-
Notifications
You must be signed in to change notification settings - Fork 0
/
discrete_BN_validation.Rmd
459 lines (387 loc) · 15.8 KB
/
discrete_BN_validation.Rmd
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
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
---
title: "Model Validation of discrete BN"
author: "Matteo Delucchi"
output:
rmarkdown::html_document:
toc: yes
toc_float:
collapsed: true
smooth_scroll: false
toc_depth: 3
rmarkdown::html_vignette: default
vignette: >
%\VignetteIndexEntry{Model Validation of discrete BN}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>"
)
```
```{r development, eval=FALSE, include=FALSE}
# devtools::document()
# devtools::load_all()
# renv::snapshot(prompt = F)
```
```{r setup}
# Clear working environment
rm(list=ls())
# Load libraries
library(bnaiaR)
library(tidyr)
library(dplyr)
library(ggplot2)
library(parallel)
library(bnlearn)
library(forcats)
# Save plots as files?
SAVEPLOTS <- F
PLOTPATH <- Sys.getenv("PLOTPATH")
PLOTFORMAT <- "png" # "svg" or "png"
```
# Load discrete BNs
The object `discrete_bns` was generated in vignette `discrete_BN_SL`.
```{r message=FALSE, warning=FALSE}
data <- exp11_dat$abndata
bl <- exp11_dat$bl
avgnet <- discrete_bns$avgnet
arcstren <- discrete_bns$arcstren
# str(avgnet)
```
```{r}
strength.plot(avgnet[["tabu.th03"]],
arcstren[["tabu"]],
main = "DAG",
sub = "SL: tabu\nth.=0.3\nScore=BIC",
shape = "rectangle")
strength.plot(avgnet[["tabu.bic.custom"]],
arcstren[["tabu.bic.custom"]],
main = "DAG",
sub = "SL: tabu\nth.=sign.level\nScore=BIC-custom",
shape = "rectangle")
```
# Cross-validation for predictive accuracy
## Classification Error of rupture status.
```{r}
xval.rupture.bic.part <- bn.cv(data = data,
bn = avgnet$tabu,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred",
loss.args = list(target="Ruptured_IA"))
xval.rupture.bic <- bn.cv(
data = data,
bn = avgnet$tabu.th03,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred",
loss.args = list(target = "Ruptured_IA")
)
xval.rupture.bic.custom <- bn.cv(
data = data,
bn = avgnet$tabu.bic.custom,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred",
loss.args = list(target = "Ruptured_IA")
)
df.cvmetrics <- data.frame(loocv.dbn.clerr.bic.part = c(avgClassError = mean(loss(xval.rupture.bic.part)), sdClassError = sd(loss(xval.rupture.bic.part)), cv.metrics(xval.rupture.bic.part, returnConfMat = T)$carret_confmat$overall, cv.metrics(xval.rupture.bic.part, returnConfMat = T)$carret_confmat$byClass),
loocv.dbn.clerr.bic = c(avgClassError = mean(loss(xval.rupture.bic)), sdClassError = sd(loss(xval.rupture.bic)),cv.metrics(xval.rupture.bic, returnConfMat = T)$carret_confmat$overall, cv.metrics(xval.rupture.bic, returnConfMat = T)$carret_confmat$byClass),
loocv.dbn.clerr.custom = c(avgClassError = mean(loss(xval.rupture.bic.custom)), sdClassError = sd(loss(xval.rupture.bic.custom)),cv.metrics(xval.rupture.bic.custom, returnConfMat = T)$carret_confmat$overall, cv.metrics(xval.rupture.bic.custom, returnConfMat = T)$carret_confmat$byClass))
df.cvmetrics
```
```{r}
if (SAVEPLOTS) {
FILE <- paste0(PLOTPATH, "/", "classification_error_ruptureStatus_all", ".")
if (PLOTFORMAT == "svg") {
svg(paste0(FILE, PLOTFORMAT))
} else if (PLOTFORMAT == "png") {
png(paste0(FILE, PLOTFORMAT))
}
plot(xval.rupture.bic.part, xval.rupture.bic, xval.rupture.bic.custom,
xlab = c("BIC.signTH", "BIC.th03", "BIC.custom"),
main = "Classification Error of Rupture Status")
dev.off()
} else {
plot(xval.rupture.bic.part, xval.rupture.bic, xval.rupture.bic.custom,
xlab = c("BIC.signTH", "BIC.th03", "BIC.custom"),
main = "Classification Error of Rupture Status")
}
```
## Averaged, Overall Classification Error of Rupture Status
Average predictive accuracy over all nodes.
