. What do you suggest? Would it make sense to move it to Math.Net F#?

[ghost]

If you follow my advice about "no math", then just remove the monad instance for BigRational :)

[ovatsus]

I'm thinking of submiting a pull request to Math.Net to add the BigRational & Complex from PowerPack and DistributionMonad from FSharpx, but I'd like to hear @forki opinion first, I don't want to delete his work just like that

[ghost]

Another option is to include BigRational and the DistributionMonad but immediately mark them [<Obsolete>].

[ovatsus]

Sent pull request 172

[forki]

You can move the DistributionMonad over. I'm ok with that.
The DistributionMonad is more a sample. I don't think it's that usefule since it's O(2^n)