NEWS | R Documentation |
News for Package naivebayes
Changes in version 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.
Changes in version 0.9.7
Improvement:
multinomial_naive_bayes()
,bernoulli_naive_bayes()
,poisson_naive_bayes()
andgaussian_naive_bayes()
now support sparse matrices (dgCMatrix
class from theMatrix
Package).Improvement: updated documentation.
Improvement: better informative errors.
Changes in version 0.9.6
Improvements:
Enhanced documentation - this includes a new webpage: https://majkamichal.github.io/naivebayes/
-
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 parameterusepoisson = TRUE
innaive_bayes
function. By defaultusepoisson = FALSE
. Allnaive_bayes
objects created with previous versions are fully compatible with the 0.9.6 version. -
predict.naive_bayes()
has new parametereps
that specifies a value of an epsilon-range to replace zero or close to zero probabilities by specified threshold. It applies to metric variables. -
predict.naive_bayes()
is now more efficient and more 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 parameterprob
with two possible values: "marginal" or "conditional".
New functions:
-
bernoulli_naive_bayes()
- specialised version of thenaive_bayes()
, where all features take on 0-1 values and each feature is modelled with the Bernoulli distribution. -
gaussian_naive_bayes()
- specialised version of thenaive_bayes()
, where all features are real valued and each feature is modelled with the Gaussian distribution. -
poisson_naive_bayes()
- specialised version of thenaive_bayes()
, where all features take are non-negative integers and each feature is modelled with the Poisson distribution. -
nonparametric_naive_bayes()
- specialised version of thenaive_bayes()
, where all features take 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.
Changes in version 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.
Changes in version 0.9.4
Fixed: plot crashes when missing data present in training set (bug found by Mark van der Loo).
Changes in version 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
inpredict.naive_bayes
function do not match these defined in the naive_bayes object, then the calculation based on prior probabilities is done only for one row ofnewdata
.Improvement: better handling (informative warnings/errors) of not correct inputs in 'predict.naive_bayes' function.
Improvement:
print.naive_bayes
fits now the console width.
Changes in version 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).
Changes in version 0.9.1
Fixed: when the prediction data has one row, the column names get dropped (bug found by Max Kuhn).