ensemble_randForest: Bagged Ensemble Random Forests Model.
In oislen/BuenaVista: Functions for Everyday Data Science Tasks

Description Usage Arguments Value See Also Examples

This function creates a bagged random forests model on a given a dataset. This function effictively acts as a wrapper for the random forest package. The random forests model is a very versitile classifcation and prediction model that incorporates feature selection using shrinkage parametres and cross validation.

ensemble_randForest(y_index, train, valid_size = NULL, test = NULL,
  type = c("prediction", "classification"), ntree = 500,
  importance = FALSE, n = 10, r = NULL, r_replace = FALSE, c = NULL,
  c_replace = FALSE, seed = TRUE)

`y_index`	A column index representing the response variable of the model.
`train`	A dataset for the random forest model to be trained on. The order and names of train set should be the exact same as the test set.
`valid_size`	A natural number indicating the number of observations to be randomly sampled from the training data for model validation.
`test`	A dataset for the random forest model to predict for. The order and names of test set should be the exact same as the train set.
`type`	The type of response variable, either 'prediction' or 'classification'.
`n`	A natural number indicating the number of random forest models to be built.
`r`	The number of rows to be bagged. Note r < nrow(train).
`r_replace`	A logical object allow resampling when bagging rows. Default is FALSE.
`c`	The number of columns to be bagged Note c < ncol(train)
`c_replace`	A logical object allowing resampling when bagging columns Default is FALSE.
`seed`	Logical, indicating whether a random seed should be implemented.
`file_name`	A character object indicating the file name when saving the data frame. The default is NULL. The name must include the .csv suffixs.
`directory`	A character object specifying the directory where the data frame is to be saved as a .csv file.

Outputs a list of information related to the ensemble GLMNET model. The first object of the list is a data frame of the response observations, the corresponding predictions and the error associated with the prediction. The second object of the list is a data frame of model performance metrics. The third object of the list is a vector of predictions / classifications for the specified test set.

ensemble_mars, ensemble_mlr

# Example 1
# Numeric prediction example with Iris Data
data <- iris[sample(1:150, size = 150, replace = FALSE),]
# subset training data
train_data <- data[1:150,-5]
# bag rows only
ensemble_randForest(y_index = 1, train = train_data, valid_size = 50, n = 10, r = 80, type = "prediction")
# bag columns only
ensemble_randForest(y_index = 1, train = train_data, valid_size = 50, n = 10, c = 2, type = "prediction")
# bagg rows and columns
ensemble_randForest(y_index = 1, train = train_data, valid_size = 50, n = 10, r = 80, c = 2, type = "prediction")

# Example 2
# Binomial classication example with irisData
data <- iris[sample(1:150, size = 150, replace = FALSE),]
# Dummy encode the Species
data <- derive_variables(dataset = data, type = "dummy", integer = TRUE, return_dataset = TRUE)
# Convert the response variable into a binary factor with two class
data$Species_setosa <- as.factor(data$Species_setosa)
# Extract the test data
test <- data[101:50,c(1,2,3,4,6,7)]
# move Species_setosa to the front of the data frame
data <- data[,c(5,1,2,3,4,6,7)]
# fit a model
ensemble_randForest(y_index = 1, train = data, valid_size = 50, n = 10, type = "classification")
# fit a model
ensemble_randForest(y_index = 1, train = data, valid_size = 50, n = 20, r = 80, type = "classification")
# fit a model
ensemble_randForest(y_index = 1, train = data, valid_size = NULL, test = test, n = 10, type = "classification")

# Example 3
# Validation set with Multinomial Class prediction example using iris
data <- iris
data <- data[sample(1:150, size = 150, replace = FALSE),c(5,1,2,3,4)]
# plots
ense <- glmnet(y = as.vector(data[,1]), x = as.matrix(data[,-1]), alpha = 1, family = "binomial")
plot(ense)
ense <- cv.glmnet(y = as.vector(data[,1]), nfolds = 10, x = as.matrix(data[,-1]), alpha = 1, family = "binomial")
plot(ense)
# raw calculations
predict.cv.glmnet(object = cv.glmnet(y = as.vector(data[1:100,1]), x = as.matrix(data[1:100,-1]), alpha = 1, family = "multinomial"), newx = as.matrix(data[101:50,-1]), type = "class")

# Example 4
# Test set with numeric prediction using titanic
descriptive_statistics(dataset = titanic)
str(titanic)
data <- titanic[,c(6,2,3,7,8,10)]
data <- data[order(data$Age),]
data$Age
train <- data[1:714, ]
valid_size <- 50
test <- data[715:891, ]
ensemble_glmnet(y_index = 1, train = train, valid = valid_size, test = test, n = 10, r = 600, c = 4)
ensemble_glmnet(y_index = 1, train = train, valid = valid_size, n = 10)

# Example 5 
# Possion Prediction with IrisData
counts = rpois(n = 150, lambda = 3)
data <- iris[sample(1:150, 150, FALSE), ]
data = cbind(counts, data)
train <- data[1:100,]
# test <- data[101:150,]
test <- data[101:150, -1]
ensemble_glmnet(y_index = 1, train = train, test = test, valid_size = 50, family = "poisson", type = "response")