-
Notifications
You must be signed in to change notification settings - Fork 3
/
sim_basic.R
151 lines (125 loc) · 4.17 KB
/
sim_basic.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
# This simulation is for examining variance estimates under a basic exogenous LIM model.
library(igraph)
library(foreach)
library(doParallel)
library(dplyr)
library(tidyr)
library(broom)
source('functions/data_generators.R')
source('functions/covariate_functions.R')
source('functions/response_functions.R')
source('functions/existing_estimators.R')
source('functions/proposed_estimators.R')
source('functions/variance_estimators.R')
source('functions/precompute_matrices.R')
run_sim = function(param, g, variance_factor, n_reps, n_cores, pid=NULL) {
# param: row of parameters
# creates a list containing the data
covariate_fns_for_response=list(
frac_nbh = fraction_trt_nbrs,
frac_nbh2 = fraction_trt_nbrs2
)
data = generate_covariate_data(g, covariate_fns_for_response)
# generate response
data$y = linear_response(data$w, data$x_obs, param)
return(c(
pid=pid,
dm=data %>% difference_in_means,
dm_var_est = data %>% dm_variance_estimate,
adjusted=data %>% linear_adjustment,
var_est=data %>% linear_variance_estimate(variance_factor)
))
}
load('data/caltech.Rdata')
n_reps = 1000
n_cores = 30
registerDoParallel(cores=n_cores)
covariate_fns_for_estimator=list(
frac_nbh = fraction_trt_nbrs,
frac_nbh2 = fraction_trt_nbrs2
)
start = proc.time()
vf = precompute_variance(g_fb, covariate_fns_for_estimator, n_boot_reps=200, n_cores=n_cores)
vf
print(proc.time() - start)
params = purrr::cross(list(
beta0 = list(c(0, 0, 0), c(0, 0, 0.01), c(0, 0.1, 0), c(0, 0.1, 0.01)),
beta1 = list(c(1, 0, 0), c(1, 0, 0.05), c(1, 0.2, 0), c(1, 0.2, 0.05)),
noise_sd = 1
))
print('Running simulation...')
# Run simulation
start = proc.time()
estimates = foreach(i = 1:length(params), .combine=rbind) %do% {
param = params[[i]]
print(unlist(param))
foreach(rep = 1:n_reps, .combine=rbind) %dopar% {
run_sim(param, g_fb, vf, n_reps, n_cores, pid = i)
}
} %>% data.frame
print(proc.time())
write.csv(estimates, file='results/sim_basic.csv')
d_bar = mean(degree(g_fb))
truth = sapply(params, function(p) {with(p, beta1[1] - beta0[1] + beta1[2] + beta1[3])})
df_dm = estimates %>%
select(pid, estimate=dm, var_est=dm_var_est) %>%
mutate(
truth=truth[pid],
accepts=abs(estimate - truth) / sqrt(var_est) < qnorm(0.95)
) %>%
group_by(pid) %>%
summarise(
truth = first(truth),
mean=mean(estimate),
bias=mean - truth,
sd=sd(estimate),
avg_sd_est = mean(sqrt(var_est)),
sd_ratio = avg_sd_est / sd,
coverage=mean(accepts)
) %>%
ungroup
df_adjusted = estimates %>%
select(pid, estimate=adjusted, var_est=var_est) %>%
mutate(
truth=truth[pid],
accepts=abs(estimate - truth) / sqrt(var_est) < qnorm(0.95)
) %>%
group_by(pid) %>%
summarise(
truth = first(truth),
mean=mean(estimate),
bias=mean - truth,
sd=sd(estimate),
avg_sd_est = mean(sqrt(var_est)),
sd_ratio = avg_sd_est / sd,
coverage=mean(accepts)
) %>%
ungroup
cbind(
df_dm %>% select(pid, truth, bias_dm=bias, sd_dm = sd, sd_ratio_dm=sd_ratio, coverage_dm=coverage),
df_adjusted %>% select(bias_adj=bias, sd_adj = sd, sd_ratio_adj=sd_ratio, coverage_adj=coverage)
) %>% filter(pid < 17) %>%
select(pid, truth, bias_dm, bias_adj, sd_dm, sd_adj, sd_ratio_dm, sd_ratio_adj, coverage_dm, coverage_adj) %>%
xtable(digits=3) %>% print.xtable(include.rownames=FALSE)
foreach(p = params[1:16], .combine=rbind) %do% {c(p$beta0[2:3], p$beta1[2:3])} %>% xtable %>% print.xtable(include.rownames=FALSE)
summarised = rbind(
estimates %>% select(pid, estimate=dm, var_est=dm_var_est) %>% mutate(estimator='dm'),
estimates %>% select(pid, estimate=adjusted, var_est=var_est) %>% mutate(estimator='adjusted')
) %>% mutate(
truth=truth[pid],
accepts=abs(estimate - truth) / sqrt(var_est) < qnorm(0.95)
) %>%
group_by(pid, estimator) %>%
summarise(
truth = first(truth),
mean=mean(estimate),
bias=mean - truth,
sd=sd(estimate),
avg_sd_est = mean(sqrt(var_est)),
sd_ratio = avg_sd_est / sd,
coverage=mean(accepts)
) %>%
ungroup
summarised %>%
select(estimator, bias, sd, sd_ratio, coverage)
tmp = cbind(summarised %>% filter(estimator == 'adjusted'), summarised %>% filter(estimator == 'dm'))