```{r}
xval.list.bic <- list()
for (node in nodes(avgnet$tabu.th03)) {
xval <- bn.cv(
data = data,
bn = avgnet$tabu.th03,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred",
loss.args = list(target = node)
)
xval.list.bic[[node]] <- list(xval = xval,
cv.metrics = cv.metrics(xval))
}
xval.list.bic.custom <- list()
for (node in nodes(avgnet$tabu.th03)) {
xval <- bn.cv(
data = data,
bn = avgnet$tabu.bic.custom,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred",
loss.args = list(target = node)
)
xval.list.bic.custom[[node]] <- list(xval = xval,
cv.metrics = cv.metrics(xval))
}
accs.bic <- unlist(lapply(unlist(xval.list.bic, recursive = F), `[[`, "acc"))
summary(accs.bic, na.rm = T)
accs.bic.custom <- unlist(lapply(unlist(xval.list.bic.custom, recursive = F), `[[`, "acc"))
summary(accs.bic.custom, na.rm = T)
```
BIC.th03 used three parents (pos.fam.hist, size, location) to predict rupture
whereas BIC.custom uses only the later two parent nodes to predict rupture.
From `?bnlearn::bn.cv()`:
"...
Classification Error (pred): the prediction error for a single node in a discrete network. Frequentist predictions are used, so the values of the target node are predicted using only the information present in its local distribution (from its parents). Lower values are better.
Posterior Classification Error (pred-lw and pred-lw-cg): similar to the above, but predictions are computed from an arbitrary set of nodes using likelihood weighting to obtain Bayesian posterior estimates. pred-lw applies to discrete Bayesian networks, pred-lw-cg to (discrete nodes in) hybrid networks. Lower values are better.
..."
## Posterior Classification Error of rupture status
So, we can use a Bayesian predictions that use a random set of available nodes for the prediction.
```{r}
xval.bic.bayes.part <- bn.cv(data = data,
bn = avgnet$tabu,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred-lw",
loss.args = list(target="Ruptured_IA"))
xval.bic.bayes <- bn.cv(data = data,
bn = avgnet$tabu.th03,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred-lw",
loss.args = list(target="Ruptured_IA"))
xval.bic.custom.bayes <- bn.cv(data = data,
bn = avgnet$tabu.bic.custom,
method = "hold-out",
runs = 50,
fit = "mle",
loss = "pred-lw",
loss.args = list(target="Ruptured_IA"))
cv.metrics(xval.bic.bayes.part)
cv.metrics(xval.bic.bayes)
cv.metrics(xval.bic.custom.bayes)
df.cvmetrics.postclerr <- data.frame(loocv.dbn.postclerr.bic.part = c(avgPostClassError = mean(loss(xval.bic.bayes.part)), sdClassError = sd(loss(xval.bic.bayes.part)), cv.metrics(xval.bic.bayes.part, returnConfMat = T)$carret_confmat$overall, cv.metrics(xval.bic.bayes.part, returnConfMat = T)$carret_confmat$byClass),
loocv.dbn.postclerr.bic = c(avgPostClassError = mean(loss(xval.bic.bayes)), sdClassError = sd(loss(xval.bic.bayes)), cv.metrics(xval.bic.bayes, returnConfMat = T)$carret_confmat$overall, cv.metrics(xval.bic.bayes, returnConfMat = T)$carret_confmat$byClass),
loocv.dbn.postclerr.custom = c(avgPostClassError = mean(loss(xval.bic.custom.bayes)), sdClassError = sd(loss(xval.bic.custom.bayes)), cv.metrics(xval.bic.custom.bayes, returnConfMat = T)$carret_confmat$overall, cv.metrics(xval.bic.custom.bayes, returnConfMat = T)$carret_confmat$byClass))
df.cvmetrics <- cbind(df.cvmetrics, df.cvmetrics.postclerr)
df.cvmetrics <- round(df.cvmetrics, 2)
df.cvmetrics
```
Save classification metrics
```{r}
write.csv(df.cvmetrics, file = paste0(PLOTPATH, "/prediction_metrics_dbn.csv"))
```
```{r}
if (SAVEPLOTS) {
FILE <- paste0(PLOTPATH, "/", "post_classification_error_ruptureStatus_all", ".")
