-
Notifications
You must be signed in to change notification settings - Fork 0
/
PrecRecF1.R
executable file
·151 lines (132 loc) · 6.29 KB
/
PrecRecF1.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
#!/usr/bin/Rscript
require(ggplot2)
require(dplyr)
require(RColorBrewer)
#source("~/Projects/R/fs_.R")
#setwd("..")
args <- commandArgs(trailingOnly = TRUE)
target_alpha <- args[1] %>% as.numeric
cat("target alpha", target_alpha, "\n")
# Precision, recall and F1 for interaction terms
ans <- lapply(list.files(path = "./fits_proper/", pattern = "", full.names = TRUE), function(f) {
ans <- readRDS(f)
n <- regmatches(x = f, m = regexpr(f, pattern = "(?<=n)\\d+(?=_)", perl = TRUE)) %>% as.numeric
p <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_p)\\d+(?=_)", perl = TRUE)) %>% as.numeric
SNR <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_SNR)\\d+(?=_)", perl = TRUE)) %>% as.numeric
nbi <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_nbi)\\d+(?=_)", perl = TRUE)) %>% as.numeric
nbij <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_nbij)\\d+(?=_)", perl = TRUE)) %>% as.numeric
perc_viol <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_viol)\\d+(?=_)", perl = TRUE)) %>% as.numeric
regression_alpha <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_alpha)[\\d.]+(?=_)", perl = TRUE)) %>% as.numeric
ID <- regmatches(x = f, m = regexpr(f, pattern = "(?<=_)\\d+(?=\\.rds)", perl = TRUE)) %>% as.numeric
tp <- ans$tp
count <- ans$count
notest <- data.frame(n = n, p = p, SNR = SNR, regression_alpha, nbi = nbi, nbij = nbij,
precision = tp / count,
recall = tp / nbij) %>%
mutate(F1 = 2 * (precision * recall) / (precision + recall),
test = "no")
test <- data.frame(n = n, p = p, SNR = SNR, regression_alpha, nbi = nbi, nbij = nbij,
precision = tp / count,
recall = tp / nbij) %>%
mutate(F1 = 2 * (precision * recall) / (precision + recall),
test = "yes")
rbind(notest, test) %>% data.frame(., id = ID)
}) %>% do.call("rbind", .) %>%
tbl_df %>%
mutate(n = factor(n),
p = factor(p),
SNR = factor(SNR),
regression_alpha = factor(regression_alpha),
nbi = factor(nbi),
nbij = factor(nbij))
saveRDS(ans, file = "PrecRecF1/dat_precrecf1.rds")
# or we could calculate dat_precrecf1.rds beforehand.
for (numrows in c(1000)) { #400
for (t in c("yes", "no")) {
dat_precrecf1 <- readRDS(file = "PrecRecF1/dat_precrecf1.rds") %>%
filter(n == numrows) %>%
filter(test == t) %>%
filter(nbi %in% c("0", "20", "50", "100")) %>%
filter(nbij %in% c("5", "20", "50", "100")) %>%
filter(SNR != "1") %>%
filter(regression_alpha == target_alpha) %>%
mutate(SNR = factor(SNR, levels = levels(SNR), labels = paste0("SNR = ", levels(SNR)))) %>%
group_by(n, p, SNR, nbi, nbij, test) #%>% sample_n(10, replace = TRUE)
#TODO: replace = FALSE when we have a large enough data set.
pl <- group_by(dat_precrecf1, SNR, nbi, nbij) %>%
summarise(mean = mean(precision, na.rm = TRUE), sd = sd(precision, na.rm = TRUE) / sqrt(n())) %>%
mutate(Measure = "Precision") %>%
rbind(., group_by(dat_precrecf1, SNR, nbi, nbij) %>%
summarise(mean = mean(recall, na.rm = TRUE), sd = sd(recall, na.rm = TRUE) / sqrt(n())) %>%
mutate(Measure = "Recall")) %>%
rbind(., group_by(dat_precrecf1, SNR, nbi, nbij) %>%
summarise(mean = mean(F1, na.rm = TRUE), sd = sd(F1, na.rm = TRUE) / sqrt(n())) %>%
mutate(Measure = "F1")) %>%
mutate(Measure = factor(Measure, levels = c("Precision", "Recall", "F1"))) %>%
ggplot(aes(x = nbij,
y = mean,
group = nbi,
ymax = mean + sd,
ymin = mean - sd)) +
facet_grid(Measure ~ SNR) +
# geom_bar(aes(fill = nbi), stat = "identity", position = "dodge") +
geom_line(aes(colour = nbi), position=position_dodge(.35)) +
geom_point(aes(colour = nbi), position=position_dodge(.35), size = 1) +
geom_errorbar(colour = "darkgrey", width=.3, position=position_dodge(.35)) +
scale_color_discrete(name = "True additional\nmain effects") +
# scale_colour_manual(name = "Additional main effects", values = c("#7fcdbb", "#1d91c0", "#253494")) + #brewer.pal(7, "Set3")[-1]) + #
ylim(c(0,1)) +
xlab("True interactions") +
ylab("") +
# theme_fs() +
theme(legend.position = "bottom")
pl
ggsave(pl, file = sprintf("PrecRecF1/PrecRecF1_n%d_t%s_a%.2f.pdf", numrows, t, target_alpha), width = 3, height = 4)
}
}
#
#
#
require(tidyr)
require(gridExtra)
#source("~/Projects/R/fs_.R")
for (numrows in c(400, 1000)) {
dat <- readRDS(file = "PrecRecF1/dat_precrecf1.rds") %>%
filter(n == numrows) %>%
mutate(test = factor(test)) %>%
group_by(n, p, SNR, nbi, nbij, test) %>%
summarise(precision = mean(precision, na.rm = TRUE), recall = mean(recall, na.rm = TRUE)) %>%
gather(measure, value, precision:recall) %>%
ungroup %>%
spread(test, value) %>%
filter(SNR == 5) %>%
filter(nbi %in% c(0, 20, 50, 100)) %>%
filter(nbij %in% c("5", "20", "50", "100"))
pl.prec <- group_by(subset(dat, measure == "precision"), n, p, nbi, nbij, measure) %>%
summarise(change = yes - no) %>%
ggplot(aes(x = nbij, y = change)) +
geom_bar(aes(fill = nbi), stat = "identity", position = "dodge") +
# facet_wrap(~measure, scale = "free_y", ncol = 1) +
# scale_fill_manual(name = "Additional\nmain effects", values = c("#7fcdbb", "#1d91c0", "#253494")) +
scale_fill_discrete(name = "True additional\nmain effects") +
ylim(c(0, 0.8)) +
xlab("True interactions") +
ylab("Precision") +
# theme_fs() +
theme(legend.position = "bottom")
pl.rec <- group_by(subset(dat, measure == "recall"), n, p, nbi, nbij, measure) %>%
summarise(change = yes - no) %>%
ggplot(aes(x = nbij, y = change)) +
geom_bar(aes(fill = nbi), stat = "identity", position = "dodge") +
# facet_wrap(~measure, scale = "free_y", ncol = 1) +
# scale_fill_manual(name = "Additional\nmain effects", values = c("#7fcdbb", "#1d91c0", "#253494")) +
scale_fill_discrete(name = "True additional\nmain effects") +
ylim(c(-0.8, 0)) +
xlab("True interactions") +
ylab("Recall") +
# theme_fs() +
theme(legend.position = "bottom")
pdf(sprintf("PrecRecF1/test_analysis_n%d.pdf", numrows), width = 3, width = 3)
grid.arrange(pl.prec, pl.rec, ncol = 1)
dev.off()
}