FuzzyClass 0.3.0

R6 classes

• All classifiers are now implemented as R6 classes that inherit from the new base class FuzzyClassifier. Each classifier has a generator named after its function with the suffix Model (e.g. FuzzyGaussianNaiveBayesModel, GauNBFuzzyParamModel, DWFuzzyGammaNaiveBayesModel) and the methods $new(), $fit(train, cl), $predict(newdata, type), $set_params(...), $is_fitted() and $print(). Calls can be chained: FuzzyGammaNaiveBayesModel$new()$fit(x, y)$predict(x_new).
• Training now reports its progress with an informative message (classifier, number of observations, predictors and classes, and the training time). It can be silenced with verbose = FALSE or globally with options(FuzzyClass.verbose = FALSE) (see fuzzy_verbose()).
• The functional interface is unchanged: FuzzyGaussianNaiveBayes(train, cl) and the other functions train and return the R6 object, and predict(fit, newdata) / predict(fit, newdata, type = "matrix") keep working. type = "prob" is now accepted as a synonym of "matrix". Predictions are identical to version 0.2.0.
• The estimated quantities are now stored in fit$estimates (they were the elements of the S3 list returned in previous versions), together with fit$params, fit$predictors, fit$classes, fit$n_obs and fit$fit_time.
• Informative error messages (powered by cli) for untrained models, train/cl with different sizes and missing required arguments (wdelta, weta).

tidymodels

• New parsnip model fuzzy_naive_bayes(), so every classifier can be used with parsnip, workflows, recipes, rsample, tune and yardstick. The classifier is chosen by the engine ("gaussian", "gamma", "beta", "exponential", "binomial", "poisson", "geometric", "trapezoidal", "triangular", "hypergeometric", "bayes_rule", "naive", "dw_gamma", "dw_hypergeometric", "gaussian_fuzzy_param", "exponential_fuzzy_param" and "poisson_fuzzy_param").
• The main arguments fuzzy, alpha_cut and ranking_method (metd) can be tuned with tune(); their dials parameter objects are fuzzy_membership(), alpha_cut() and ranking_method().
• parsnip, dials and the other tidymodels packages are only suggested: the model is registered when parsnip is loaded.

Other changes

• Requires R >= 4.1.0. New imports: R6 and cli.
• The tests now use the standard testthat structure (tests/testthat/), so they run with R CMD check and devtools::test().

FuzzyClass 0.2.0

• Added new method for fuzzy number ordering as metd = 5 in functions GauNBFuzzyParam, ExpNBFuzzyParam, and PoiNBFuzzyParam.
• Significant performance optimization across all classifiers: vectorized predictions using native matrix operations and max.col for substantial speedups.
• Fixed discrete variables validation (verifyNumbersFunction) ensuring compatibility with R 4.2+ and resolving missing argument errors in DWFuzzyHipergeometricNaiveBayes, FuzzyBinomialNaiveBayes, FuzzyHipergeometricNaiveBayes, and FuzzyPoissonNaiveBayes.
• Fixed mathematical consistency in FuzzyHipergeometricNaiveBayes by applying joint membership once to the class score, consistent with other models.
• Fixed prior class probability multiplication in FuzzyGeoNaiveBayes applied once per class.
• Cleaned up package dependencies: removed unused dependencies (dplyr, tibble, tidyr, rlang, purrr) from Imports.
• Fixed class/label misalignment: parameters were estimated in the order the classes appear in the data while predictions were labelled with the sorted factor levels, so predictions were wrong whenever the training data were not sorted by class. All classifiers now use the factor levels (get_classes()) for estimation and prediction, and the columns of type = "matrix" follow the levels.
• FuzzyGammaNaiveBayes / DWFuzzyGammaNaiveBayes: exact Gamma maximum likelihood estimates (the previous optimisation was bounded by max(x), which truncated the shape/scale).
• FuzzyBetaNaiveBayes: removed the artificial upper bound (10) on the Beta shapes; the method of moments is used as starting value.
• FuzzyGeoNaiveBayes: probability mass function consistent with the estimator (dgeom, support 0, 1, 2, …); memberships use the same intervals in training and prediction, empty intervals (and values outside the training range) receive 0.001.
• FuzzyBinomialNaiveBayes: closed-form method of moments with n >= max(x).
• GauNBFuzzyParam / ExpNBFuzzyParam: confidence intervals (alpha-cuts) use the class sample size instead of the total sample size. ExpNBFuzzyParam ranks densities (as the Gaussian and Poisson versions), since the ranking methods assume non-negative fuzzy numbers.
• PoiNBFuzzyParam: alpha-cuts are now nested (they shrink as alpha increases).
• Ranking methods (metd = 1..5) vectorized; the closed form of the Q distance replaces one eigen decomposition per observation. doParallel, foreach and parallel are no longer needed; cores is kept for backward compatibility.
• FuzzyTriangularNaiveBayes / FuzzyTrapezoidalNaiveBayes: densities outside the support are floored at 1e-10, avoiding ties decided by class order.
• FuzzyNaiveBayes: the arguments m and Pi are now passed to the estimators (they were ignored).

FuzzyClass 0.1.7

• Add the “Fuzzy Hipergeometric Naive Bayes classifier” as FuzzyHipergeometricNaiveBayes
• Add the “Double Weighted Fuzzy Hipergeometric Naive Bayes classifier” as DWFuzzyHipergeometricNaiveBayes
• Fix: memberships functions. Now they work correctly with dataframes

FuzzyClass 0.1.6

• Add the “Fuzzy Rule-based System” with some functions
• Fix: FuzzyNaiveBayes to one column in the data
• Fix: “membership” to dataframe with one column in the data
• Fix: “Frequency” to discrete distributions

FuzzyClass 0.1.5

• Add the “double weighted fuzzy gamma naive bayes classifier” as DWFuzzyGammaNaiveBayes
• Add the “geometric fuzzy gamma naive bayes classifier” as GeometricFuzzyGammaNaiveBayes
• Add the “Fuzzy Bayes Rule classifier” as FuzzyBayesRule
• Remove the ordering method for functions with fuzzy parameters metd=4.

FuzzyClass 0.1.4

• Added a new ordering method for functions with fuzzy parameters: GauNBFuzzyParam, ExpNBFuzzyParam,GamNBFuzzyParam and PoiNBFuzzyParam. The form can be used considering metd = 4.

FuzzyClass 0.1.3

• Update in parameters estimation in function FuzzyTriangularNaiveBayes.
• Update predict has been optimized to functions FuzzyGammaNaiveBayes.
• Update functions of GamNBFuzzyParam and GauNBFuzzyParam.
• predict function has been optimized to FuzzyGammaNaiveBayes, FuzzyBetaNaiveBayes, FuzzyBinomialNaiveBayes,FuzzyExponencialNaiveBayes,FuzzyTriangularNaiveBayes,FuzzyPoissonNaiveBayes

FuzzyClass 0.1.2

New features

• Changed \() by function(.) in function FuzzyNaiveBayes, line 262.
• Changed description in FuzzyClass Documentation

FuzzyClass 0.1.1

New features

• Estimation new form in function FuzzyGammaNaiveBayes.