| Title: | Extensible Package for Parallel, Batch Training of Base Learners for Ensemble Modeling |
|---|---|
| Description: | Extensible S4 classes and methods for batch training of regression and classification algorithms such as Random Forest, Gradient Boosting Machine, Neural Network, Support Vector Machines, K-Nearest Neighbors, Penalized Regression (L1/L2), and Bayesian Additive Regression Trees. These algorithms constitute a set of 'base learners', which can subsequently be combined together to form ensemble predictions. This package provides cross-validation wrappers to allow for downstream application of ensemble integration techniques, including best-error selection. All base learner estimation objects are retained, allowing for repeated prediction calls without the need for re-training. For large problems, an option is provided to save estimation objects to disk, along with prediction methods that utilize these objects. This allows users to train and predict with large ensembles of base learners without being constrained by system RAM. |
| Authors: | Alireza S. Mahani, Mansour T.A. Sharabiani |
| Maintainer: | Alireza S. Mahani <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 1.0.2 |
| Built: | 2024-11-20 06:40:31 UTC |
| Source: | CRAN |
"KNN.Regression.Config", "NNET.Regression.Config", "RF.Regression.Config", "SVM.Regression.Config", "GBM.Regression.Config", "PENREG.Regression.Config", "BART.Regression.Config"
These base learner configuration objects contain tuning parameters needed for training base learner algorithms. Names are identical to those used in implementation packages. See documentation for those packages for detailed definitions.
These objects are typically constructed via calls to make.configs and make.instances.
For KNN.Regression.Config:
kernel:Object of class "character", defining the weighting function applied to neighbors as a function of distance from target point. Options include "rectangular", "epanechnikov", "triweight", and "gaussian".
k:Object of class "numeric", defining the number of nearest neighbors to include in prediction for each target point.
For NNET.Regression.Config:
decay:Object of class "numeric", defining the weight decay parameter.
size:Object of class "numeric", defining the number of hidden-layer neurons.
maxit:Object of class "numeric", defining the maximum number of iterations in the training.
For RF.Regression.Config:
ntree:Object of class "numeric", defining the number of trees in the random forest.
nodesize:Object of class "numeric", defining the minimum size of terminal nodes.
mtry.mult:Object of class "numeric", defining the multiplier of the default value for mtry parameter in the randomForest function call.
For SVM.Regression.Config:
cost:Object of class "numeric", defining the cost of constraint violation.
epsilon:Object of class "numeric", the parameter of insensitive-loss function.
kernel:Object of class "character", the kernel used in SVM training and prediction. Options include "linear", "polynomial", "radial", and "sigmoid".
For GBM.Regression.Config:
n.trees:Object of class "numeric", defining the number of trees to fit.
interaction.depth:Object of class "numeric", defining th maximum depth of variable interactions.
Object of class "numeric", defining the shrinkage parameter applied to each tree in expansion.
bag.fraction:Object of class "numeric", defining the fraction of training set observations randomly selected to propose the next tree in the expansion.
For PENREG.Regression.Config:
alpha:Object of class "numeric", defining the mix of L1 and L2 penalty. Must be between 0.0 and 1.0.
lambda:Object of class "numeric", defining the shrinkage parameter. Must be non-negative.
For BART.Regression.Config:
num_trees:Object of class "numeric", defining the number of trees to be grown in the sum-of-trees model. Must be a positive integer.
k:Object of class "numeric", controlling the degree of shrinkage and hence conservativeness of the fit. Must be positive.
q:Object of class "numeric", defining quantile of the prior on the error variance at which the data-based estimate is placed. Higher values of this parameter lead to a more aggressive fit.
nu:Object of class "numeric", defining degrees of freedom for the inverse chi-squared prior. Must be a positive integer.
Class "Regression.Config", directly.
Class "BaseLearner.Config", by class "Regression.Config", distance 2.
signature(object = "KNN.Regression.Config"): ...
signature(object = "NNET.Regression.Config"): ...
signature(object = "RF.Regression.Config"): ...
signature(object = "SVM.Regression.Config"): ...
signature(object = "GBM.Regression.Config"): ...
signature(object = "PENREG.Regression.Config"): ...
signature(object = "BART.Regression.Config"): ...
