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
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
|
###################################################################
##
## Todo:
## X. Make an argument for what to save ("prediction", "probabilities", "both")
## X. Adapt code to average predictions/probabilities
## 1. Use train/rfe/sbf saved prediction function to make predictions
## for the ensemble.
## X. Change `nrow` to an average of `complete.cases`
## X. Detect class prob column names and drop off the last one
## X. Breakout the 'n' code so that it isn't converted to character
## X. Call it "oob" or "oob_pred"
## 2. Make `oob_pred.list` smaller and create a new function to
## do the real work
## 3. TEST TEST TEST
###################################################################
##
oob_pred <- function (x, ...) UseMethod("oob_pred")
#' @export
oob_pred.train <- function(x, best = TRUE, average = TRUE, ...) {
if(is.null(x$pred))
stop("re-fit the model using 'trainControl(savePredictions=TRUE)'")
prd <- x$pred
pname <- as.character(x$modelInfo$parameters$parameter)
if(best) {
prd <- merge(prd, x$bestTune)
prd <- prd[, !(colnames(prd) %in% pname)]
}
bycol <- "rowIndex"
if(!best) bycol <- c(bycol, pname)
if(average) prd <- get_averages(x, prd, bycol)
if(x$modelType == "Classification") {
lev <- train_lev(x)
if(!is.null(lev)) {
if(is.character(prd$obs)) prd$obs <- factor(prd$obs, levels = lev)
if(is.character(prd$pred)) prd$pred <- factor(prd$pred, levels = lev)
}
}
prd
}
#' @export
oob_pred.rfe <- function(x, best = TRUE, average = TRUE, ...) {
if(is.null(x$pred))
stop("re-fit the model using 'rfeControl(saveDetails=TRUE)'")
prd <- x$pred
if(best) {
prd <- subset(prd, Variables == x$bestSubset)
prd <- prd[, colnames(prd) != "Variables"]
bycol <- "rowIndex"
} else bycol <- c("rowIndex", "Variables")
## Temp kludge since I think there is an issue with
## how `rfe` saves the data
prd <- prd[!duplicated(prd),]
if(average) prd <- get_averages(x, prd, bycol)
if(is.character(prd$obs) && !is.null(x$obsLevels))
prd$obs <- factor(prd$obs, levels = x$obsLevels)
if(is.character(prd$pred) && !is.null(x$obsLevels))
prd$pred <- factor(prd$pred, levels = x$obsLevels)
prd
}
#' @export
oob_pred.sbf <- function(x, average = TRUE, ...) {
if(is.null(x$pred))
stop("re-fit the model using 'rfeControl(saveDetails=TRUE)'")
prd <- x$pred[names(x$pred) == "predictions"]
prd <- rbind.fill(prd)
prd <- prd[!duplicated(prd),]
if(average) prd <- get_averages(x, prd, bycol = "rowIndex")
if(is.character(prd$obs) && !is.null(x$obsLevels))
prd$obs <- factor(prd$obs, levels = x$obsLevels)
if(is.character(prd$pred) && !is.null(x$obsLevels))
prd$pred <- factor(prd$pred, levels = x$obsLevels)
prd
}
#' @importFrom stats reshape
#' @export
oob_pred.list <- function(x, direction = "wide", what = "both", ...) {
num <- length(x)
oob <- lapply(x, oob_pred, ...)
