-
Notifications
You must be signed in to change notification settings - Fork 39
/
04-complementarity.R
354 lines (293 loc) · 13.6 KB
/
04-complementarity.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
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
#' Try to find good combinations of methods given a set of datasets, which could indicate some complementarity between methods
library(tidyverse)
library(dynbenchmark)
experiment("10-benchmark_interpretation")
methods <- load_methods()
data <- read_rds(result_file("benchmark_results_normalised.rds", "06-benchmark"))$data %>%
filter(method_id %in% methods$id)
trajectory_type_colours <- dynwrap::trajectory_types %>% select(id, colour) %>% deframe()
# remove datasets on which all methods failed
data_succeeded <- data %>% group_by(dataset_id) %>% summarise(pct_succeeded = mean(error_status == "no_error")) %>% arrange(desc(pct_succeeded))
dataset_sel <- data_succeeded %>% filter(pct_succeeded > 0) %>% pull(dataset_id)
data <- data %>% filter(dataset_id %in% dataset_sel)
## ............................................................................
## Find a "good-enough" subset of methods ####
# several functions to define when a particular method has succeeded on a particular dataset
# is it in range of the top method?
method_subset_preparer_range <- function(data_oi, range = 0.05) {
data_oi <- data_oi %>%
group_by(dataset_id) %>%
mutate(success = overall >= max(overall) * (1-range)) %>%
ungroup()
data_oi
}
# has it a high overall score?
method_subset_preparer_overall_cutoff <- function(data_oi, overall_cutoff = 0.75) {
data_oi <- data_oi %>%
group_by(dataset_id) %>%
mutate(success = overall >= overall_cutoff) %>%
ungroup()
data_oi
}
# does it have the top score?
method_subset_preparer_top <- function(data_oi) {
data_oi <- data_oi %>%
group_by(dataset_id) %>%
mutate(success = overall == max(overall)) %>%
ungroup()
data_oi
}
# score a given set of methods on the datasets, by summing over the weights of the successful datasets
scorer <- function(method_ids, data_oi) {
data_oi %>%
filter(method_id %in% !!method_ids) %>%
filter(success) %>%
group_by(dataset_id) %>%
summarise(weight = first(weight)) %>%
pull(weight) %>%
sum()
}
#' @example
#' trajectory_types_oi <- "tree"
#' data_oi <- data
#' preparer <- method_subset_preparer_range
get_top_methods <- function(data_oi, trajectory_types_oi, preparer = method_subset_preparer_range) {
# filter methods:
# - can detect at least one of the requested trajectory type(s)
# - does not require any hard prior information
hard_priors <- dynwrap::priors %>% filter(type == "hard") %>% pull(prior_id)
relevant_method_ids <- load_methods() %>%
filter(
map_lgl(trajectory_types, ~any(. %in% trajectory_types_oi)),
map_lgl(required_priors, ~!any(. %in% hard_priors))
) %>%
pull(id)
data_oi <- data_oi %>% filter(method_id %in% relevant_method_ids)
# filter datasets
data_oi <- data_oi %>%
filter(dataset_trajectory_type %in% !!trajectory_types_oi) %>% # filter datasets on trajectory type
complete(dataset_id, method_id, fill = list(overall = 0)) # make sure all methods have all datasets, even if this means that they get a zero overall score
# add dataset weights
dataset_weights_oi <- get_dataset_weighting(data_oi %>% distinct(dataset_id, dataset_source, dataset_trajectory_type))
data_oi <- data_oi %>%
left_join(dataset_weights_oi, "dataset_id")
if (nrow(data_oi) == 0) stop("No data found!")
all_method_ids <- unique(data_oi$method_id)
# get the successes for this data_oi
data_oi <- preparer(data_oi)
# add one method step by step
step_ix <- 1
steps <- list()
method_ids <- character()
while (length(method_ids) != length(all_method_ids)) {
print(length(method_ids))
next_steps <- tibble(
method_id = setdiff(all_method_ids, method_ids)
) %>%
mutate(
score = map(method_id, c, method_ids) %>% map_dbl(scorer, data_oi = data_oi)
) %>%
arrange(-score) %>%
mutate(step_ix = step_ix)
chosen_method_id <- next_steps$method_id[[1]]
method_ids <- c(method_ids,chosen_method_id )
steps <- c(
steps,
list(next_steps %>% mutate(chosen = method_id == chosen_method_id))
)
step_ix <- step_ix + 1
}
steps <- bind_rows(steps)
steps
}
## ............................................................................
## Compute step data ####
steps_combinations <- tibble(
trajectory_types_oi = list(
dynwrap::trajectory_types$id,
c("linear", "bifurcation", "multifurcation", "tree"),
c("cycle"),
c("linear"),
c("bifurcation"),
c("multifurcation"),
c("tree"),
c("graph"),
c("disconnected_graph")
)
) %>%
mutate(
combination_id = map_chr(trajectory_types_oi, function(trajectory_types_oi) {
if (all(dynwrap::trajectory_types$id %in% trajectory_types_oi)) {
"All trajectory types"
} else if (length(trajectory_types_oi) > 1) {
paste0(first(trajectory_types_oi), " \U2192 ", last(trajectory_types_oi))
} else {
trajectory_types_oi
}
}) %>% fct_inorder(),
most_complex_trajectory_type = map_chr(trajectory_types_oi, last)
) %>%
mutate(steps = map(trajectory_types_oi, get_top_methods, data = data))
all_steps <-
steps_combinations %>%
unnest(steps)
