/
data_pdp.R
320 lines (298 loc) · 10.3 KB
/
data_pdp.R
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
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
#' Calculate data to construct partial dependence plots
#'
#' @description Calculate data to construct partial dependence plots for a given predictor
#'
#' @param model A model object of class "gam", "gbm", "glm", "graf", "ksvm", "ksvm", "maxnet”,
#' “nnet", and "randomForest" This model can be found in the first element of the list returned
#' by any function from the fit_, tune_, or esm_ function families
#' @param predictors character. Vector with a predictor name.
#' @param resolution numeric. Number of equally spaced points at which to predict continuous predictors. Default 50
#' @param resid logical. Calculate residuals based on training data. Default FALSE
#' @param training_data data.frame. Database with response (0,1) and predictor values used
#' to fit a model. Default NULL
#' @param projection_data SpatRaster. Raster layer with environmental variables used for model
#' projection. When this argument is used, function will calculate partial dependence curves
#' distinguishing conditions used in training and projection conditions
#' (i.e., projection data present in projection area but not training). Default NULL
#' @param clamping logical. Perform clamping. Only for maxent models. Default FALSE
#'
#' @return A list with two tibbles "pdpdata" and "resid".
#' \itemize{
#' \item pdpdata: has data to construct partial dependence plots, the first column includes values of the selected environmental
#' variable, the second column with predicted suitability, and the third
#' column with range type, with two values Training and Projecting, referring to suitability
#' calculated within and outside the range of training conditions. Third column is only returned
#' if "projection_data" argument is used
#' \item resid: has data to plot residuals. The first column includes values of the selected environmental
#' variable and the second column with predicted suitability.
#' }
#'
#'
#' @seealso {\code{\link{data_bpdp}}, \code{\link{p_bpdp}}, \link{p_pdp}}
#'
#' @export
#' @importFrom dplyr select as_tibble
#' @importFrom gbm predict.gbm
#' @importFrom kernlab predict
#' @importFrom mgcv predict.gam
#' @importFrom stats na.omit
#' @importFrom terra minmax
#'
#' @examples
#' \dontrun{
#' library(terra)
#' library(dplyr)
#'
#' somevar <- system.file("external/somevar.tif", package = "flexsdm")
#' somevar <- terra::rast(somevar) # environmental data
#' names(somevar) <- c("aet", "cwd", "tmx", "tmn")
#' data(abies)
#'
#' abies2 <- abies %>%
#' select(x, y, pr_ab)
#'
#' abies2 <- sdm_extract(abies2,
#' x = "x",
#' y = "y",
#' env_layer = somevar
#' )
#' abies2 <- part_random(abies2,
#' pr_ab = "pr_ab",
#' method = c(method = "kfold", folds = 5)
#' )
#'
#' svm_t1 <- fit_svm(
#' data = abies2,
#' response = "pr_ab",
#' predictors = c("aet", "cwd", "tmx", "tmn"),
#' partition = ".part",
#' thr = c("max_sens_spec")
#' )
#'
#' df <- data_pdp(
#' model = svm_t1$model,
#' predictors = c("aet"),
#' resolution = 100,
#' resid = TRUE,
#' projection_data = somevar,
#' training_data = abies2,
#' clamping = FALSE
#' )
#'
#' df
#' names(df)
#' df$pdpdata
#' df$resid
#'
#' plot(df$pdpdata[1:2], type = "l")
#' points(df$resid[1:2], cex = 0.5)
#'
#' # see p_pdp to construct partial dependence plot with ggplot2
#' }
data_pdp <-
function(model,
predictors,
resolution = 50,
resid = FALSE,
training_data = NULL,
projection_data = NULL,
clamping = FALSE) {
# Extract training data
if (class(model)[1] == "gam") {
x <- model$model[attr(model$terms, "term.labels")]
}
if (class(model)[1] == "graf") {
x <- model$obsx
x <- x[names(model$peak)]
}
if (class(model)[1] == "glm") {
flt <- grepl("[I(]", attr(model$terms, "term.labels")) |
grepl(":", attr(model$terms, "term.labels"))
flt <- attr(model$terms, "term.labels")[!flt]
x <- model$model[flt]
}
if (any(class(model)[1] == c("nnet.formula", "randomForest.formula", "ksvm", "gbm", "maxnet"))) {
if (is.null(training_data)) {
stop(
"For estimating partial plot data for Neural Networks, Random Forest, Support Vector Machine it is necessary to provide calibration data in 'training_data' argument"