if (PLOTFORMAT == "svg") {
svg(paste0(FILE, PLOTFORMAT))
} else if (PLOTFORMAT == "png") {
png(paste0(FILE, PLOTFORMAT), res = 300)
} else if (PLOTFORMAT == "pdf") {
pdf(paste0(FILE, PLOTFORMAT))
}
plot(xval.bic.bayes.part, xval.bic.bayes, xval.bic.custom.bayes,
xlab = c("BIC.signTH", "BIC.th03", "BIC.custom"),
main = "Posterior Classification Error of Rupture Status")
dev.off()
} else {
plot(xval.bic.bayes.part, xval.bic.bayes, xval.bic.custom.bayes,
xlab = c("BIC.signTH", "BIC.th03", "BIC.custom"),
main = "Posterior Classification Error of Rupture Status")
}
```
# Inference by querying
```{r}
tabu_th03_fit <- bn.fit(data = data, x = avgnet$tabu.th03)
```
```{r}
plot(tabu_th03_fit$location.grouped$prob)
```
Males tend to have an IA more often at a high risk location but
Females tend to develop an IA in high risk location more often if they suffer HBP than males who suffer HBP.
```{r}
plot(tabu_th03_fit$IAsize.groups.merged$prob)
```
Being aware of pos. fam. hist. reduces the porbability of developing larger IAs.
```{r}
(tabu_th03_fit$Ruptured_IA$prob)
```
Larger IAs tend to rupture more frequently.
```{r cpdist_age_hypertension_rupture}
par(mfrow = c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("Hypertension", "AgeDiag.group"), n = 10^4, evidence = (Ruptured_IA == "Yes"))
plot(sim, col = "grey",
main = "Ruptured_IA == Yes")
sim = cpdist(tabu_th03_fit, nodes = c("Hypertension", "AgeDiag.group"), n = 10^4, evidence = (Ruptured_IA == "No"))
plot(sim, col = "grey",
main = "Ruptured_IA == No")
par(mfrow = c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group", "Hypertension"), n = 10^4, evidence = (Ruptured_IA == "Yes"))
plot(sim, col = "grey",
main = "Ruptured_IA == Yes")
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group", "Hypertension"), n = 10^4, evidence = (Ruptured_IA == "No"))
plot(sim, col = "grey",
main = "Ruptured_IA == No")
par(mfrow = c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group"), n = 10^4, evidence = (Hypertension == "AnyType"))
plot(sim, col = "grey",
main = "Hypertension == AnyType")
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group"), n = 10^4, evidence = (Hypertension == "Never"))
plot(sim, col = "grey",
main = "Hypertension == Never")
par(mfrow = c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group"), n = 10^5, evidence = (Ruptured_IA == "Yes"))
plot(prop.table(table(sim)), col = "grey",
main = "Ruptured_IA == Yes")
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group"), n = 10^5, evidence = (Ruptured_IA == "No"))
plot(prop.table(table(sim)), col = "grey",
main = "Ruptured_IA == No")
SAVEPLOTS <- F
if (SAVEPLOTS){
PLOTNAME <- "infquerry_rupture-age"
dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
res = 600, width = 6000)
dev.off()
}
```
```{r}
SAVEPLOTS <- F
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("Smoking_Current_Former_No", "IAsize.groups.merged"), n = 10^5, evidence = (Ruptured_IA == "Yes"))
plot(prop.table(table(sim)), col = "grey",
main = 'Ruptured_IA == "Yes"')
sim = cpdist(tabu_th03_fit, nodes = c("Smoking_Current_Former_No", "IAsize.groups.merged"), n = 10^5, evidence = (Ruptured_IA == "No"))
plot(prop.table(table(sim)), col = "grey",
main = 'Ruptured_IA == "No"')
if (SAVEPLOTS){
PLOTNAME <- "infquerry_rupture-smoking-size"
dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
res = 600, width = 6000)
dev.off()
}
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("Smoking_Current_Former_No", "AgeDiag.group"), n = 10^5, evidence = (Ruptured_IA == "Yes"))
plot(prop.table(table(sim)), col = "grey",
main = 'Ruptured_IA == "Yes"')
sim = cpdist(tabu_th03_fit, nodes = c("Smoking_Current_Former_No", "AgeDiag.group"), n = 10^5, evidence = (Ruptured_IA == "No"))
plot(prop.table(table(sim)), col = "grey",
main = 'Ruptured_IA == "No"')