Alireza S. Mahani, Mansour T.A. Sharabiani
make.configs, make.instances, make.configs.knn.regression, make.configs.nnet.regression, make.configs.rf.regression, make.configs.svm.regression, make.configs.gbm.regression, "Regression.Config", "BaseLearner.Config"
"KNN.Regression.FitObj", "NNET.Regression.FitObj", "RF.Regression.FitObj", "SVM.Regression.FitObj", "GBM.Regression.FitObj", "PENREG.Regression.FitObj", "BART.Regression.FitObj"
Objects returned by base learner training functions.
Objects can be created by calls of the form new("KNN.Regression.FitObj", ...).
All classes inherit slots config, est, and pred from "Regression.FitObj". Some base learners may have additional slots as described below.
For KNN.Regression.FitObj:
formula:Object of class "formula", copy of same argument from training call BaseLearner.Fit.
data:Object of class "data.frame", copy of same argument from training call BaseLearner.Fit.
For NNET.Regession.FitObj:
y.range:Object of class "numeric", range of response variable in training data. This is used for scaling of data during prediction so that it falls between 0 and 1 for regression tasks.
y.min:Object of class "numeric", minimum of response variable in training data. This is used for scaling of data during prediction so that it falls between 0 and 1 for regression tasks.
For PENREG.Regression.FitObj and BART.Regression.FitObj:
mm:A list containing data structures needed for creating the matrix of predictors and the response variable from the formula and data frame.
Class "Regression.FitObj", directly.
Class "BaseLearner.FitObj", by class "Regression.FitObj", distance 2.
None.
Alireza S. Mahani, Mansour T.A. Sharabiani
"BaseLearner.FitObj", "Regression.FitObj"
showClass("KNN.Regression.FitObj")showClass("KNN.Regression.FitObj")
"BaseLearner.Batch.FitObj" and "Regression.Batch.FitObj"
Classes for containing base learner batch training output.
Class "BaseLearner.Batch.FitObj" is virtual; therefore No objects may be created from it. Class "Regression.Batch.FitObj" extends "BaseLearner.Batch.FitObj" and is the output of function Regression.Batch.Fit.
fitobj.list:Object of class "list", containing the BaseLearner.FitObj outputs of lower-level BaseLearner.Fit function calls.
config.list:Object of class "list", containing the list of configuration objects for each base learner fit. This list is typically the output of make.configs function call.
filemethod:Object of class "logical", indicating whether file method is used for storing the estimation objects.
tmpfiles:Object of class "OptionalCharacter", containing (if applicable) the vector of filepaths used for storing estimation objects, if filemethod==TRUE.
For Regression.Batch.FitObj (in addition to above slots):
pred:Object of class "matrix", with each column containing the predictions of one base learner.
y:Object of class "numeric", containing the response variable corresponding to the training set.
No methods defined with class "BaseLearner.Batch.FitObj" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
showClass("BaseLearner.Batch.FitObj")showClass("BaseLearner.Batch.FitObj")
"BaseLearner.Config", "Regression.Config"
Base classes in the configuration class hierarchy.
"BaseLearner.Config" is a virtual Class: No objects may be created from it. "Regression.Config" is a base class for configuration classes of specific base learners, such as SVM.Regression.Config; therefore, there is typically no need to generate objects from this base class directly.
Class "Regression.Config" extends class "BaseLearner.Config", directly.
No methods defined with class "BaseLearner.Config" or "Regression.Config" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
KNN.Regression.Config, RF.Regression.Config, NNET.Regression.Config, GBM.Regression.Config, SVM.Regression.Config
showClass("BaseLearner.Config")showClass("BaseLearner.Config")
"BaseLearner.CV.Batch.FitObj" and "Regression.CV.Batch.FitObj"
Classes for containing base learner batch CV training output.
