Copyright	(c) Fabricio Olivetti de Franca 2022
License	GPL-3
Maintainer	fabricio.olivetti@gmail.com
Stability	experimental
Portability	POSIX
Safe Haskell	Safe-Inferred

MachineLearning.TIR

Description

The TIR expression represents a function of the form:

\[ f(x) = g(\sum_{i}{w_i \cdot t_i(\prod_{j}{x_j^{k_{ij}})}} / (1 + \sum_{i}{w_i \cdot t_i(\prod_{j}{x_j^{k_{ij}})}}) \]

with \(t_i\) being a transformation function, \(g\) an invertible function, \(w_i\) a linear coefficient, \(k_{ij}\) the interaction strength.

Any given expression can be represented by two lists of terms, with each term being composed of a transformatioon function and an interaction. The transformation function is represented by a Function sum type. The interaction is represented as a list of tuple of ints where the key is the predictor index and the value is the strength of the predictor in this term. Strengths with a value of zero are omitted.

Synopsis

data TIR = TIR {
- _funY :: Function
- _p :: Sigma
- _q :: Sigma
}
type Sigma = [Pi]
type Pi = (Double, Function, [(Int, Int)])
randomRng :: (Int, Int) -> Rnd Int
randomRngNZ :: (Int, Int) -> Rnd Int
randomFrom :: [a] -> Rnd a
randomVar :: MutationCfg -> Rnd (Maybe Int, MutationCfg)
randomVars :: MutationCfg -> Rnd [(Int, Int)]
randomPi :: MutationCfg -> Rnd (Maybe Pi)
randomSigma :: MutationCfg -> Int -> Rnd (Sigma, Int)
randomTIR :: MutationCfg -> Rnd TIR
type Column a = Vector a
type Dataset a = Vector (Column a)
type Constraint = SRTree Int Double -> Double
data Individual = Individual {
- _chromo :: TIR
- _fit :: [Double]
- _weights :: [Vector Double]
- _constr :: Double
- _len :: Int
- _penalty :: Double
}
createIndividual :: TIR -> Individual
penalizedFit :: Individual -> Double
replaceConsts :: TIR -> Vector Double -> TIR
replaceWeight :: Pi -> State [Double] Pi
assembleTree :: Double -> TIR -> SRTree Int Double
prettyPrintsolution :: Individual -> String

Documentation

data TIR Source #

TIR is a record type composed of the external function _funY of type Function, numerator _p and denominator _q of type Sigma

Constructors

TIR
Fields _funY :: Function _p :: Sigma _q :: Sigma

Instances

Instances details

Show TIR Source #
Instance details Defined in MachineLearning.TIR Methods showsPrec :: Int -> TIR -> ShowS show :: TIR -> String showList :: [TIR] -> ShowS
NFData TIR Source #
Instance details Defined in MachineLearning.TIR Methods rnf :: TIR -> ()

type Sigma = [Pi] Source #

Sigma is just a list of terms Pi

type Pi = (Double, Function, [(Int, Int)]) Source #

Pi is a triple composed of a coefficient, a Function and a list of tuples where `(ix, k)` represents `x ! ix ^ k`.

randomRng :: (Int, Int) -> Rnd Int Source #

generates a random integer within the specified range.

randomRngNZ :: (Int, Int) -> Rnd Int Source #

generates a random integer within the specified range excluding zero.

randomFrom :: [a] -> Rnd a Source #

picks a random element from a list.

randomVar :: MutationCfg -> Rnd (Maybe Int, MutationCfg) Source #

returns a random index of variables provided by the mutation configuration.

randomVars :: MutationCfg -> Rnd [(Int, Int)] Source #

returns a list of random interactions (tuples of variables indeces and exponentes) with parameters provided by the mutation configuration.

randomPi :: MutationCfg -> Rnd (Maybe Pi) Source #

returns a random Pi

randomSigma :: MutationCfg -> Int -> Rnd (Sigma, Int) Source #

returns a random Sigma

randomTIR :: MutationCfg -> Rnd TIR Source #

returns a random TIR expression

type Column a = Vector a Source #

We store thee dataset as a vector of columns. Each vector is stored a Storable-based vector.

type Dataset a = Vector (Column a) Source #

A dataset is a Vector of Column

type Constraint = SRTree Int Double -> Double Source #

A constraint is a function that gets a symbolic tree as an input and returns non negative Double representing how much a constraint was violated.

data Individual Source #

An individual in the population is composed of the chromossome, a vector of fitness, a list of coefficients (for multiclass problems it stores one vector of coefficient per class), the constraint violation, the size of the expression, and the penalty value.

Constructors

Individual
Fields _chromo :: TIR _fit :: [Double] _weights :: [Vector Double] _constr :: Double _len :: Int _penalty :: Double

Instances

Instances details

NFData Individual Source #
Instance details Defined in MachineLearning.TIR Methods rnf :: Individual -> ()
Solution Individual Source #
Instance details Defined in MachineLearning.TIR Methods _getFitness :: Individual -> Double _isFeasible :: Individual -> Bool
Eq Individual Source #
Instance details Defined in MachineLearning.TIR Methods (==) :: Individual -> Individual -> Bool (/=) :: Individual -> Individual -> Bool
Ord Individual Source #
Instance details Defined in MachineLearning.TIR Methods compare :: Individual -> Individual -> Ordering (<) :: Individual -> Individual -> Bool (<=) :: Individual -> Individual -> Bool (>) :: Individual -> Individual -> Bool (>=) :: Individual -> Individual -> Bool max :: Individual -> Individual -> Individual min :: Individual -> Individual -> Individual

createIndividual :: TIR -> Individual Source #

creates an unevaluated individual.

penalizedFit :: Individual -> Double Source #

calculates the penalized fitness.

replaceConsts :: TIR -> Vector Double -> TIR Source #

replaces the coefficients of a TIR expression

replaceWeight :: Pi -> State [Double] Pi Source #

assembleTree :: Double -> TIR -> SRTree Int Double Source #

creates a symbolic tree from a TIR expression.

prettyPrintsolution :: Individual -> String Source #

pretty print a solution.