numerical: core package for Numerical Haskell project

[ bsd2, library, math ] [ Propose Tags ]
Versions [RSS] 0.0.0.0
Change log CHANGELOG.md
Dependencies base (>=4.10 && <5), ghc-prim (>=0.2 && <0.6), monad-ste (>=0.1 && <0.2), primitive (>=0.5 && <0.8), tagged (>=0.7 && <0.9), transformers (>=0.4 && <0.6), transformers-compat (>=0.4 && <0.6), vector (>=0.11 && <0.13), vector-algorithms (>=0.6.0.1 && <0.9) [details]
License BSD-2-Clause
Author Carter Tazio Schonwald
Maintainer carter at wellposed dot com copyright: Carter Schonwald
Category Math
Source repo head: git clone http://github.com/wellposed/numerical.git
Uploaded by CarterSchonwald at 2019-01-01T03:28:35Z
Distributions
Downloads 552 total (4 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs available [build log]
Last success reported on 2019-01-01 [all 1 reports]

Readme for numerical-0.0.0.0

[back to package description]

Wellposed

Currently in Pre alpha release engineering, so nearly ready for experimental consumption

(though please dont publicize yet)

About Numerical-Core

This is the core Package for Numerical Haskell, a project by Carter Schonwald aka Wellposed Ltd, and (soon I hope!) other contributors.

Numerical-Core is an open source component of the Wellposed® Numerical Haskell software suite.

##Build Status

Build Status

#note

this library is pre alpha release so not all examples / codes may work as expected. That said, the current api should be enough to prototype and typecheck algorithms.

Numerical Haskell

Numerical Haskell is an effort to bring great numerical computation and data analysis tooling to haskell, and be the best possible platform for sophisticated efforts in those same domains

What array Formats are Supported

The initial

What convention is used for indexing?

When you have an index tuple, just think x,y,z to keep track of the meaning. Indexing tuples are written as statically sized lists, eg x:*y:*z:*Nil. This follows the tradition of x,y,z axes used in plotting. Note well: the underlying memory order can be row OR column major or other!

All the computations on these static sized lists get specialized away into nonrecursive computations at their use sites. So in this special scenario, lists aren't a problem!

Contributing

Great! Theres so many awesome ways you could help out. Look at CONTRIBUTING.md for more details. Right now theres a lot of low hanging fruit in improving test coverage, and soon there'll be many opportunities on the performance tuning and numerical algorithms/tooling areas.

bug reports

see bug.md for how to file a bug report

Performance FAQ

  1. How do I use Numerical haskell to write fast code thats outstandingly high level !?
    • The leading cause of poor performance in numerical routines (aside from poor choice in algorithms) is bad memory locality, which has but a single easy cure: ** block recursive algorithms **
    • Yes, you heard me, in compiled languages recursion is pretty cheap outside of the inner most loops! It also is a fantastic tool for facilitating good memory locality!
    • I'm totally serious, try out the benchmarks for the various versions of the same routines we provide!
  2. But, what about fusion?
    • Because of certain aspects of the numerical haskell design, we can't automagically use the fusion optimization facilities of the underlying array representations such as Vector.

Community

Many member of the Numerical Haskell community can be found on #numerical-haskell on freenode IRC There is also the numericalhaskell mailing list

Support

The community provides some basic support through the IRC channel, Mailing list, and the relevant project issue trackers.

If your support needs can't be resolved though those channels, please do not hesistate to contact Wellposed (aka Carter) to find out more about our support and professional services options.