BaseLearner.CV.Batch.FitObj is virtual Class; therefore, no objects may be created from it. Class Regression.CV.Batch.FitObj is the output of Regression.CV.Batch.Fit function.
fitobj.list:Object of class "list", contains a list of objects of class BaseLearner.CV.FitObj, one per base learner instance.
instance.list:Object of class "Instance.List", the list of base learner instance passed to the function Regression.CV.Batch.Fit that produces the object.
filemethod:Object of class "logical", the boolean flag indicating whether estimation objects are saved to files or help in memory.
tmpfiles:Object of class "OptionalCharacter", list of temporary files used for storing estimation objects (if any).
tmpfiles.index.list:Object of class "list", with elements start and end, holding the start and end indexes into the tempfiles vector for each of the base learner instances trained.
tvec:Execution times for each base learner in the batch. Note: Currently implemented for serial execution only.
In addition, Regression.CV.Batch.FitObj contains the following slots:
pred:Object of class "matrix", with each column being the training-set prediction for one of base learner instances.
y:Object of class "OptionalNumeric", holding the response variable values for training set. This slot can be NULL for memory efficiency purposes.
No methods defined with class "BaseLearner.CV.Batch.FitObj" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
Regression.CV.Batch.Fit
"BaseLearner.CV.FitObj" and "Regression.CV.FitObj"
Classes for containing base learner CV training output.
"BaseLearner.CV.FitObj" is a virtual class: No objects may be created from it. "Regression.CV.FitObj" is the output of Regression.CV.Fit function.
fitobj.list:Object of class "list", contains a list of objects of class BaseLearner.Fit, one per partition fold.
partition:Object of class "OptionalInteger", representing how data must be split across folds during cross-validation. This is typically the output of generate.partition function.
filemethod:Object of class "logical", determining whether to save individual estimation objects to file or not.
In addition, Regression.CV.FitObj contains the following slot:
Object of class "OptionalNumeric", containing the prediction from the CV fit object for training data. This slot is allowed to take on a "NULL" value to reduce excess memory use by large ensemble models.
No methods defined with class "BaseLearner.CV.FitObj" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
Each base learner must provide its concrete implementation of this generic method.
BaseLearner.Fit(object, formula, data, tmpfile=NULL, print.level=1, ...)BaseLearner.Fit(object, formula, data, tmpfile=NULL, print.level=1, ...)
object |
An object of class |
formula |
Formula object expressing response and covariates. |
data |
Data frame containing response and covariates. |
tmpfile |
Filepath to save the estimation object to. If |
print.level |
Controlling verbosity level during fitting. |
... |
Arguments to be passed to/from other methods. |
signature(object = "GBM.Regression.Config")signature(object = "KNN.Regression.Config")signature(object = "NNET.Regression.Config")signature(object = "RF.Regression.Config")signature(object = "SVM.Regression.Config")signature(object = "PENREG.Regression.Config")signature(object = "BART.Regression.Config")"BaseLearner.FitObj" and "Regression.FitObj"
Base class templates for containing base learner training output.
"BaseLearner.FitObj" is a virtual class: No objects may be created from it. "Regression.FitObj" is a base class for objects representing trained models for individual base learners.
config:Object of class "BaseLearner.Config"; often one of the derived configuration classes belonging to a particular base learner.
For Regression.FitObj, we have the following additional fields:
est:Object of class "RegressionEstObj", typically containing the low-level list coming out of the training algorithm. If filemethod=TRUE during the fit, this object will be of class "character", containing the filepath to where the estimation object is stored.
pred:Object of class "OptionalNumeric", fitted values of the model for the training data. It is allowed to be "NULL" in order to reduce memory footrpint during cross-validated ensemble methods.
No methods defined with these classes in their signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
"KNN.Regression.FitObj", "RF.Regression.FitObj", "SVM.Regression.FitObj", "GBM.Regression.FitObj", "NNET.Regression.FitObj"
"Instance" and "Instance.List"
A base learner Instance is a combination of a base learner configuration and data partition. Instances constitute the major input into the cross-validation-based functions such as Regression.CV.Batch.Fit. An Instance.List is a collection of instances, along with the underlying definition of data partitions referenced in the instance objects. The function make.instances is a convenient function for generating an instance list from all permutations of a given list of base learner configurations and data partitions.
Objects can be created by calls of the form new("Instance", ...).
Instance has the following slots:
config:Object of class "BaseLearner.Config" ~~
partid:Object of class "character" ~~
Instance.List has the following slots:
instances:Object of class "list", with each element being an object of class Instance.
partitions:Object of class "matrix", defining data partitions referenced in each instance. This object is typically the output of generate.partitions.
