Skip to contents

Fits a classification random forest that predicts which model produced a simulation from its statistics (Pudlo et al. 2016), with the linear discriminant axes of the statistics added as predictors. The out-of-bag error and the confusion matrix measure, before looking at the data, how often the models can be told apart. predict() gives the chosen model for observed statistics, the votes of the trees and the posterior probability of the chosen model, which is more reliable than the proportion of votes.

Usage

abc_choose(tables, statistics = NULL, ntree = 500, lda = TRUE, ncores = 1)

# S3 method for class 'mc_abc_choice'
predict(object, newdata, ntree = 500, ncores = 1, ...)

Arguments

tables

Named list of reference tables, one per model, as from abc_table(), or a data frame with a column model. Only the statistics shared by all models are used.

statistics

Names of the statistics. By default all statistics shared by the models.

ntree

Number of trees.

lda

Add the linear discriminant axes as predictors?

ncores

Number of threads.

object

An mc_abc_choice.

newdata

Named numeric vector of observed statistics or a data frame with one observation per row.

...

Not used.

Value

An mc_abc_choice, a list with forest, training, statistics, models, error (out-of-bag error) and confusion (rows: true model; columns: chosen model).

For predict(), a data frame with observation, model, probability and one column of votes per model.

Functions

  • predict(mc_abc_choice): Chosen model, votes and posterior probability of the chosen model.

References

Pudlo, P., Marin, J.-M., Estoup, A., Cornuet, J.-M., Gautier, M., & Robert, C. P. (2016). Reliable ABC model choice via random forests. Bioinformatics, 32, 859–866. doi:10.1093/bioinformatics/btv684

Examples

set.seed(1)
a <- data.frame(s1 = rnorm(200, 0), s2 = rnorm(200))
b <- data.frame(s1 = rnorm(200, 1.5), s2 = rnorm(200))
choice <- abc_choose(list(neutral = a, niche = b), ntree = 100)
choice$confusion
#>         neutral niche
#> neutral     140    60
#> niche        53   147
predict(choice, c(s1 = 1.2, s2 = 0))
#>   observation model probability votes_neutral votes_niche
#> 1           1 niche      0.8764            70         430