)
}
if (class(model)[1] == "ksvm") {
x <- training_data[names(attr(model@terms, "dataClasses")[-1])]
} else if (class(model)[1] == "gbm") {
x <- training_data[, c(model$response.name, model$var.names)]
} else if (class(model)[1] == "maxnet") {
x <- training_data[, names(model$samplemeans)]
} else {
x <- training_data[names(attr(model$terms, "dataClasses")[-1])]
}
}
x <- stats::na.omit(x)
# Control average factor level
fact <- sapply(x, is.factor)
suit_c <- which(!fact)
fact <- which(fact)
suit_c <- data.frame(t(apply(x[suit_c], 2, mean)))
# suit_c <- data.frame((x[1, ])) For residuals
if (sum(fact) > 0) {
for (i in 1:length(fact)) {
ff <- sort(data.frame(unique(x[names(fact[i])]))[, 1])
ff <- ff[as.integer(length(ff) / 2)]
suit_c[names(fact)[i]] <- ff
}
}
if (predictors %in% names(fact)) {
rng <- sort(data.frame(unique(x[names(fact)]))[, predictors])
suit_c <- suit_c %>% dplyr::select(-{{ predictors }})
} else {
if (is.null(projection_data)) {
rng <- range(x[, predictors])
rng <- seq(rng[1], rng[2], length.out = resolution)
} else {
# Range extrapolation
rng <- terra::minmax(projection_data[[predictors]])
rng <- seq(rng[1], rng[2], length.out = resolution)
}
}
suit_c <- data.frame(rng, suit_c)
suit_c[predictors] <- NULL
names(suit_c)[1] <- predictors
# Predict model
if (class(model)[1] == "gam") {
suit_c <-
data.frame(suit_c[1],
Suitability = mgcv::predict.gam(model, newdata = suit_c, type = "response")
)
if (resid) {
suit_r <-
data.frame(x[predictors], Suitability = mgcv::predict.gam(model, type = "response"))
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "graf") {
suit_c <-
data.frame(
suit_c[1],
Suitability = predict.graf(
object = model,
newdata = suit_c[names(model$peak)],
type = "response",
CI = NULL
)[, 1]
)
if (resid) {
suit_r <-
data.frame(x[predictors],
Suitability = predict.graf(
object = model,
type = "response",
CI = NULL
)[, 1]
)
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "glm") {
suit_c <-
data.frame(suit_c[1],
Suitability = stats::predict.glm(model, newdata = suit_c, type = "response")
)
if (resid) {
suit_r <-
data.frame(x[predictors], Suitability = stats::predict.glm(model, type = "response"))
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "gbm") {
suit_c <-
data.frame(suit_c[1],
Suitability = suppressMessages(gbm::predict.gbm(model, newdata = suit_c, type = "response"))
)
if (resid) {
suit_r <-
data.frame(x[predictors], Suitability = suppressMessages(gbm::predict.gbm(model, newdata = x, type = "response")))
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "maxnet") {
suit_c <-
data.frame(suit_c[1],
Suitability = predict_maxnet(
model,
newdata = suit_c,
type = "cloglog",
clamp = clamping
)
)
if (resid) {
suit_r <-
data.frame(x[predictors], Suitability = predict_maxnet(
object = model,
newdata = data.frame(x),
type = "cloglog",
clamp = clamping
))
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "nnet.formula") {
suit_c <-
data.frame(suit_c[1], Suitability = stats::predict(model, newdata = suit_c, type = "raw"))
if (resid) {
suit_r <-
data.frame(x[predictors], Suitability = stats::predict(model, type = "raw"))
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "randomForest.formula") {
suit_c <-
data.frame(suit_c[1], Suitability = stats::predict(model, suit_c, type = "prob")[, 2])
if (resid) {
suit_r <-
data.frame(x[predictors], Suitability = stats::predict(model, type = "prob")[, 2])
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
if (class(model)[1] == "ksvm") {
suit_c <-
data.frame(suit_c[1],
Suitability = kernlab::predict(model, suit_c, type = "probabilities")[, 2]
)
if (resid) {
suit_r <-
data.frame(x[predictors],
Suitability = kernlab::predict(model, x, type = "probabilities")[, 2]
)
result <- list("pdpdata" = suit_c, "resid" = suit_r)
} else {
result <- list("pdpdata" = suit_c, "resid" = NA)
}
}
# Category of training and projection data
if (!is.null(projection_data)) {
if (!predictors %in% names(fact)) {
result$pdpdata$Type <-
ifelse(suit_c[, 1] >= min(x[, predictors]) &
suit_c[, 1] <= max(x[, predictors]),
"Training",
"Projection"
)
}
}
result <- lapply(result, function(x) if (is.data.frame(x)) dplyr::tibble(x))
return(result)
}