No methods defined with class "Instance" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
make.instances, generate.partitions, Regression.CV.Batch.Fit
showClass("Instance")showClass("Instance")
Helper Functions for Manipulating Base Learner Configurations
make.configs(baselearner=c("nnet","rf","svm","gbm","knn","penreg") , config.df, type = "regression") make.configs.knn.regression(df=expand.grid( kernel=c("rectangular","epanechnikov","triweight","gaussian") , k=c(5,10,20,40))) make.configs.gbm.regression(df=expand.grid( n.trees=c(1000,2000) , interaction.depth=c(3,4) , shrinkage=c(0.001,0.01,0.1,0.5) , bag.fraction=0.5)) make.configs.svm.regression(df=expand.grid( cost=c(0.1,0.5,1.0,5.0,10,50,75,100) , epsilon=c(0.1,0.25) , kernel="radial")) make.configs.rf.regression(df=expand.grid( ntree=c(100,500) , mtry.mult=c(1,2) , nodesize=c(2,5,25,100))) make.configs.nnet.regression(df=expand.grid( decay=c(1e-4,1e-2,1,100) , size=c(5,10,20,40) , maxit=2000)) make.configs.penreg.regression(df = expand.grid( alpha = 0.0 , lambda = 10^(-8:+7))) make.configs.bart.regression(df = rbind(cbind(expand.grid( num_trees = c(50, 100), k = c(2,3,4,5)), q = 0.9, nu = 3) , cbind(expand.grid( num_trees = c(50, 100), k = c(2,3,4,5)), q = 0.75, nu = 10) )) make.instances(baselearner.configs, partitions) extract.baselearner.name(config, type="regression")make.configs(baselearner=c("nnet","rf","svm","gbm","knn","penreg") , config.df, type = "regression") make.configs.knn.regression(df=expand.grid( kernel=c("rectangular","epanechnikov","triweight","gaussian") , k=c(5,10,20,40))) make.configs.gbm.regression(df=expand.grid( n.trees=c(1000,2000) , interaction.depth=c(3,4) , shrinkage=c(0.001,0.01,0.1,0.5) , bag.fraction=0.5)) make.configs.svm.regression(df=expand.grid( cost=c(0.1,0.5,1.0,5.0,10,50,75,100) , epsilon=c(0.1,0.25) , kernel="radial")) make.configs.rf.regression(df=expand.grid( ntree=c(100,500) , mtry.mult=c(1,2) , nodesize=c(2,5,25,100))) make.configs.nnet.regression(df=expand.grid( decay=c(1e-4,1e-2,1,100) , size=c(5,10,20,40) , maxit=2000)) make.configs.penreg.regression(df = expand.grid( alpha = 0.0 , lambda = 10^(-8:+7))) make.configs.bart.regression(df = rbind(cbind(expand.grid( num_trees = c(50, 100), k = c(2,3,4,5)), q = 0.9, nu = 3) , cbind(expand.grid( num_trees = c(50, 100), k = c(2,3,4,5)), q = 0.75, nu = 10) )) make.instances(baselearner.configs, partitions) extract.baselearner.name(config, type="regression")
baselearner |
Name of base learner algorithm. Currently, seven base learners are included: 1) Neural Network ( |
df, config.df
|
Data frame, with columns named after tuning parameters belonging to the base learner, and each row indicating a tuning-parameter combination to include in the configuration list. |
type |
Type of base learner. Currently, only "regression" is supported. |
baselearner.configs |
Base learner configuration list to use in generating instances. |
partitions |
A matrix whose columns define data partitions, usually the output of generate.partitions. |
config |
Base learner configuration object. |
The make.configs family of functions return a list of objects of various base learner config classes, such as KNN.Regression.Config. Function make.instances returns an object of class Instance.List. Function extract.baselearner.name returns a character object representing the name of the base learner associated with the passed-in config object. For example, for a KNN.Regression.Config object, we get back "KNN". This utility function can be used in printing base learner names based on class of a config object.
