Package 'learningmachine'

Title: Machine Learning with Explanations and Uncertainty Quantification
Description: Regression-based Machine Learning with explanations and uncertainty quantification.
Authors: T. Moudiki
Maintainer: T. Moudiki <[email protected]>
License: BSD_3_clause + file LICENSE
Version: 2.6.0
Built: 2024-09-20 16:21:07 UTC
Source: https://github.com/Techtonique/learningmachine

Help Index

'Base' class

Description

the 'Base' class used by other objects; useful for extensions of the package, not for basic interactions with the package

Methods

Public methods

Method new()

Create a new object.

Usage
Base$new(
  name = "Base",
  type = "none",
  model = NULL,
  method = NULL,
  X_train = NULL,
  y_train = NULL,
  pi_method = c("none", "splitconformal", "kdesplitconformal", "bootsplitconformal",
    "jackknifeplus", "kdejackknifeplus", "bootjackknifeplus", "surrsplitconformal",
    "surrjackknifeplus"),
  level = 95,
  B = 100,
  nb_hidden = 0,
  nodes_sim = c("sobol", "halton", "unif"),
  activ = c("relu", "sigmoid", "tanh", "leakyrelu", "elu", "linear"),
  engine = NULL,
  params = NULL,
  seed = 123
)
Arguments
name

name of the class

type

type of supervised learning method implemented

model

fitted model

method

supevised learning method

X_train

training set features

y_train

training set response

pi_method

type of prediction interval in c("splitconformal", "kdesplitconformal", "bootsplitconformal", "jackknifeplus", "kdejackknifeplus", "bootjackknifeplus", "surrsplitconformal", "surrjackknifeplus")

level

an integer; the level of confidence

B

an integer; the number of simulations when level is not NULL

nb_hidden

number of nodes in the hidden layer, for construction of a quasi- randomized network

nodes_sim

type of 'simulations' for hidden nodes, if nb_hidden > 0; takes values in c("sobol", "halton", "unif")

activ

activation function's name for the hidden layer, in the construction of a quasi-randomized network; takes values in c("relu", "sigmoid", "tanh", " leakyrelu", "elu", "linear")

engine

contains fit and predict lower-level methods for the given method; do not modify by hand

params

additional parameters passed to method when calling fit

seed

an integer; reproducibility seed for methods that include randomization

Returns

A new 'Base' object.

Method get_name()

Usage
Base$get_name()

Method get_type()

Usage
Base$get_type()

Method get_model()

Usage
Base$get_model()

Method set_model()

Usage
Base$set_model(model)

Method get_method()

Usage
Base$get_method()

Method set_method()

Usage
Base$set_method(method)

Method get_pi_method()

Usage
Base$get_pi_method()

Method set_pi_method()

Usage
Base$set_pi_method(pi_method)

Method get_level()

Usage
Base$get_level()

Method set_level()

Usage
Base$set_level(level)

Method get_B()

Usage
Base$get_B()

Method set_B()

Usage
Base$set_B(B)

Method get_nb_hidden()

Usage
Base$get_nb_hidden()

Method set_nb_hidden()

Usage
Base$set_nb_hidden(nb_hidden)

Method get_nodes_sim()

Usage
Base$get_nodes_sim()

Method set_nodes_sim()

Usage
Base$set_nodes_sim(nodes_sim)

Method get_activ()

Usage
Base$get_activ()

Method set_activ()

Usage
Base$set_activ(activ)

Method set_engine()

Usage
Base$set_engine(engine)

Method get_engine()

Usage
Base$get_engine()

Method get_params()

Usage
Base$get_params()

Method get_seed()

Usage
Base$get_seed()

Method set_seed()

Usage
Base$set_seed(seed)

Method summary()

Usage
Base$summary(
  X,
  show_progress = TRUE,
  class_name = NULL,
  class_index = NULL,
  y = NULL,
  type_ci = c("student", "nonparametric", "bootstrap"),
  cl = NULL
)

Method clone()

The objects of this class are cloneable with this method.

