FunctionalModel.fit.transformed: Fit the Given Model Blueprint to the Specified Data

In thomasWeise/regressoR.functional: An R Package for Fitting Functional Models to 2-Dimensional Data

Description

Fit the specified model to the given data. First, we generate a starting guess about the parameterization via FunctionalModel.par.estimate (or accept it via the parameter par). From then on, we apply the different function fitters one by one. All the fitters who have not produced the current best solution are applied again, to the now-best guess. However, we do not apply the fitters that have produced that very guess in the next round. (They may get a chance again in a later turn.) Anyway, this procedure is iterated until no improvement can be made anymore. After finishing the fitting, we attempt whether rounding the fitted parameters to integers can improve the fitting quality.

Usage

FunctionalModel.fit.transformed(metric, model, transformation.x = NULL,
  transformation.y = NULL, metric.transformed = NULL, par = NULL,
  q = 0.75, fitter = FunctionalModel.fit)

Arguments

metric	an instance of RegressionQualityMetric
model	an instance of FunctionalModel
transformation.x	the transformation along the x-axis, or NULL if none was applied to the data
transformation.y	the transformation along the y-axis, or NULL if none was applied to the data
metric.transformed	the transformed metric for the first fitting step
par	the initial starting point
q	the effort to spent in learning, a value between 0 (min) and 1 (max). Higher values may lead to much more computational time, lower values to potentially lower result quality.
fitter	the model fitter to use

Value

On success, an instance of FittedFunctionalModel. NULL on failure.

Examples

set.seed(234345L)

orig.f.x.par <- function(x, par) exp(par[1] + par[2]*x - par[3]*x*x);
orig.par <- c(0.2, -0.3, 0.4);
orig.f.x <- function(x) orig.f.x.par(x, orig.par);
orig.x <- (-100:100)*0.05;
orig.y <- orig.f.x(orig.x);
noisy.x <- rnorm(n=length(orig.x), mean=orig.x, sd=0.05);
noisy.y <- rnorm(n=length(orig.y), mean=orig.y, sd=0.05*orig.y);

transformed.data <- dataTransformeR::TransformedData2D.new(
  dataTransformeR::Transformation.normalize(noisy.x),
  dataTransformeR::Transformation.log(noisy.y));

metric <- regressoR.quality::RegressionQualityMetric.default(noisy.x, noisy.y);
metric.transformed <- regressoR.quality::RegressionQualityMetric.default(
                                                 transformed.data@x@data,
                                                 transformed.data@y@data);
model <- regressoR.functional.models::FunctionalModel.quadratic();
result <- FunctionalModel.fit.transformed(metric, model,
                                transformed.data@x@transformation,
                                transformed.data@y@transformation,
                                metric.transformed);

result.2 <- learnerSelectoR::learning.Result.finalize(result)
plot(noisy.x, noisy.y)
lines(noisy.x, result.2@f(noisy.x), col="red")

thomasWeise/regressoR.functional documentation built on May 10, 2019, 10:24 a.m.