-
Notifications
You must be signed in to change notification settings - Fork 1
/
buildModelsForDeployment.R
166 lines (124 loc) · 6.42 KB
/
buildModelsForDeployment.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
# buildModelsForDeployment.R
# this script (and similar others?) controls standardized database queries and model training for web deployment
library(dbViewR)
library(incidenceMapR)
library(modelServR)
library(modelTestR)
library(dplyr)
library(ggplot2)
## factors and levels to be modeled ##
#####################
### REAL DATA #######
#####################
db <- selectFromDB(queryIn= list(SELECT =c("*")), source = 'production')
pathogens <- c('all', unique(db$observedData$pathogen))
factors <- c('site_type','sex','flu_shot')
geoLevels <- list( seattle_geojson = c('residence_puma','residence_neighborhood_district_name','residence_cra_name','residence_census_tract'),
wa_geojson = c('residence_puma'), # census tract impossible due to memory limits
king_county_geojson = c('residence_puma','residence_census_tract')
)
##############################
## time-independent maps #####
##############################
# catchments: number of subjects with pathogen and factor at residence location
for (SOURCE in names(geoLevels)){
for (PATHOGEN in pathogens){
for (FACTOR in factors){
for (GEO in geoLevels[[SOURCE]]){
queryIn <- list(
SELECT =list(COLUMN=c('pathogen', FACTOR, GEO)),
WHERE =list(COLUMN='pathogen', IN=PATHOGEN),
GROUP_BY =list(COLUMN=c(FACTOR,GEO)),
SUMMARIZE=list(COLUMN=FACTOR, IN= 'all')
)
shp <- masterSpatialDB(shape_level = gsub('residence_','',GEO), source = SOURCE)
db <- expandDB( selectFromDB( queryIn, source='production', na.rm=TRUE ), shp=shp )
modelDefinition <- smoothModel(db=db, shp=shp)
# training occassionaly segfaults but it does not appear to be deterministic...
tryCatch(
{
model <- modelTrainR(modelDefinition)
print(summary(model$inla))
saveModel(model)
dir.create('/home/rstudio/seattle_flu/plots/', showWarnings = FALSE)
for(k in unique(model$modeledData[[FACTOR]])){
tmp<-list(modeledData = model$modeledData[model$modeledData[[FACTOR]]==k,])
fname <- paste('/home/rstudio/seattle_flu/plots/',paste(PATHOGEN,SOURCE,GEO,FACTOR,k,sep='-'),'.png',sep='')
png(filename = fname,width = 6, height = 5, units = "in", res = 300)
print(ggplotSmoothMap(tmp,shp,title=k,shape_level = GEO))
dev.off()
}
}, error=function(e){cat("ERROR :",conditionMessage(e), "\n")}
)
}
}
}
}
# age-distributions by pathogen and factor
# eventually this should be multinomial models, but indepdendent binomial for now
### BROKEN BECAUSE NEED TO PROPOGATE AGE_RANGE from DATABASE THROUGH SYSTEM
for (PATHOGEN in pathogens){
for (FACTOR in factors){
queryIn <- list(
SELECT =list(COLUMN=c('pathogen',FACTOR,'age_range_fine_lower')),
GROUP_BY =list(COLUMN=c('pathogen',FACTOR,'age_range_fine_lower')),
SUMMARIZE=list(COLUMN='pathogen', IN= PATHOGEN)
)
# queryIn <- list(
# SELECT =list(COLUMN=c(FACTOR,'age_range_fine_lower')),
# GROUP_BY =list(COLUMN=c(FACTOR,'age_range_fine_lower')),
# SUMMARIZE=list(COLUMN=FACTOR, IN= 'all')
# )
db<- selectFromDB( queryIn, source='production', na.rm=TRUE )
# get all ages denominator. I'm not sure how to implement this as single query..
tmp<-db$observedData %>% group_by_(.dots=c(FACTOR,'age_range_fine_lower')) %>% summarize(n=sum(n))
db$observedData <- db$observedData %>% select(-n) %>% left_join(tmp)
db <- expandDB(db)
modelDefinition <- smoothModel(db=db, shp=shp)
model <- modelTrainR(modelDefinition)
print(summary(model$inla))
saveModel(model)
p1<-ggplot(model$modeledData) + geom_line(aes(x=age_range_fine_lower ,y=modeled_count_mode, group=pathogen)) +
geom_point(aes(x=age_range_fine_lower, y=positive, group=pathogen)) +
geom_ribbon(aes(x=age_range_fine_lower ,ymin=modeled_count_0_025quant, ymax=modeled_count_0_975quant , group=pathogen)) +
facet_wrap(FACTOR) +
ylim(c(0,2*max(model$modeledData$modeled_count_mode)))
fname <- paste('/home/rstudio/seattle_flu/data/plots/',paste(PATHOGEN,FACTOR,'age_range_fine',sep='-'),'.png',sep='')
png(filename = fname,width = 6, height = 5, units = "in", res = 300)
print(p1)
dev.off()
}
}
#####################################
###### timeseries models ############
#####################################
# number of subjects with pathogen and factor at residence location
for (SOURCE in names(geoLevels)){
for (PATHOGEN in pathogens){
for (GEO in geoLevels[[SOURCE]]){
queryIn <- list(
SELECT =list(COLUMN=c('pathogen', factors, GEO,'encountered_week')),
WHERE =list(COLUMN='pathogen', IN=PATHOGEN),
GROUP_BY =list(COLUMN=c('pathogen',factors,GEO,"encountered_week")),
SUMMARIZE=list(COLUMN='pathogen', IN= PATHOGEN)
)
shp <- masterSpatialDB(shape_level = gsub('residence_','',GEO), source = SOURCE)
db <- expandDB( selectFromDB( queryIn, source='production', na.rm=TRUE ), shp=shp )
# training occassionaly segfaults on but it does not appear to be deterministic...
tryCatch(
{
model <- modelTrainR(modelDefinition)
db <- appendCatchmentModel(db,shp=shp, source='production', na.rm=TRUE )
modelDefinition <- latentFieldModel(db=db, shp=shp)
model <- modelTrainR(modelDefinition)
print(summary(model$inla))
saveModel(model)
fname <- paste('/home/rstudio/seattle_flu/data/plots/',paste(PATHOGEN,SOURCE,paste(factors,collapse='-'),GEO,'encountered_week',sep='-'),'.png',sep='')
png(filename = fname,width = 6, height = 5, units = "in", res = 300)
print(ggplot(model$latentField) + geom_line(aes_string(x='encountered_week',y="modeled_intensity_mode", color=GEO,group =GEO)) )
dev.off()
}, error=function(e){cat("ERROR :",conditionMessage(e), "\n")}
)
}
}
}