subtree: Extract a subset of a tree of nested lists

Description Usage Arguments Details Value Author(s) See Also Examples

View source: R/reshape-result.r

Description

Modeling results produced by evaluate comes in the form of nested lists. This function can be used to subset or rearrange parts of the results into vectors, matrices or data frames. Also note the select function that provides an extension to the dplyr package for data manipulation.

Usage

1
subtree(x, i, ..., error_value, warn, simplify = TRUE)

Arguments

x

List of lists.

i

Indexes to extract on the first level of the tree. Can also be a function that will be applied to the downstream result of the function.

...

Indexes to extract on subsequent levels.

error_value

A template for the return value in case it is missing or invalid. Note that NA is a logical by default, causing subtree to also convert existing results to logicals. To get around this, please specify it as as.numeric(NA), as.character(NA), or similar (see the example below).

warn

Specifies whether warnings should be displayed (0), ignored (-1), or break execution (1). Works like the options parameter warn.

simplify

Whether to collapse results into vectors or matrices when possible (TRUE) or to preserve the original tree structure as a list (FALSE).

Details

This function can only be used to extract data, not to assign.

Value

A subset of the list tree.

Author(s)

Christofer B├Ącklin

See Also

select, get_prediction, get_importance, get_tuning.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
l <- list(A=list(a=0:2, b=3:4, c=023-22030),
          B=list(a=5:7, b=8:9))
subtree(l, 1:2, "b")
subtree(l, TRUE, mean, "a")

# More practical examples
x <- iris[-5]
y <- iris$Species
cv <- resample("crossvalidation", y, nfold=5, nrep=3)
procedure <- modeling_procedure("pamr")

# To illustrate the error handling capacities of subtree we'll introduce some
# spurious errors in the pre-processing function. By setting .return_error=TRUE
# they wont break the execution, but will instead be return in the results.
pre_error <- function(data, risk=.1){
    if(runif(1) < risk)
        stop("Oh no! Random error!")
    data
}
result <- evaluate(procedure, x, y, resample=cv,
    .save=c(importance=TRUE), .return_error=TRUE,
    pre_process = function(...){
        pre_split(...) %>%
            pre_error(risk=.3) %>%
            pre_pamr
    }
)
message(sum(sapply(result, inherits, "error")),
        " folds did not complete successfully!")

# Extract error rates. Since some folds fail it will be an ugly list with both
# numeric estimates and NULL values (for the failed folds).
subtree(result, TRUE, "error")

# To put it on a more consistent form we can impute the missing error rates
# with NA to allow automatic simplification into a vector (since it requires
# all values to be on the same form, i.e. numeric(1) rather than a mix
# between numeric(1) and NULL as in the previous example).
subtree(result, TRUE, "error", error_value=as.numeric(NA), warn=-1)

# Sum up feature importance for all classes within each fold and extract.
# Note that the lengths (= 4) must match between the folds for the automatic
# simplification to work.
subtree(result, TRUE, "importance", function(x){
    if(is.null(x)){
        rep(NA, 3)
    } else {
        colMeans(x[2:4])
    }
})

# The equivalent 'select' command would be ...
require(tidyr)
imp <- result %>% select(fold = TRUE, "importance", function(x){
    if(is.null(x)) return(NULL)
    x %>% gather(Species, Importance, -feature)
})
require(ggplot2)
ggplot(imp, aes(x=Species, y=Importance)) +
    geom_abline(intercept=0, slope=0, color="hotpink") +
    geom_boxplot() + facet_wrap(~feature)

emil documentation built on Aug. 1, 2018, 1:03 a.m.

Related to subtree in emil...