-
Notifications
You must be signed in to change notification settings - Fork 0
/
simulations.R
231 lines (205 loc) · 9.57 KB
/
simulations.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
##' With this script the simulations from Laabs et al. "Identification of
##' representative trees in random forests based on a new tree-based distance
##' measure" can be reproduced.
##' Please note, that the simulations in the paper were performed using
##' batchtools on a high throughout batch system. This script will implement
##' the same calculations on your local system, which may lead to a high
##' computation time. Comments will show you where you can save time or
##' incorporate your own batch system.
## Define directories
## Please define your main directory here.
## This should be the directory you cloned the git repository into.
main_dir <- "~"
## Create and define registry directory
dir.create(file.path(main_dir, "registries"), showWarnings = FALSE)
reg_dir <- file.path(main_dir, "registries")
## Create and define functions directory
dir.create(file.path(main_dir, "functions"), showWarnings = FALSE)
fun_dir <- file.path(main_dir, "functions")
## Create and define proc directory
dir.create(file.path(main_dir, "proc"), showWarnings = FALSE)
proc_dir <- file.path(main_dir, "proc")
## Load libraries
if (!"pacman" %in% installed.packages()){
install.packages("pacman")
}
pacman::p_load(batchtools)
pacman::p_load(ranger)
pacman::p_load(devtools)
pacman::p_load(rpart)
if("timbR" %in% installed.packages()){
library(timbR)
} else {
devtools::install_github("imbs-hl/timbR", "master")
library(timbR)
}
# --------------------------------------------------- #
# Data Simulation #
# --------------------------------------------------- #
## In this part the data will be simulated and saved
## for later use
## Define constants ----
n <- 1000 ## Number of samples in random forest training data sets. You can save time by reducing this.
## Load functions ----
source(file.path(fun_dir, "simulate_rf_setting_1.R"))
source(file.path(fun_dir, "simulate_rf_setting_2.R"))
source(file.path(fun_dir, "simulate_rf_setting_3.R"))
source(file.path(fun_dir, "simulate_rf_setting_4.R"))
source(file.path(fun_dir, "simulate_rf_setting_5.R"))
source(file.path(fun_dir, "calculate_decision_tree.R"))
source(file.path(fun_dir, "calculate_distances.R"))
## Create registry ----
reg_name <- "simulate_metrics"
reg <- batchtools::makeExperimentRegistry(
file.dir = file.path(reg_dir, reg_name),
work.dir = main_dir,
conf.file = NA, ## If you have a batch system, please enter conf file here,
packages = c("ranger", "timbR", "rpart") ## Define which packages to use in your simulations
)
## Add problems ----
## There is a separate function for generating each of the settings from the paper.
## You can save time, excluding settings you are not interested in.
batchtools::addProblem(name = "simulate_setting_1",
reg = reg,
fun = simulate_rf_setting_1,
data = n,
seed = 12345)
# batchtools::addProblem(name = "simulate_setting_2",
# reg = reg,
# fun = simulate_rf_setting_2,
# data = n,
# seed = 12345)
# batchtools::addProblem(name = "simulate_setting_3",
# reg = reg,
# fun = simulate_rf_setting_3,
# data = n,
# seed = 12345)
# batchtools::addProblem(name = "simulate_setting_4",
# reg = reg,
# fun = simulate_rf_setting_4,
# data = n,
# seed = 12345)
# batchtools::addProblem(name = "simulate_setting_5",
# reg = reg,
# fun = simulate_rf_setting_5,
# data = n,
# seed = 12345)
## Add algorithms to solve the problem ----
batchtools::addAlgorithm(reg = reg,
name = "distances",
fun = calculate_distances
)
batchtools::addAlgorithm(reg = reg,
name = "decision_tree",
fun = calculate_decision_tree
)
## define problem and algorithm designs ----
n_test <- 100 ## Number of samples in test data set
n_val <- 1000 ## Number of samples in validation data set
p <- 100 ## Number of variables
num_trees <- 500 ## Number of trees in random forest
eps <- 1 ## Simulated noise in data set
mtry <- sqrt(p) ## Mtry for random forest
prob.designs <- list(
simulate_setting_1 = data.frame(p_eff = 5, ## Number of true effect variables
beta_eff = 2, ## Effect size of true effect variables
n_test = n_test,
n_val = n_val,
p = p,
num.trees = num_trees,
eps = 1,
mtry = mtry,
min_node_size = c(10, 50, 100, 200), ## Minimal node sizes in random forest
stringsAsFactors = FALSE
)#,
# simulate_setting_2 = data.frame(p_eff = 50,
# beta_eff = 0.2,
# n_test = n_test,
# n_val = n_val,
# p = p,
# num.trees = num.trees,
# eps = eps,
# mtry = mtry,
# min_node_size = c(10, 50, 100, 200),
# stringsAsFactors = FALSE
# ),
# simulate_setting_3 = data.frame(p_eff = 5,
# beta_eff = 2,
# n_test = n_test,
# n_val = n_val,
# p = p,
# p_corr = 5,
# n_blocks = 5,
# cor = 0.3,
# num.trees = num.trees,
# eps = eps,
# mtry = mtry,
# min_node_size = c(10, 50, 100, 200),
# stringsAsFactors = FALSE
# ),
# simulate_setting_4 = data.frame(p_eff = 5,
# beta_eff = 2,
# n_test = n_test,
# n_val = n_val,
# p = p,
# p_int = 5,
# beta_int = 2,
# num.trees = num.trees,
# eps = eps,
# mtry = mtry,
# min_node_size = c(10, 50, 100, 200),
# stringsAsFactors = FALSE
# ),
# simulate_setting_5 = data.frame(p_eff_bin = 5,
# p_eff_con = 5,
# beta_eff = 2,
# n_test = n_test,
# n_val = n_val,
# p = p,
# num.trees = num.trees,
# eps = eps,
# mtry = mtry,
# min_node_size = c(10, 50, 100, 200),
# stringsAsFactors = FALSE
# )
)
##' For the algorithms you can only define, which metrics to use. We recommend
##' to exclude "terminal nodes" to save time.
algo.designs <- list(
distances = data.frame(metric = c("splitting variables", "weighted splitting variables", "prediction"),
stringsAsFactors = FALSE),
decision_tree = data.frame()
)
# algo.designs <- list(
# distances = data.frame(metric = c("splitting variables", "weighted splitting variables", "prediction", "terminal nodes"),
# stringsAsFactors = FALSE),
# decision_tree = data.frame()
# )
## Add experiments ----
ids = batchtools::addExperiments(reg = reg,
prob.designs = prob.designs,
algo.designs = algo.designs,
repls = 10 ## Number of times each experiment is repeated. You can save time here
)
summarizeExperiments(reg = reg)
## Submit jobs ----
ids <- findNotDone()
ids[, chunk := 1]
## Test jobs before submission
# testJob(id = 1, reg = reg)
## Please change this if you have a batch system.
submitJobs(ids = ids, reg = reg)
##' With pre selected parameters it will take around 10 min to complete.
##' The different metrics will have the following computation time for one replication of setting 1.
##' splitting variables - 20s
##' weighted splitting variables - 21.6s
##' prediction - 24.4s
##' terminal nodes - 2103s
##' decision tree - 4.7s
##' Please note, the run times for the other setting could differ.
##' Anyway simulating data for the figures in the paper will probably run for >100 days on you computer.
getStatus()
## Collect and save results ----
results <- reduceResultsList(reg = reg, missing.val = 0)
## Save results
saveRDS(results, file = file.path(proc_dir, "results.Rds"))