Usage
Base$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

'Classifier' class

Description

The 'Classifier' class contains supervised classification models

Details

This class implements models:

lm

Linear model

bcn

see https://www.researchgate.net/publication/380760578_Boosted_Configuration_neural_Networks_for_supervised_classification

extratrees

Extremely Randomized Trees; see https://link.springer.com/article/10.1007/s10994-006-6226-1

glmnet

Elastic Net Regression; see https://glmnet.stanford.edu/

krr

Kernel Ridge Regression; see for example https://www.jstatsoft.org/article/view/v079i03

(but the implementation is different)

ranger

Random Forest; see https://www.jstatsoft.org/article/view/v077i01

ridge

Ridge regression; see https://arxiv.org/pdf/1509.09169

xgboost

a scalable tree boosting system see https://arxiv.org/abs/1603.02754

rvfl

Random Vector Functional Network, see https://www.researchgate.net/publication/332292006_Online_Bayesian_Quasi-Random_functional_link_networks_application_to_the_optimization_of_black_box_functions

Super class

learningmachine::Base -> Classifier

Public fields

name

name of the class

type

type of supervised learning method implemented

model

fitted model

method

supervised learning method in c('lm', 'ranger', 'extratrees', 'ridge', 'bcn', 'glmnet', 'krr', 'xgboost')

X_train

training set features; do not modify by hand

y_train

training set response; do not modify by hand

pi_method

type of prediction set in c("splitconformal", "kdesplitconformal", "bootsplitconformal", "surrsplitconformal")

level

an integer; the level of confidence (default is 95, for 95 per cent) for prediction sets

type_prediction_set

a string; the type of prediction set (currently, only "score" method)

B

an integer; the number of simulations when level is not NULL

nb_hidden

number of nodes in the hidden layer, for construction of a quasi- randomized network

nodes_sim

type of 'simulations' for hidden nodes, if nb_hidden > 0; takes values in c("sobol", "halton", "unif")

activ

activation function's name for the hidden layer, in the construction of a quasi-randomized network; takes values in c("relu", "sigmoid", "tanh", " leakyrelu", "elu", "linear")

engine

contains fit and predic lower-level methods for the given method; do not modify by hand

params

additional parameters passed to method when calling fit do not modify by hand

seed

an integer; reproducibility seed for methods that include randomization

Methods

Public methods

Inherited methods

Method new()

Create a new object.

Usage
Classifier$new(
  name = "Classifier",
  type = "classification",
  model = NULL,
  method = NULL,
  X_train = NULL,
  y_train = NULL,
  pi_method = c("none", "kdesplitconformal", "bootsplitconformal", "surrsplitconformal"),
  level = 95,
  type_prediction_set = c("none", "score"),
  B = 100,
  nb_hidden = 0,
  nodes_sim = c("sobol", "halton", "unif"),
  activ = c("relu", "sigmoid", "tanh", "leakyrelu", "elu", "linear"),
  engine = NULL,
  params = NULL,
  seed = 123
)
Returns

A new 'Classifier' object.

Method get_type_prediction_set()

Usage
Classifier$get_type_prediction_set()

Method set_type_prediction_set()

Usage
Classifier$set_type_prediction_set(type_prediction_set)

Method fit()

Fit model to training set

Usage
Classifier$fit(X, y, ...)
Arguments
X

a matrix of covariates (i.e explanatory variables)

y

a vector, the response (i.e variable to be explained)

...

additional parameters to learning algorithm (see vignettes)

Method predict_proba()

Usage
Classifier$predict_proba(X)

Method predict()

Usage
Classifier$predict(X)

Method clone()

The objects of this class are cloneable with this method.

Usage
Classifier$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.

'Regressor' class

Description

The 'Regressor' class contains supervised regression models

Details

This class implements models:

lm

Linear model

bcn

see https://www.researchgate.net/publication/380760578_Boosted_Configuration_neural_Networks_for_supervised_classification

extratrees

Extremely Randomized Trees; see https://link.springer.com/article/10.1007/s10994-006-6226-1

glmnet

Elastic Net Regression; see https://glmnet.stanford.edu/

krr

Kernel Ridge Regression; see for example https://www.jstatsoft.org/article/view/v079i03