## check column names and get those common to everything
cnames <- lapply(oob, colnames)
cnames <- table(unlist(cnames))
if(any(cnames < num)) {
cnames <- names(cnames[cnames == num])
for(i in 1:num) oob[[i]] <- oob[[i]][, cnames]
}
nms <- names(oob)
if(is.null(nms)) nms <- well_numbered("Model", length(oob))
for(i in seq(along.with = nms)) oob[[i]]$.label <- nms[i]
oob <- rbind.fill(oob)
if(length(table(table(oob$n))) > 1)
stop("Some averages have different sample sizes than others")
if(direction == "wide") {
vert_names <- colnames(oob)
vert_names <- vert_names[!(vert_names %in% c("rowIndex", "n", "obs", ".label"))]
oob <- reshape(oob, direction = "wide",
v.names = vert_names,
idvar = c("rowIndex", "obs", "n"),
timevar = ".label")
vert_names <- vert_names[vert_names != "pred"]
exclude <- vert_names[length(vert_names)]
oob <- oob[, !grepl(paste0("^", exclude, "\\."), names(oob))]
if(!is.null(what) && !("both" %in% what)) {
if(!is.null(what) && !("pred" %in% what))
oob <- oob[, !grepl("^pred\\.", names(oob))]
if(!is.null(what) && !("prob" %in% what)) {
for(i in vert_names)
oob <- oob[, !grepl(paste0("^", i, "\\."), names(oob))]
}
}
}
oob
}
###################################################################
##
get_averages <- function (x, ...) UseMethod("get_averages")
#' @importFrom stats complete.cases
get_averages.train <- function(x, prd, bycol = "rowIndex", ...) {
if("Regression" %in% x$modelType) {
out <- ddply(prd, bycol,
function(x) c(colMeans(x[, c("pred", "obs")])))
} else {
out <- ddply(prd, bycol,
function(x) c(pred = char_mode(x$pred),
obs = as.character(x$obs)[1]))
if(x$control$classProbs) {
lev <- train_lev(x)
cprobs <- ddply(prd, bycol,
function(x, lev) c(colMeans(x[, lev])),
lev = lev)
out <- merge(out, cprobs)
}
}
n <- ddply(prd, bycol, function(x) c(n = sum(complete.cases(x))))
out <- merge(out, n)
out
}
#' @importFrom stats complete.cases
get_averages.rfe <- function(x, prd, bycol = "rowIndex", ...) {
if(is.null(x$obsLevels)) {
out <- ddply(prd, bycol,
function(x) c(colMeans(x[, c("pred", "obs")])))
} else {
out <- ddply(prd, bycol,
function(x) c(pred = char_mode(x$pred),
obs = as.character(x$obs)[1]))
if(all(x$obsLevels %in% colnames(prd))) {
lev <- x$obsLevels
cprobs <- ddply(prd, bycol,
function(x, lev) c(colMeans(x[, lev])),
lev = lev)
out <- merge(out, cprobs)
}
}
n <- ddply(prd, bycol, function(x) c(n = sum(complete.cases(x))))
out <- merge(out, n)
out
}
#' @importFrom stats complete.cases
get_averages.sbf <- function(x, prd, bycol = "rowIndex", ...) {
if(is.null(x$obsLevels)) {
out <- ddply(prd, bycol,
function(x) c(colMeans(x[, c("pred", "obs")])))
} else {
out <- ddply(prd, bycol,
function(x) c(pred = char_mode(x$pred),
obs = as.character(x$obs)[1]))
if(all(x$obsLevels %in% colnames(prd))) {
lev <- x$obsLevels
cprobs <- ddply(prd, bycol,
function(x, lev) c(colMeans(x[, lev])),
lev = lev)
out <- merge(out, cprobs)
}
}
n <- ddply(prd, bycol, function(x) c(n = sum(complete.cases(x))))
out <- merge(out, n)
out
}
###################################################################
##
#' @importFrom stats complete.cases
char_mode <- function(x, random = TRUE, na.rm = FALSE) {
if(na.rm) x <- x[complete.cases(x)]
tab <- table(x)
tab <- tab[tab == max(tab)]
tab <- if(length(tab) > 1 & random) sample(tab, 1) else tab[1]
as.vector(names(tab))
}
train_lev <- function(x) {
if(x$modelType == "Classification") {
if(!is.null(x$modelInfo$levels)) {
lev <- x$modelInfo$levels(x$finalModel)
} else {
lev <- if(!isS4(x)) x$finalModel$obsLevel else unique(x$pred$obs)
}
} else lev <- NULL
lev
}
#' @importFrom stats cor
corr_mat <- function (object, metric = object$metrics,
...) {
dat <- object$values[, grepl(paste0("~", metric[1]),
colnames(object$values))]
colnames(dat) <- gsub(paste0("~", metric[1]), "", colnames(dat))
dat <- cor(dat, ...)
dat
}
|