if (SAVEPLOTS){
PLOTNAME <- "infquerry_rupture-smoking-age"
dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
res = 600, width = 6000)
dev.off()
}
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("IAsize.groups.merged", "AgeDiag.group"), n = 10^5, evidence = (Smoking_Current_Former_No == "Current"))
plot(prop.table(table(sim)), col = "grey",
main = 'Smoking_Current_Former_No == "Current"')
sim = cpdist(tabu_th03_fit, nodes = c("IAsize.groups.merged", "AgeDiag.group"), n = 10^5, evidence = (Smoking_Current_Former_No == "Current" & Ruptured_IA == "Yes"))
plot(prop.table(table(sim)), col = "grey",
main = 'Smoking_Current_Former_No == "Current" & Ruptured_IA == "Yes"')
if (SAVEPLOTS){
PLOTNAME <- "infquerry_smoking-age-size"
dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
res = 600, width = 6000)
dev.off()
}
```
```{r}
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("Gender", "location.grouped"), n = 10^4, evidence = (Ruptured_IA == "Yes"))
plot(prop.table(table(sim)),
main = 'Ruptured_IA == "Yes"')
prop.table(table(sim))
sim = cpdist(tabu_th03_fit, nodes = c("Gender", "location.grouped"), n = 10^4, evidence = (Ruptured_IA == "No"))
plot(prop.table(table(sim)),
main = 'Ruptured_IA == "No"')
prop.table(table(sim))
if (SAVEPLOTS){
PLOTNAME <- "infquerry_rupture-location-gender"
dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
res = 600, width = 6000)
dev.off()
}
```
```{r}
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group", "Ruptured_IA"), n = 10^4, evidence = (Smoking_Current_Former_No == "Current" & Hypertension == "AnyType" & Positive.famillial.history == "Yes"))
plot(prop.table(table(sim)), col = "grey",
main = 'Smoking_Current_Former_No == "Current", Hypertension = "AnyType", Positive.famillial.history = "Yes"')
sim = cpdist(tabu_th03_fit, nodes = c("AgeDiag.group", "Ruptured_IA"), n = 10^4, evidence = (Smoking_Current_Former_No == "No" & Hypertension == "Never" & Positive.famillial.history == "No"))
plot(prop.table(table(sim)), col = "grey",
main = 'Smoking_Current_Former_No == "No" & Hypertension == "Never" & Positive.famillial.history == "No"')
if (SAVEPLOTS){
PLOTNAME <- "infquerry_rupture-age-smoking-hbp-family"
dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
res = 600, width = 6000)
dev.off()
}
```
```{r}
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("Multiple.IAs"), n = 10^4, evidence = (Hypertension == "AnyType"))
barplot(prop.table(table(sim)), col = "grey",
ylim = c(0,0.8),
main = "P(Multiple.IAs | Hypertension == 'AnyType')")
prop.table(table(sim))
sim = cpdist(tabu_th03_fit, nodes = c("Multiple.IAs"), n = 10^4, evidence = (Hypertension == "Never"))
barplot(prop.table(table(sim)), col = "grey",
ylim = c(0,0.8),
main = "P(Multiple.IAs | Hypertension == 'Never')")
prop.table(table(sim))
if (SAVEPLOTS){
PLOTNAME <- "infquerry_multiple_hbp"
# dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
# res = 600, width = 6000)
dev.print(pdf, file =paste0(PLOTPATH, "/", PLOTNAME, ".pdf"),
width = 16, height = 9)
dev.off()
}
par(mfrow=c(1,2))
sim = cpdist(tabu_th03_fit, nodes = c("location.grouped"), n = 10^4, evidence = (Hypertension == "AnyType"))
barplot(prop.table(table(sim)), col = "grey",
ylim = c(0,0.5),
main = "P(IA Location | Hypertension == 'AnyType')")
prop.table(table(sim))
sim = cpdist(tabu_th03_fit, nodes = c("location.grouped"), n = 10^4, evidence = (Hypertension == "Never"))
barplot(prop.table(table(sim)), col = "grey",
ylim = c(0,0.5),
main = "P(IA Location | Hypertension == 'Never')")
prop.table(table(sim))
if (SAVEPLOTS){
PLOTNAME <- "infquerry_location_hbp"
# dev.print(png, filename =paste0(PLOTPATH, "/", PLOTNAME, ".png"),
# res = 600, width = 6000)
dev.print(pdf, file =paste0(PLOTPATH, "/", PLOTNAME, ".pdf"),
width = 16)
dev.off()
}
```
```{r}
str(data)
```