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
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
|
#' @title Aggregation methods.
#'
#' @description
#' \describe{
#' \item{**test.mean**}{Mean of performance values on test sets.}
#' \item{**test.sd**}{Standard deviation of performance values on test sets.}
#' \item{**test.median**}{Median of performance values on test sets.}
#' \item{**test.min**}{Minimum of performance values on test sets.}
#' \item{**test.max**}{Maximum of performance values on test sets.}
#' \item{**test.sum**}{Sum of performance values on test sets.}
#' \item{**train.mean**}{Mean of performance values on training sets.}
#' \item{**train.sd**}{Standard deviation of performance values on training sets.}
#' \item{**train.median**}{Median of performance values on training sets.}
#' \item{**train.min**}{Minimum of performance values on training sets.}
#' \item{**train.max**}{Maximum of performance values on training sets.}
#' \item{**train.sum**}{Sum of performance values on training sets.}
#' \item{**b632**}{Aggregation for B632 bootstrap.}
#' \item{**b632plus**}{Aggregation for B632+ bootstrap.}
#' \item{**testgroup.mean**}{Performance values on test sets are grouped according
#' to resampling method. The mean for every group is calculated, then the mean of those means.
#' Mainly used for repeated CV.}
#' \item{**testgroup.sd**}{Similar to **testgroup.mean** - after
#' the mean for every group is calculated, the standard deviation of those means is obtained.
#' Mainly used for repeated CV.}
#' \item{**test.join**}{Performance measure on joined test sets.
#' This is especially useful for small sample sizes where unbalanced group sizes have a significant impact
#' on the aggregation, especially for cross-validation test.join might make sense now.
#' For the repeated CV, the performance is calculated on each repetition and then aggregated
#' with the arithmetic mean.}
#' }
#' @seealso [Aggregation]
#' @name aggregations
#' @usage NULL
#' @format NULL
#' @export
#' @rdname aggregations
test.mean = makeAggregation(
id = "test.mean",
name = "Test mean",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) mean(perf.test)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.sd = makeAggregation(
id = "test.sd",
name = "Test sd",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) sd(perf.test)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.median = makeAggregation(
id = "test.median",
name = "Test median",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) median(perf.test)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.min = makeAggregation(
id = "test.min",
name = "Test minimum",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) min(perf.test)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.max = makeAggregation(
id = "test.max",
name = "Test maximum",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) max(perf.test)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.sum = makeAggregation(
id = "test.sum",
name = "Test sum",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) sum(perf.test)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.range = makeAggregation(
id = "test.range",
name = "Test range",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) diff(range(perf.test))
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.rmse = makeAggregation(
id = "test.rmse",
name = "Test RMSE",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) sqrt(mean(perf.test^2))
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.mean = makeAggregation(
id = "train.mean",
name = "Training mean",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) mean(perf.train)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.sd = makeAggregation(
id = "train.sd",
name = "Training sd",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) sd(perf.train)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.median = makeAggregation(
id = "train.median",
name = "Training median",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) median(perf.train)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.min = makeAggregation(
id = "train.min",
name = "Training min",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) min(perf.train)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.max = makeAggregation(
id = "train.max",
name = "Training max",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) max(perf.train)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.sum = makeAggregation(
id = "train.sum",
name = "Training sum",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) sum(perf.train)
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.range = makeAggregation(
id = "train.range",
name = "Training range",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) diff(range(perf.train))
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
train.rmse = makeAggregation(
id = "train.rmse",
name = "Training RMSE",
properties = "req.train",
fun = function(task, perf.test, perf.train, measure, group, pred) sqrt(mean(perf.train^2))
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
b632 = makeAggregation(
id = "b632",
name = ".632 Bootstrap",
properties = c("req.train", "req.test"),
fun = function(task, perf.test, perf.train, measure, group, pred) {
mean(0.632 * perf.test + 0.368 * perf.train)
}
)
# FIXME: read this again properly and double check it
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
b632plus = makeAggregation(
id = "b632plus",
name = ".632 Bootstrap plus",
properties = c("req.train", "req.test"),
fun = function(task, perf.test, perf.train, measure, group, pred) {
df = as.data.frame(pred)
a = numeric(length(perf.test))
for (i in seq_along(a)) {
df2 = df[df$iter == i, , drop = FALSE]
y1 = df2$truth
y2 = df2$response
grid = expand.grid(y1, y2, KEEP.OUT.ATTRS = FALSE)
pred2 = makePrediction(
task.desc = pred$task.desc, row.names = rownames(grid),
id = NULL, truth = grid[, 1L], predict.type = "response", y = grid[, 2L],
time = NA_real_
)
gamma = performance(pred2, measures = measure)
R = (perf.test[i] - perf.train[i]) / (gamma - perf.train[i])
w = 0.632 / (1 - 0.368 * R)
a[i] = (1 - w) * perf.train[i] + w * perf.test[i]
}
return(mean(a))
}
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
testgroup.mean = makeAggregation(
id = "testgroup.mean",
name = "Test group mean",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) {
mean(vnapply(split(perf.test, group), mean))
}
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
testgroup.sd = makeAggregation(
id = "testgroup.sd",
name = "Test group standard deviation",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) {
sd(BBmisc::vnapply(split(perf.test, group), mean))
}
)
#' @export
#' @usage NULL
#' @format NULL
#' @rdname aggregations
test.join = makeAggregation(
id = "test.join",
name = "Test join",
properties = "req.test",
fun = function(task, perf.test, perf.train, measure, group, pred) {
df = as.data.frame(pred)
f = if (length(group)) group[df$iter] else factor(rep(1L, nrow(df)))
mean(vnapply(split(df, f), function(df) {
if (pred$predict.type == "response") y = df$response
if (pred$predict.type == "prob") {
y = df[, stri_startswith_fixed(colnames(df), "prob."), drop = FALSE]
colnames(y) = stri_sub(colnames(y), 6L)
}
npred = makePrediction(
task.desc = pred$task.desc, row.names = rownames(df),
id = NULL, truth = df$truth, predict.type = pred$predict.type, y = y,
time = NA_real_
)
performance(npred, measure)
}))
}
)
|