Alireza S. Mahani, Mansour T.A. Sharabiani
"OptionalInteger"
Utility classes to allow for inclusion of "NULL" an a class instance, for memory efficiency. Each one of these is a class union between the underlying class ("integer", "character" and "numeric") and "NULL".
These classes are typically part of more complex classes representing outputs of ensemble fit functions.
No methods defined with class "OptionalInteger" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
Regression.FitObj, BaseLearner.CV.FitObj, Regression.CV.FitObj
Batch Training, Prediction and Diagnostics of Regression Base Learners.
Regression.Batch.Fit(config.list, formula, data, ncores = 1 , filemethod = FALSE, print.level = 1) ## S3 method for class 'Regression.Batch.FitObj' predict(object, ..., ncores=1) ## S3 method for class 'Regression.Batch.FitObj' plot(x, errfun=rmse.error, ...)Regression.Batch.Fit(config.list, formula, data, ncores = 1 , filemethod = FALSE, print.level = 1) ## S3 method for class 'Regression.Batch.FitObj' predict(object, ..., ncores=1) ## S3 method for class 'Regression.Batch.FitObj' plot(x, errfun=rmse.error, ...)
config.list |
List of configuration objects for batch of base learners to be trained. |
formula |
Formula objects expressing response and covariates. |
data |
Data frame containing response and covariates. |
ncores |
Number of cores to use during parallel training. |
filemethod |
Boolean indicator of whether to save estimation objects to disk or not. |
print.level |
Determining level of command-line output verbosity during training. |
object |
Object of class |
... |
Arguments to be passed from/to other functions. |
x |
Object of class |
errfun |
Error function to use for calculating errors plotted. |
Function Regression.Batch.Fit returns an object of class Regression.Batch.FitObj. Function predict.Regression.Batch.FitObj returns a matrix of predictions, each column corresponding to one base learner in the trained batch. Function plot.Regression.Batch.FitObj creates a plot of base learner errors over the training set, grouped by type of base learner (all configurations within a given base learner using the same symbol).
Alireza S. Mahani, Mansour T.A. Sharabiani
data(servo) myformula <- class~motor+screw+pgain+vgain myconfigs <- make.configs("knn") perc.train <- 0.7 index.train <- sample(1:nrow(servo), size = round(perc.train*nrow(servo))) data.train <- servo[index.train,] data.predict <- servo[-index.train,] ret <- Regression.Batch.Fit(myconfigs, myformula, data.train, ncores=2) newpred <- predict(ret, data.predict)data(servo) myformula <- class~motor+screw+pgain+vgain myconfigs <- make.configs("knn") perc.train <- 0.7 index.train <- sample(1:nrow(servo), size = round(perc.train*nrow(servo))) data.train <- servo[index.train,] data.predict <- servo[-index.train,] ret <- Regression.Batch.Fit(myconfigs, myformula, data.train, ncores=2) newpred <- predict(ret, data.predict)
CV Batch Training and Diagnostics of Regression Base Learners.
Regression.CV.Batch.Fit(instance.list, formula, data , ncores = 1, filemethod = FALSE, print.level = 1 , preschedule = TRUE , schedule.method = c("random", "as.is", "task.length") , task.length) ## S3 method for class 'Regression.CV.Batch.FitObj' predict(object, ..., ncores=1 , preschedule = TRUE) ## S3 method for class 'Regression.CV.Batch.FitObj' plot(x, errfun=rmse.error, ylim.adj = NULL, ...)Regression.CV.Batch.Fit(instance.list, formula, data , ncores = 1, filemethod = FALSE, print.level = 1 , preschedule = TRUE , schedule.method = c("random", "as.is", "task.length") , task.length) ## S3 method for class 'Regression.CV.Batch.FitObj' predict(object, ..., ncores=1 , preschedule = TRUE) ## S3 method for class 'Regression.CV.Batch.FitObj' plot(x, errfun=rmse.error, ylim.adj = NULL, ...)
instance.list |
An object of class |
formula |
Formula object expressing response variable and covariates. |
data |
Data frame expressing response variable and covariates. |
ncores |
Number of cores in parallel training. |
filemethod |
Boolean flag, indicating whether to save estimation objects to file or not. |
print.level |
Verbosity level. |
preschedule |
Boolean flag, indicating whether parallel jobs must be scheduled statically ( |
schedule.method |
Method used for scheduling tasks across threads. In |
task.length |
Estimation task execution times, to be used for loading balancing during parallel execution. |
object |
Output of |
... |
Arguments passed from/to other functions. |
x |
Object of class |
errfun |
Error function used in generating plot. |
ylim.adj |
Optional numeric argument to use for adjusting the range of y-axis. |
Function Regression.CV.Batch.Fit produces an object of class Regression.CV.Batch.FitObj. The predict method produces a matrix, whose columns each represent training-set predictions from one of the batch of base learners (in CV fashion).
