hs-carbon alternatives and similar packages
Based on the "hs" category.
Alternatively, view hs-carbon alternatives based on common mentions on social networks and blogs.
hs-carbon-examples3.2 0.0 hs-carbon VS hs-carbon-examplesExamples of Monte Carlo simulations written with Carbon
Do you think we are missing an alternative of hs-carbon or a related project?
A Haskell framework for (parallel) monte-carlo simulations.
What is it?
Carbon is an open-source, Haskell framework aiming to provide easy access to parallel Monte Carlo simulations by providing a simple, but powerful compositional method for building simulations, high-level functions for running them, and efficient data structures for aggregating common data types.
Examples of simulations written in Carbon can be found at: https://github.com/icasperzen/hs-carbon-examples
Carbon allows users to quickly describe Monte Carlo simulations using monadic composition (similar to parser combinators).
For example, starting from a Carbon primitive action,
randomR, we can build a simulation which tests for membership in the unit circle.
mcSquareD :: RandomGen g => MonteCarlo g (Double,Double) mcSquareD = liftM2 (,) (randomR (-1,1)) (randomR (-1,1)) inUnitCircle :: RandomGen g => MonteCarlo g Bool inUnitCircle = do (x,y) <- mcSquareD return $ x*x + y*y <= 1
mcSquareD, which returns a random 2D coordinate in the unit square, from
Next, we build
mcSquareD, returning a
Bool if the sampled 2D point lies within the unit circle.
High Level Function
Carbon aims to capture common usage patterns in high level "skeletons". Currently, just one is provided (and its sequential equivalent):
experimentP :: (R.RandomGen g, Result s) => MonteCarlo g (Obs s) -> Int -> Int -> g -> s
For example, using
inUnitCircle, we can compute the value of pi.
let res = experimentP inUnitCircle noRuns chSize gen pi = 4 * sampleMean res
Descriptions of simulations should be completely independent of the results desired.
Carbon enforces this separation of concerns by using type families.
Every simulation will require a type annotation describing the desired output.
The particular instance of the type family describes how data aggregation is to occur (what to do with each observation).
For example, the following two simulations use the same
MonteCarlo action, but deliver very different results.
let bl = experimentP inUnitCircle noRuns chnkSize gen :: [Bool] let bs = experimentP inUnitCircle noRuns chnkSize gen :: BoolSumm
[Bool] is a list containing every observation from the simulation.
This is inefficient, however, as it is rarely necessary to maintain each and every observation in memory.
Instead, we can fold it into a clever data structure.
BoolSumm is a space-efficient representation of binary observations.
Result typeclass captures this functionality.
Carbon provides functionality for computing basic statistics on some instances of
Summary typeclass is meant to capture some of the common statistical measures one might be interested in.
Creating new instances of
Summary is easy; and extra functions can be written for a
Result instance to capture even more complex statistical functionality.
Version 0.1.0 released!
- support for parallelism (using GpH Strategies)
- high-level function for describing a simulation
- efficient data aggregation techniques