## ............................................................................
## Plot a single combination of trajectory types ####
plot1_width <- 7
plot_height <- 6
relevant_steps_ex <-
all_steps %>%
filter(combination_id == "All trajectory types") %>%
group_by(step_ix) %>%
filter(any(score < 0.9)) %>%
ungroup() %>%
mutate(y = -step_ix + ifelse(step_ix == 1, .75, 0))
# determine relevant step labels
# some funky plotmath wizardry is performed here
# the labels can look something like this:
#
# > phantom("Running ") * phantom(bold("Slingshot")) * phantom(" on all datasets")
# > "will result in a top model 33% of the time"
relevant_steps_labels_ex <-
relevant_steps_ex %>%
filter(chosen) %>%
mutate(
score_start = lag(score, default = 0),
label_method = label_method(method_id),
label_score = paste0(round(score * 100), "%"),
label = case_when(
step_ix == 1 ~ paste0('phantom("Running ") * phantom(bold("', label_method, '")) * phantom(" on all datasets")\n"will result in a top model ', label_score, ' of the time"'),
step_ix == 2 ~ paste0('"Running both " * bold("', label_method[[1]], '") * " and " * bold("', label_method, '")\n"will result in at least one top model ', label_score, ' of the time"'),
step_ix == 3 ~ paste0('bold("', label_method[[1]], '") * ", " * bold("', label_method[[2]], '") * " and " * bold("', label_method, '")\n"\u21B3 \u2265 1 top model ', label_score, ' of the time"'),
TRUE ~ paste0('"Add " * bold("', label_method, '")\n"\u21B3 ', label_score, '"')
),
label_top = gsub("\\n.*", "", label),
label_bot = gsub(".*\\n", "", label)
)
# compute bracket positions
source(scripts_file("helper-04-complementarity.R"))
bracket_range <- relevant_steps_ex %>% filter(step_ix == 1, !chosen) %>% pull(score) %>% range()
bracket_data <-
bracket(0, 1, 0, .3, flip = TRUE) %>%
mutate(
x = x * diff(bracket_range) + bracket_range[[1]],
y = y - .9
)
plot_complementarity_example <-
ggplot(relevant_steps_ex, aes(score, y)) +
# plot first sentence of the labels to the right. allow phantom(...) text to be plotted, but don't plot phantom(bold(...)) text
geom_text(
aes(x = 1.25, y = y + .375, label = label_top),
relevant_steps_labels_ex %>% mutate(label_top = gsub("phantom\\((\"[^)]*\")\\)", "\\1", label_top)),
hjust = 1, lineheight = 1, parse = TRUE, vjust = 1
) +
# plot first sentence of the labels to the right. allow phantom(bold(...)) text to be plotted in a trajtype colour, but don't plot phantom(...) text
geom_text(
aes(x = 1.25, y = y + .375, label = label_top, colour = most_complex_trajectory_type),
relevant_steps_labels_ex %>% filter(grepl("phantom\\(bold\\(", label_top)) %>% mutate(label_top = gsub("phantom\\((bold\\(\"[^)]*\"\\))\\)", "\\1", label_top)),
hjust = 1, lineheight = 1, parse = TRUE, vjust = 1
) +
# plot second sentence of the labels to the right
geom_text(
aes(x = 1.25, y = y - .375, label = label_bot),
relevant_steps_labels_ex,
hjust = 1, lineheight = 1, parse = TRUE, vjust = 0
) +
# first label line
geom_path(aes(x, y), data_frame(x = c(.31, .6), y = c(-.25, -.25))) +
# bracket and extra text
geom_text(aes(x = 1.25, y = -1.125, label = 'phantom("While ") * bold("other methods") * phantom(" perform less well")'), data_frame(x = 1), hjust = 1, vjust = .5, parse = TRUE, colour = "#AAAAAA") +
geom_text(aes(x = 1.25, y = -1.125, label = '"While " * phantom(bold("other methods")) * " perform less well"'), data_frame(x = 1), hjust = 1, vjust = .5, parse = TRUE) +
geom_path(aes(x, y), bracket_data) +
geom_path(aes(x, y), data_frame(x = c(mean(bracket_range), mean(bracket_range), .6), y = c(-.9, -1.125, -1.125))) +
# red baseline lines
geom_segment(aes(x = score, xend = score, y = -(step_ix + .5 + .1), yend = -(step_ix + .5 + .9), colour = most_complex_trajectory_type), data = filter(relevant_steps_ex, chosen)) +
# points
ggbeeswarm::geom_quasirandom(aes(color = chosen), data = filter(relevant_steps_ex, !chosen), color = "#AAAAAA", size = 1, groupOnX = FALSE) +
# big best point
geom_point(aes(colour = most_complex_trajectory_type), data = relevant_steps_ex %>% filter(chosen)) +
# theme settings
scale_x_continuous(limits = c(0, 1.25), breaks = c(0, 0.25, 0.5, 0.75, 1), labels = scales::percent, expand = c(0, 0)) +
scale_y_continuous(breaks = relevant_steps_labels_ex$y, labels = relevant_steps_labels_ex$step_ix, expand = c(0, 0), limits = range(relevant_steps_ex$y) + c(-.5, .5)) +
scale_color_manual(values = trajectory_type_colours, guide = FALSE) +
theme_pub() +
labs(
x = "Likelihood of obtaining a top model",
y = label_long("n_methods")
) +
lemon::coord_capped_cart(bottom = "both", left = lemon::brackets_vertical(length = .04))
plot_complementarity_example
ggsave(result_file("complementarity_example.pdf"), device = cairo_pdf, plot_complementarity_example, width = plot1_width, height = plot_height)
write_rds(plot_complementarity_example, derived_file("complementarity_example.rds"))
