ELM-class: Class 'elm'

In mugiro/elm: Extreme Learning Machine for regression and classification

Description

A S4 class to represent an Extreme Learning Machine (ELM) model

Usage

## S4 method for signature 'elm'
add_neurons(object, act_fun, nn, w_in = NULL, b = NULL)

## S4 method for signature 'elm'
initialize(.Object = object, inputs = 0, outputs = 0)

## S4 method for signature 'elm'
show(object)

## S4 method for signature 'elm'
get_error(object, n_sel, h, y, h_val = NULL, y_val = NULL,
  cv_rows = NULL)

## S4 method for signature 'elm'
mse(object, y, yp, x)

## S4 method for signature 'elm'
class_postprocess(object, yp, class_output, ml_threshold)

## S4 method for signature 'elm'
rank_neurons(object, nn_max, h = NULL, y = NULL)

## S4 method for signature 'elm'
train(object, x, y, x_val = NULL, y_val = NULL,
  type = "reg", tune = "none", ranking = "random", validation = "none",
  folds = 10, class_weights = NULL, ...)

## S4 method for signature 'elm'
project(object, x, rbf_dist = "euclidean")

## S4 method for signature 'elm'
solve_system(object, h, y, solve = TRUE)

## S4 method for signature 'elm'
train_pruning(object, h, y, h_val = NULL, y_val = NULL,
  cv_rows = NULL)

## S4 method for signature 'elm'
prune(object, n_sel)

Methods (by generic)

• add_neurons: add neurons of the same type of activation function to the hidden layer

• initialize: initalize an object of class elm

• show: display an object of class elm

• get_error: implement a validation procedure

• mse: MSE error

• class_postprocess: Description of class_postprocess for ELM (origin predic.R)

• rank_neurons: rank neurons of a ELM

• train: train the elm

• project: project form input-space to neuron-space. Compute H

• solve_system: solve linear system H x Wout = Y

• train_pruning: Optimization procedure for obtaining the optimial number of neurons for pruning.

• prune: Prune the hidden layer of a elm

Slots

inputs

The number of input features.

outputs

The number of outputs.

h_neurons

An object of classs hiddenlayer

w_out

The weight output vector that includes the computed weights between the hidden and the output layer.

type

The type of model implemented:

• "reg": regression problem.

• "class_mc": multi-class: the sample belongs to 1 class out of n.

• "class_ml": multi-label: the sample can belong to m classes out of n (m<n).

• "class_w": weigted classification

tune

Parameter to define the model structure selection method implemented to tune the model hyper-parameters #'

• "none": no model selection

• "pruning": pruning of neurons of the hidden layer: P-ELM, if "ridge = 0 & ranking = "random", OP-ELM, if "ridge = 0 & ranking = lars", TROP-ELM, if ("ridge != 0 & ranking = lars)

ranking

A character to select the type of ranking implemented when prunning option is selected.

• "random" - random ranking

• "lars" - ranking based on lars - L1 penalty

results

The error used to evaluate model performance. mse c(mse_train, mse_val)

ridge

The regularization parameter used to include the L2 penalty the#'

validation

The validation procedure used for developing the model. #'

• "none" - no validation process <<<<<<ANDRES<<<<<<

• "v" - validation. Xv and Yv are required

• "cv" - cross validation. The number of folds is required

• "loo" - leave one out based on the PRESS statistic

folds

The number number of folds for the cross-validation procedure.

class_weights

numeric vector of length = number_of_classes with the weigths for weighted type

batch

The size of the bacth in an adaptative ELM.

time_exec

The time of calculation for training the model.

bigdata

An logical parameter to select the kind of acceleration used in case of solving big data problems.

mugiro/elm documentation built on May 23, 2019, 8:21 a.m.