iqnn_tune: Tuning function for iqnn model
In kmaurer/iqbin: Iterative Quantile Binning

Description Usage Arguments Value See Also Examples

View source: R/IterativeQuantileNearestNeighbors.R

Identify optimal number of bins per dimension for iterative quantile nearest-neighbor model using k-fold cross validation. Search through possible numbers of bins per dimension using a iterative rules defined in the make_nbins_list function.

iqnn_tune(data, y, mod_type = "reg", bin_cols, nbins_range,
  jit = rep(0, length(bin_cols)), stretch = FALSE, tol = rep(0,
  length(bin_cols)), strict = FALSE, cv_k = 10, oom_search = FALSE,
  oom_base = 2)

`data`	Data frame containing the response variable and numeric input variables from the training data
`y`	Name of response variable column
`mod_type`	Depends on response variables type: "reg" creates iqnn-regression for predicting numeric values, "class" creates iqnn-classifier for predicting categorical values
`bin_cols`	vector of column names of variables to iteratively bin, ordered first to last
`nbins_range`	positive integer vector containing lower and upper bounds on number of bins in each dimension
`jit`	vector of margins for uniform jitter to each dimension to create seperability of tied obs due to finite precision
`stretch`	TRUE/FALSE if will bins be given tolerance buffer
`tol`	vector of tolerance values to stretch each dimension for future binning
`strict`	TRUE/FALSE: If TRUE Observations must fall within existing bins to be assigned; if FALSE the outer bins in each dimension are unbounded to allow outlying values to be assigned.
`cv_k`	integer specifying number of folds
`oom_search`	TRUE/FALSE: Only consider nbins that change number of neighbors by order of magnitudes
`oom_base`	order of magnitude base value (default to powers of 2)

data frame with one row per binning specification with desriptive and performance statistics: bin dimensitions, number of bins, equivalent k-nearest neightbor size, performance (mean squared error or class error rate)

Other iterative quantile nearest-neighbors functions: iqnn_cv_predict, iqnn_predict, iqnn

# 10-fold CV
cv_tune1 <- iqnn_tune(data=iris, y="Petal.Length", mod_type="reg",  
                      bin_cols=c("Sepal.Length","Sepal.Width","Petal.Width"),
                      nbins_range=c(2,5), jit=rep(0.001,3), strict=FALSE, cv_k=10)
cv_tune1
# LOO CV
cv_tune2 <- iqnn_tune(data=iris, y="Petal.Length", mod_type="reg",  
                      bin_cols=c("Sepal.Length","Sepal.Width","Petal.Width"),
                      nbins_range=c(2,5), jit=rep(0.001,3), strict=FALSE, cv_k=nrow(iris))
cv_tune2