agglda: LDA model aggregation

In jinyung/otolith: Identification of fish otolith using digital images

Description

aggregate LDA models based on iterated k-fold resampling method

Usage

agglda(X, Y, newdata = NULL, type = c("vote", "post"), k = 5, run = 100,
  threshold, prior = c("equal", "proportion"), suppress = FALSE)

Arguments

X	matrix/ dataframe of predictors, e.g. EFA coefficients/ PC scores selected using selectdim
Y	factor/ character giving the class, e.g. value obtained from getclass or sp value from routine1 object
newdata	matrix/ dataframe of newdata to be predicted. see details.
type	prediction based on the majority vote ("vote") from the submodels or based on the mean posterior probability from the submodels ("post")
k	the fold number for k-fold resampling
run	the iteration number for iteration of k-fold resampling
threshold	single numeric value of < 1. threshold of the proportion of majority vote or the mean posterior probability. predictions with less than the value will be reported as NA
prior	the prior used in lda models, "equal" means all classes have same prior, "proportion" means prior according to the classes weight in the training data.
suppress	logical. whether to suppress the progress monitoring output

Details

If newdata is provided, the function is in the prediction mode, the aggregated model will be built from X and Y and predicition is performed on newdata. Otherwise, if newdata = NULL (default) the function is in evaluation mode.

In evaluation mode, overall accuracy of the model and the by-class statistics are calculated, similar to that of mrkfcv2. However, the statistics are calculated based on the aggregated prediction. See reference for explanation on model aggregation and thresholding.

Value

accuracy	[evaluation mode] the overall accuracy in percent
conmat	[evaluation mode] confusion matrix
stat	[evaluation mode] matrix containing the statistics of each class, see details
total	[evaluation mode] the total percent of reported prediction after threshold. give NULL if threshold is not given
ind.prediction	matrix containing the prediction result on each training/ new specimens

References

Beleites, C., & Salzer, R. (2008). Assessing and improving the stability of chemometric models in small sample size situations. Analytical and Bioanalytical Chemistry, 390(5), 1261-1271.

See Also

kfcv, mrkfcv

jinyung/otolith documentation built on May 19, 2019, 10:36 a.m.