NEWS.md

In majkamichal/naivebayes: High Performance Implementation of the Naive Bayes Algorithm

naivebayes 1.0.0

Major Release: Maturity and Stability

The package has reached a significant milestone of maturity and stability, leading to the version update to 1.0.0.

• Improvement: enhanced print methods.
• Improvement: updated documentation.
• Improvement: minor internal enhancements.

naivebayes 0.9.7

• Improvement: multinomial_naive_bayes(), bernoulli_naive_bayes(), poisson_naive_bayes() and gaussian_naive_bayes() now support sparse matrices (dgCMatrix class from the Matrix Package).

• Improvement: updated documentation.

• Improvement: better informative errors.

naivebayes 0.9.6

Improvements:

• Enhanced documentation - this includes a new webpage: https://majkamichal.github.io/naivebayes/

• In naive_bayes() Poisson distribution is now available to model class conditional probabilities of non-negative integer predictors. It is applied to all vectors with class "integer" via a new parameter usepoisson = TRUE. By default usepoisson = FALSE. All naive_bayes objects created with previous versions are fully compatible with the 0.9.6 version.

• predict.naive_bayes() has a new parameter eps that specifies a value of an epsilon-range to replace zero or close to zero probabilities by specified threshold. It applies to metric variables as well as to discrete variables, but only when laplace = 0.

• predict.naive_bayes() is now more efficient and reliable.

• print() method has been enhanced for better readability.

• plot() method allows now visualising class marginal and class conditional distributions for each predictor variable via new parameter prob with two possible values: "marginal" or "conditional".

New functions

• bernoulli_naive_bayes() - specialised version of naive_bayes(), where all features take on 0-1 values and each feature is modelled with the Bernoulli distribution.

• gaussian_naive_bayes() - specialised version of naive_bayes(), where all features are real valued and each feature is modelled with the Gaussian distribution.

• poisson_naive_bayes() - specialised version of naive_bayes(), where all features are non-negative integers and each feature is modelled with the Poisson distribution.

• nonparametric_naive_bayes() - specialised version of naive_bayes(), where all features are real valued and distribution of each is estimated with kernel density estimation (KDE).

• multinomial_naive_bayes() - specialised Naive Bayes classifier suitable for text classification.

• %class% and %prob% - infix operators that are shorthands for performing classification and obtaining posterior probabilities, respectively.

• coef() - a generic function which extracts model coefficients from specialized Naive Bayes objects.

• get_cond_dist() - for obtaining names of class conditional distributions assigned to features.

naivebayes 0.9.5

• Fixed: when laplace > 0 and discrete feature with >2 distinct values, the probabilities in the probability table do not sum up to 1.

naivebayes 0.9.4

• Fixed: plot.naive_bayes() crashes when missing data present in training set (bug found by Mark van der Loo).

naivebayes 0.9.3

• Fixed: numerical underflow in predict.naive_bayes function when the number of features is big (bug found by William Townes).

• Fixed: when all names of features in the newdata in predict.naive_bayes() do not match these defined in the naive_bayes object, then the calculation based on prior probabilities is done only for one row of newdata.

• Improvement: better handling (informative warnings/errors) of not correct inputs in predict.naive_bayes().

• Improvement: print.naive_bayes() is now more transparent.

naivebayes 0.9.2

• Fixed: when the data have two classes and they are not alphabetically ordered, the predicted classes are incorrect (bug found by Max Kuhn).

naivebayes 0.9.1

• Fixed: when the prediction data has one row, the column names get dropped (bug found by Max Kuhn).

majkamichal/naivebayes documentation built on March 26, 2024, 8:44 p.m.