cross.validator: Model cross-validation

In dprocter/modeid: modeid: Identification of travel modes from GPS and accelerometry data

Description

Returns three column matrix, with as many rows as there are crossvalidation subsets.

Column 1 has the fitted models

Column 2 has the confusion.matrix for each cross-validation subset

Column 3 has the model.acc for each cross-validation subset

Usage

cross.validator(training.data, label, cv.marker, method, threads = 2,
  nrounds = 10, eta = 0.1, subsample = 1, max.depth = 10,
  eval_metric = "merror", early_stopping_rounds = 50,
  colsample_bytree = 1, min_child_weight = 1, gamma = 1, seed = NULL,
  verbose = 1)

Arguments

training.data	The data.frame of training data you want to cross-validate
label	A variable of the same length as the rows of training.data, with the true mode of travel
cv.marker	A numeric marker for cross-validation subsets
method	The method to use, currently either "randomForest" or "xgboost"
threads	number of threads to pass to xgboost to allow parralel computation, default is 2
nrounds	The number of iterations for xgboost to perform, default is 10
eta	The eta value to supply to xgboost, between 0 and 1, lowe values reduce overfitting, default 0.1
subsample	The proportion of data for xgboost toapply to each tree, smaller values reduce overfitting, default 0.2
max.depth	The maximum tree depth for xgboost, default is 10
verbose	What level of printed output you want. See ?xgboost for details, 1 is default, 0 is silent

Value

A four column matrix (I realise this is not terribly elegant, it will probably be changed). The rows contain different cross-validation subsets.

The columns are as follows:

1. Fitted models for each cross-validation subset

2. The confusion.matrix for each cross-validation subset

3. The model.acc for each cross-validation subset

dprocter/modeid documentation built on May 19, 2019, 8:21 a.m.