(but the implementation is different)

ranger

Random Forest; see https://www.jstatsoft.org/article/view/v077i01

ridge

Ridge regression; see https://arxiv.org/pdf/1509.09169

xgboost

a scalable tree boosting system see https://arxiv.org/abs/1603.02754

svm

Support Vector Machines, see https://cran.r-project.org/web/packages/e1071/vignettes/svmdoc.pdf

rvfl

Random Vector Functional Network, see https://www.researchgate.net/publication/332292006_Online_Bayesian_Quasi-Random_functional_link_networks_application_to_the_optimization_of_black_box_functions

Super class

learningmachine::Base -> Regressor

Public fields

name

name of the class

type

type of supervised learning method implemented

model

fitted model

method

supervised learning method in c('lm', 'ranger', 'extratrees', 'ridge', 'bcn', 'glmnet', 'krr', 'xgboost', 'svm')

X_train

training set features; do not modify by hand

y_train

training set response; do not modify by hand

pi_method

type of prediction interval in c("splitconformal", "kdesplitconformal", "bootsplitconformal", "jackknifeplus", "kdejackknifeplus", "bootjackknifeplus", "surrsplitconformal", "surrjackknifeplus")

level

an integer; the level of confidence (default is 95, for 95 per cent) for prediction intervals

B

an integer; the number of simulations when 'level' is not NULL

nb_hidden

number of nodes in the hidden layer, for construction of a quasi- randomized network

nodes_sim

type of 'simulations' for hidden nodes, if nb_hidden > 0; takes values in c("sobol", "halton", "unif")

activ

activation function's name for the hidden layer, in the construction of a quasi-randomized network; takes values in c("relu", "sigmoid", "tanh", " leakyrelu", "elu", "linear")

engine

contains fit and predic lower-level methods for the given method; do not modify by hand

params

additional parameters passed to method when calling fit do not modify by hand

seed

an integer; reproducibility seed for methods that include randomization

Methods

Public methods

Inherited methods

Method new()

Create a new object.

Usage
Regressor$new(
  name = "Regressor",
  type = "regression",
  model = NULL,
  method = NULL,
  X_train = NULL,
  y_train = NULL,
  pi_method = c("none", "splitconformal", "jackknifeplus", "kdesplitconformal",
    "bootsplitconformal", "kdejackknifeplus", "bootjackknifeplus", "surrsplitconformal",
    "surrjackknifeplus"),
  level = 95,
  B = 100,
  nb_hidden = 0,
  nodes_sim = c("sobol", "halton", "unif"),
  activ = c("relu", "sigmoid", "tanh", "leakyrelu", "elu", "linear"),
  engine = NULL,
  params = NULL,
  seed = 123
)
Returns

A new 'Regressor' object.

Method fit()

Fit model to training set

Usage
Regressor$fit(X, y, type_split = c("stratify", "sequential"), ...)
Arguments
X

a matrix of covariates (i.e explanatory variables)

y

a vector, the response (i.e variable to be explained)

type_split

type of data splitting for split conformal prediction: "stratify" (for classical supervised learning) "sequential" (when the data sequential ordering matters)

...

additional parameters to learning algorithm (see vignettes)

Method predict()

Predict model on test set

Usage
Regressor$predict(X, ...)
Arguments
X

a matrix of covariates (i.e explanatory variables)

...

additional parameters

Method fit_predict()

Fit model to training set and predict on test set

Usage
Regressor$fit_predict(
  X,
  y,
  pct_train = 0.8,
  score = ifelse(is.factor(y), yes = function(preds, y_test) mean(preds == y_test), no =
    function(preds, y_test) sqrt(mean((preds - y_test)^2))),
  level = NULL,
  pi_method = c("none", "splitconformal", "jackknifeplus", "kdesplitconformal",
    "bootsplitconformal", "kdejackknifeplus", "bootjackknifeplus", "surrsplitconformal",
    "surrjackknifeplus"),
  B = 100,
  seed = 123,
  graph = FALSE,
  ...
)

Method update()

update model in an online fashion (for now, only implemented for 'rvfl' models")

Usage
Regressor$update(newx, newy, ...)
Arguments
newx

a vector of new covariates (i.e explanatory variables)

newy

a numeric, the new response's observation (i.e variable to be explained)

...

additional parameters to be passed to the underlying model's method 'update'

Method clone()

The objects of this class are cloneable with this method.

Usage
Regressor$clone(deep = FALSE)
Arguments
deep

Whether to make a deep clone.