## ............................................................................
## Plot the complementarity of multiple combinations of trajectory_types ####
plot2_width <- 7
comb_ypos <-
steps_combinations %>%
select(combination_id) %>%
mutate(y = as.integer(combination_id), y = max(y) - y + 1)
bar_height <- .33
text_width <- .2
relevant_steps <-
all_steps %>%
group_by(step_ix) %>%
filter(any(round(score, 5) < 1)) %>%
ungroup() %>%
left_join(comb_ypos, by = "combination_id")
relevant_steps_labels <-
relevant_steps %>%
group_by(combination_id) %>%
filter(chosen) %>%
mutate(label = label_wrap(label_method(method_id))) %>%
mutate(score_start = lag(score, default = 0)) %>%
ungroup()
repel_labels <- relevant_steps_labels %>% filter(step_ix < 3 | (score < 0.9 & step_ix < 7))
plot_complementarity_combinations <-
ggplot(relevant_steps) +
geom_rect(aes(xmin = score_start, xmax = score, ymin = y, ymax = y + bar_height, fill = factor(step_ix)), data = relevant_steps_labels) +
geom_rect(aes(xmin = 0, xmax = 1, ymin = y, ymax = y + bar_height), comb_ypos, colour = "#888888", fill = NA) +
geom_segment(aes(x = score, xend = score, y = y, yend = y + bar_height), color = "black", alpha = 0.1) +
geom_point(aes(x = score, y = y, color = most_complex_trajectory_type), data = relevant_steps %>% filter(chosen)) +
geom_text(aes(-text_width / 2, y + bar_height / 2, label = label_short(combination_id, width = 15)), comb_ypos, hjust = .5, lineheight = 1) +
ggrepel::geom_text_repel(
aes(x = score, y = y, label = label),
data = repel_labels,
nudge_y = -.1,
vjust = 1,
direction = "x",
lineheight = .85,
min.segment.length = 0,
force = 20,
max.iter = 10000
) +
geom_segment(aes(x = score, xend = score, y = y, yend = y + bar_height, color = most_complex_trajectory_type), data = relevant_steps %>% filter(chosen)) +
scale_y_continuous(limits = c(bar_height, max(comb_ypos$y) + .5), breaks = NULL, expand = c(0, 0)) +
scale_x_continuous(limits = c(-text_width, 1), breaks = c(0, 0.25, 0.5, 0.75, 1), labels = scales::percent, expand = c(0, .03)) +
scale_fill_grey(label_long("n_methods"), start = 0.95, end = 0.2, limits = c(1, 2,3,4,5,6), na.value = "#333333") +
scale_color_manual(values = trajectory_type_colours, guide = FALSE) +
theme_pub() +
theme(
axis.title.y = element_blank(),
axis.line.y = element_blank(),
strip.text.y = element_text(angle = 180, vjust = 1, size = 10),
strip.placement.y = "left",
strip.background.y = element_blank(),
panel.spacing = unit(0, "cm"),
legend.position = "top",
legend.justification = "center"
) +
guides(fill = guide_legend(ncol = 6, label.position = "right")) +
labs(x = "Likelihood of obtaining a top model") +
lemon::coord_capped_cart(bottom = "both")
write_rds(plot_complementarity_combinations, derived_file("complementarity_combinations.rds"))
## ............................................................................
## Combined complementarity plot ####
plot_width <- plot1_width + plot2_width
plot_complementarity <- patchwork::wrap_plots(
read_rds(derived_file("complementarity_example.rds")),
read_rds(derived_file("complementarity_combinations.rds")),
nrow = 1,
widths = c(plot1_width, plot2_width) / plot_width
) +
patchwork::plot_annotation(tag_levels = "a")
ggsave(result_file("complementarity.pdf"), plot_complementarity, width = plot_width, height = plot_height, device = cairo_pdf)