/
run_mixture_mcmc.R
162 lines (147 loc) · 6.61 KB
/
run_mixture_mcmc.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
source('mixture_analysis.R')
get_stan_data = function(model_file, obs){
if(model_file == 'working_countrywide_mixture.stan'){
stan_data = with(obs, list(
N = nrow(obs),
D = length(unique(location_variable)),
R = length(unique(driver_race)),
K = length(unique(state)),
state = as.integer(state),
d = as.integer(location_variable),
r = as.integer(driver_race),
n = num_stops,
s = num_searches,
h = num_hits
))
}else if(model_file %in% c(old_stop_model_name, new_stop_model_name)){
# add a couple asserts to make sure data is ordered correctly.
error_message = 'Data is not formatted correctly for the stop model. Every location should have data for all races, and dataframe should be sorted by location_variable.'
if(!(as.character(all_equal(obs, obs %>% arrange(location_variable), ignore_row_order = FALSE)) == 'TRUE')){stop(error_message)}
if(!all(obs %>% group_by(location_variable) %>% summarise(n = n()) %>% .$ n %>% unique() == length(unique(obs$driver_race)))){stop(error_message)}
if(!all(obs %>% group_by(driver_race) %>% summarise(n = n()) %>% .$ n %>% unique() == length(unique(obs$location_variable)))){stop(error_message)}
# package data for Stan.
stan_data = with(obs, list(
N = nrow(obs),
D = length(unique(location_variable)),
R = length(unique(driver_race)),
d = as.integer(location_variable),
r = as.integer(driver_race),
base_population_proportions = base_population,
s = num_searches,
h = num_hits
))
}else if(model_file %in% c(old_frisk_model_name, new_frisk_model_name)){
stan_data = with(obs, list(
N = nrow(obs),
D = length(unique(location_variable)),
R = length(unique(driver_race)),
d = as.integer(location_variable),
r = as.integer(driver_race),
n = num_stops,
s = num_searches,
h = num_hits
))
}else{
message("Error: not a valid model name.")
stopifnot(FALSE)
}
return(stan_data)
}
get_stan_initialization = function(model_file, stan_data){
if(model_file %in% c(new_frisk_model_name, new_stop_model_name)){
init_fn = 'random' # checked. No special initialization needed.
}else if(model_file %in% c(old_frisk_model_name, old_stop_model_name)){
initializer <- function(num_obs, num_races, num_locations) {
# force immediate evaluation of arguments
force(num_obs); force(num_races); force(num_locations);
function() {
list(sigma_t = runif(num_races, 0.05, 0.25),
t_r = runif(num_races, -3, -1),
t_i_raw = runif(num_obs, -0.75, 0.75),
phi_r = runif(num_races, -3.75, -3.25),
lambda_r = runif(num_races, 1.5, 2.5),
phi_d_raw = runif(num_locations-1, -0.25, 0.25),
lambda_d_raw = runif(num_locations-1, -0.25, 0.25)
)
}
}
init_fn <- initializer(stan_data$N, stan_data$R, stan_data$D)
}else if(model_file == 'working_countrywide_mixture.stan'){ # this gives an example of how to fit the model on the national traffic stops data.
initializer <- function(num_obs, num_races, num_depts, num_states) {
# force immediate evaluation of arguments
force(num_obs); force(num_races); force(num_depts); force(num_states);
function() {
list(sigma_t = runif(1, 0.1, 1),
phi_r = runif(num_races, -5, -3),
phi_state = runif(num_states, -.25, -.25),
sigma_phi = runif(num_states, 0.01, .1),
phi_d = runif(num_depts, -0.25, 0.25),
delta_r = runif(num_races, -.3, .3),
delta_state = runif(num_states, -.3, .3),
sigma_delta = runif(num_states, .01, .1),
delta_d = runif(num_depts, -0.25, .25)
)
}
}
init_fn <- initializer(stan_data$N, stan_data$R, stan_data$D, stan_data$K)
}else{
message("Error: not a valid model name.")
stopifnot(FALSE)
}
return(init_fn)
}
run_mixture_mcmc <- function(stops, output_filename, iter = 5000, warmup = NULL, chains = 5, adapt_delta = 0.95, max_treedepth = 12, sample_from_prior = FALSE, verbose = FALSE, simulation=FALSE, model_file = 'mixture_model.stan') {
# checked.
if (is.null(warmup)) {
if (sample_from_prior) {
warmup = 0
} else {
warmup = ceiling(iter/2)
}
}
warmup = ceiling(iter / 2)
obs = stops
# check model identifiability
r <- length(unique(obs$driver_race))
d <- length(unique(obs$location_variable))
if (! (r >= 3 & d >= 5)) {
stop('Not enough departments to constrain estimates')
}
# Package data for Stan
stan_data = get_stan_data(model_file, obs)
# set up parameter initialization
init_fn = get_stan_initialization(model_file, stan_data)
# fit the model
my_model = stan_model(paste0('stan_models/', model_file))
t1 = now()
message('Starting sampling now!')
fit <- sampling(my_model, data = stan_data, iter=iter, init = init_fn, chains=chains, cores=chains, refresh = 50, warmup = warmup, control = list(adapt_delta = adapt_delta, max_treedepth = max_treedepth, adapt_engaged = !sample_from_prior), verbose = verbose, diagnostic_file = paste0(output_filename, '_diag.txt'))
seconds_required = as.numeric(now() - t1, units = "secs")
post = rstan::extract(fit)
s = summary(fit)
Rhat = max(s$summary[,'Rhat'], na.rm = TRUE)
message(sprintf("RHat is %2.3f", Rhat))
if((model_file == 'model_flat.stan') | (model_file == 'multinomial_model_flat.stan')){
obs = get_thresholds_from_post_old_model(post, obs);
}else{
obs$thresholds = get_thresholds_from_post(post, obs)
}
print(get_single_threshold_from_state(obs))
save(file=paste0(output_filename, '.RData'), obs, post, fit, seconds_required, Rhat)
}
run_threshold_test <- function(file_prefix, model_file){
#Checked. This actually runs the threshold test.
input_file = paste0(base_input_dir, file_prefix, '.RData')
stopifnot(file.exists(input_file))
stopifnot(file.exists(paste0('stan_models/', model_file)))
message(sprintf("Loading %s", input_file))
load(input_file)
model_name = gsub('.stan', '', model_file)
out_name = paste0(base_output_dir, sprintf('%s_%s', file_prefix, model_name))
message(sprintf('Running threshold test on %s: %i locations, %i races, saving results to %s',
file_prefix,
length(unique(stops$location_variable)),
length(unique(stops$driver_race)),
out_name))
output = run_mixture_mcmc(stops, out_name, iter=5000, chains=5, adapt_delta=.9, max_treedepth=12, model_file=model_file)
}