Alireza S. Mahani, Mansour T.A. Sharabiani
data(servo) myformula <- class~motor+screw+pgain+vgain perc.train <- 0.7 index.train <- sample(1:nrow(servo) , size = round(perc.train*nrow(servo))) data.train <- servo[index.train,] data.predict <- servo[-index.train,] parts <- generate.partitions(1, nrow(data.train)) myconfigs <- make.configs("knn" , config.df = expand.grid(kernel = "rectangular", k = c(5, 10))) instances <- make.instances(myconfigs, parts) ret <- Regression.CV.Batch.Fit(instances, myformula, data.train) newpred <- predict(ret, data.predict)data(servo) myformula <- class~motor+screw+pgain+vgain perc.train <- 0.7 index.train <- sample(1:nrow(servo) , size = round(perc.train*nrow(servo))) data.train <- servo[index.train,] data.predict <- servo[-index.train,] parts <- generate.partitions(1, nrow(data.train)) myconfigs <- make.configs("knn" , config.df = expand.grid(kernel = "rectangular", k = c(5, 10))) instances <- make.instances(myconfigs, parts) ret <- Regression.CV.Batch.Fit(instances, myformula, data.train) newpred <- predict(ret, data.predict)
This function trains the base learner indicated in the configuration object in a cross-validation scheme using the partition argument. The cross-validated predictions are assembled and returned in the pred slot of the Regression.CV.FitObj object. Individual trained base learners are also assembled and returned in the return object, and used in the predict method.
Regression.CV.Fit(regression.config, formula, data , partition, tmpfiles = NULL, print.level = 1) ## S3 method for class 'Regression.CV.FitObj' predict(object, newdata=NULL, ...)Regression.CV.Fit(regression.config, formula, data , partition, tmpfiles = NULL, print.level = 1) ## S3 method for class 'Regression.CV.FitObj' predict(object, newdata=NULL, ...)
regression.config |
An object of class |
formula |
Formula object expressing response and covariates. |
data |
Data frame containing response and covariates. |
partition |
Data partition, typically the output of |
tmpfiles |
List of temporary files to save the |
print.level |
Integer setting verbosity level of command-line output during training. |
object |
An object of class |
newdata |
Data frame containing new observations. |
... |
Arguments passed to/from other methods. |
Function Regression.CV.Fit returns an object of class Regression.CV.FitObj. Function predict.Regression.CV.FitObj returns a numeric vector of length nrow(newdata).
Alireza S. Mahani, Mansour T.A. Sharabiani
data(servo) myformula <- class~motor+screw+pgain+vgain myconfig <- make.configs("knn", config.df=data.frame(kernel="rectangular", k=10)) perc.train <- 0.7 index.train <- sample(1:nrow(servo), size = round(perc.train*nrow(servo))) data.train <- servo[index.train,] data.predict <- servo[-index.train,] mypartition <- generate.partition(nrow(data.train),nfold=3) ret <- Regression.CV.Fit(myconfig[[1]], myformula, data.train, mypartition) newpred <- predict(ret, data.predict)data(servo) myformula <- class~motor+screw+pgain+vgain myconfig <- make.configs("knn", config.df=data.frame(kernel="rectangular", k=10)) perc.train <- 0.7 index.train <- sample(1:nrow(servo), size = round(perc.train*nrow(servo))) data.train <- servo[index.train,] data.predict <- servo[-index.train,] mypartition <- generate.partition(nrow(data.train),nfold=3) ret <- Regression.CV.Fit(myconfig[[1]], myformula, data.train, mypartition) newpred <- predict(ret, data.predict)
"Regression.Integrator.Config", "Regression.Select.Config", "Regression.Integrator.FitObj", "Regression.Select.FitObj"
Virtual base classes to contain configuration and fit objects for integrator operations.
All virtual classes; therefore, no objects may be created from them.
For config classes:
errfun:Object of class "function" ~~
For FitObj classes:
config:Object of class "Regression.Integrator.Config" or "Regression.Select.Config" for the Integrator and Select classes.
est:Object of class ANY, containing estimation objects for concrete extensions of the virtual classes.
pred:Object of class "numeric", containing the prediction of integrator operations.
No methods defined with class "Regression.Integrator.Config" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
Regression.Integrator.Fit, Regression.Select.Fit
Generic methods that can be extended and used in constructing integrator algorithms by other packages.
Regression.Integrator.Fit(object, X, y, print.level=1) Regression.Select.Fit(object, X, y, print.level=1)Regression.Integrator.Fit(object, X, y, print.level=1) Regression.Select.Fit(object, X, y, print.level=1)
object |
An object typically containing all configuration parameters of a particular integrator algorithm or operations. |
X |
Matrix of covariates. |
y |
Vector of response variables. |
print.level |
Verbosity level. |
"RegressionEstObj"
Union of (converted) S3 classes for individual base learners as defined in their corresponding packages. The special class "character" has been added to allow for returning filepaths when saving estimation objects to disk.
Objects from this class are typically returned as part of FitObj family of classes.
No methods defined with class "RegressionEstObj" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
"RegressionSelectPred"
Union of classes "NULL", "numeric" and "matrix" to hold prediction output of Select operations based on generic function Regression.Select.Fit. Class NULL is included to allow methods to save memory by not returning the prediction, espeically when a highe-level wrapper takes responsibility for holding a global copy of all prediction results. The "numeric" and "matrix" classes allow for a single predictor or multiple predictors to be produced by a Select operation.
A virtual Class: No objects may be created from it.
No methods defined with class "RegressionSelectPred" in the signature.
Alireza S. Mahani, Mansour T.A. Sharabiani
A small regression data set taken from UCI Machine Learning Repository. Response variable is "class".
data("servo")data("servo")
The format is: chr "servo"
Bache, K. & Lichman, M. (2013). UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. Irvine, CA: University of California, School of Information and Computer Science.
data(servo) lm(class~motor+screw+pgain+vgain, servo)data(servo) lm(class~motor+screw+pgain+vgain, servo)
Collection of utility functions for generating random partitions in datasets (for cross-validated operations), extracting regression response variable from dataset, loading an object from memory and assigning it to an arbitrary symbol, and error definitions.
generate.partition(ntot, nfold = 5) generate.partitions(npart=1, ntot, nfold=5, ids=1:npart) regression.extract.response(formula, data) load.object(file) rmse.error(a,b)generate.partition(ntot, nfold = 5) generate.partitions(npart=1, ntot, nfold=5, ids=1:npart) regression.extract.response(formula, data) load.object(file) rmse.error(a,b)
ntot |
Total number of observations in the data set to be partitioned. |
nfold |
Number of folds in the data partition. |
npart |
Number of random partitions to generate. |
ids |
Column names for the resulting partition matrix, used as partition ID. |
formula |
Formula object to use for extracting response variable from data set. |
data |
Data frame containing response variable as defined in |
file |
Filepath from which to read an |
a, b
|
Vectors of equal length, used to calculate their RMSE distance. |
Function generate.partition returns an integer vector of length ntot, with entries - nearly - equally split in the range 1:nfold. Function generate.partitions returns a matrix of size ntot x npart, with each column being a partition alike to the output of generate.partition. The columns are named ids. Function regression.extract.response returns a vector of length nrow(data), containing the numeric response variable for regression problems. Function load.object returns the saved object, but only works if only a single R object was saved to the file. Function rmse.error returns a single numeric value representing root-mean-squared-error distance between vectors a and b.
Alireza S. Mahani, Mansour T.A. Sharabiani
validate in Package EnsembleBase ~~~~ Methods for function validate in package EnsembleBase ~~
signature(object = "Regression.Batch.FitObj")signature(object = "Regression.CV.Batch.FitObj")