-
Notifications
You must be signed in to change notification settings - Fork 0
/
demo1.qmd
291 lines (246 loc) · 9.07 KB
/
demo1.qmd
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
---
title: "Flu Season 2022-2023"
description: "Predicting flu season 2022-2023 using a randomwalk + trend model (non-spatial)"
format:
html:
df-print: kable
code-fold: show
code-summary: "Hide code"
code-overflow: wrap
toc-title: Page Contents
toc: true
toc-depth: 2
toc-location: right
number-sections: true
html-math-method: katex
css: styles.css
theme: flatly
smooth-scroll: true
editor_options:
chunk_output_type: console
---
```{=html}
<style type="text/css">
body, td {
font-size: 13pt;
}
code.r{
font-size: 9pt;
}
pre {
font-size: 11pt
}
</style>
```
## Overview
This is a quick demonstration of using flusion data to forecast 2022-2023 influenza hospitalizations across all U.S. States and Territories. The demo includes matching flusion to *truth* data from [FluSight](https://github.com/cdcepi/Flusight-forecast-data), constructing a non-spatial randomwalk model, and then comparing the predicted values to truth data.
## Analysis Setup
### Libraries
Loading libraries.
```{r warning=FALSE, message=FALSE}
#wrangling
library(tidyverse)
library(lubridate)
#inference
library(INLA)
#use adaptive search algorithm
inla.setOption(inla.mode= "experimental")
options(dplyr.summarise.inform = FALSE)
```
### Observation Data
**flusion**
```{r warning=FALSE, message=FALSE}
#function to downlaod file
get_data <- function(url) {
df <- read_csv(url)
return(df)
}
flusion_url <- "https://github.com/JMHumphreys/flusion/raw/main/flusion/flusion_v1.csv"
flusion <- get_data(flusion_url)
head(flusion)
```
**FluSight truth data**
```{r warning=FALSE, message=FALSE}
#FluSight: 2023-06-12
flusight_url <- "https://github.com/cdcepi/Flusight-forecast-data/raw/master/data-truth/truth-Incident%20Hospitalizations.csv"
flusight_truth <- get_data(flusight_url)
head(flusight_truth)
```
### Join Data
```{r}
range(flusion$date)
range(flusight_truth$date) #1 week added since flusion.v1
flusight_truth <- flusight_truth %>%
mutate(truth = value) %>%
select(date, location, truth)
comb_data <- left_join(flusion, flusight_truth, by = c("date", "location"))
comb_data <- comb_data %>%
mutate(ts_weeks = as.integer(as.factor(year + epiweek/52)))
head(comb_data)
tail(comb_data)
```
### Overlap Period
Quick plot to compare flusion estimates to FluSight truth.
```{r fig.width=8, fig.height=6}
overlap_data <- comb_data %>%
filter(date >= min(flusight_truth$date) &
date <= max(flusight_truth$date))
overlap_natl <- overlap_data %>%
group_by(date) %>%
summarise(flusion = sum(q_0.50),
truth = sum(truth, na.rm=T))
overlap_natl <- reshape2::melt(overlap_natl, "date")
ggplot(overlap_natl, aes(date, value)) +
geom_bar(stat="identity") +
facet_grid(rows = vars(variable)) +
scale_x_date(date_breaks = "6 month", date_labels = "%b-%Y") +
theme_classic() +
ylab("Hospitalizations") +
xlab(" ") +
theme_minimal() +
theme(plot.margin = unit(c(2,0.1,2,0.1), "cm"),
panel.grid.minor = element_line(color = "gray90", linewidth = 0.25, linetype = 1),
panel.grid.major = element_line(color = "gray60", linewidth = 0.5, linetype = 1),
panel.background = element_blank(),
plot.background = element_blank(),
strip.text = element_text(size=14, face="bold"),
strip.background = element_blank(),
legend.position="none",
legend.text = element_text(size=12, face="bold"),
legend.title = element_text(size=16, face="bold"),
axis.title.x = element_text(size=16, face="bold"),
axis.title.y = element_text(size=16, face="bold"),
axis.text.x = element_text(size=14, face="bold", angle=60, hjust=1),
axis.text.y = element_text(size=12, face="bold"),
plot.title = element_text(size=22, face="bold"))
```
## Training vs Testing
Break data into testing and training sets. Attempt to forecast the most recent flu season 2022-2033.
####Notes:
+ **Y** is the target response variable in the demo model
+ Dates between Oct 2022 through May 2023 as coded as unknown (NA)
+ The demo model will attempt to predict the NA's
```{r}
comb_data$Y <- ifelse(comb_data$year >= 2022 & comb_data$epiweek >= 40, NA, comb_data$q_0.50)
```
## Organize Data
```{r}
comb_data <- comb_data %>%
mutate(intercept = 1, #intercept
Y = round(Y, 0)) #round to integer count data
# copy location index
comb_data$Region.1 <- as.integer(comb_data$location)
#copies of weekly time index
comb_data$ts_weeks.1 <- comb_data$ts_weeks.2 <- comb_data$ts_weeks.3 <- comb_data$ts_weeks
```
## Model
```{r}
#prior
pc.prior = list(prec = list(prior="pc.prec",
param = c(1, 0.5)))
#formula
form.rw <- Y ~ -1 + intercept + #use custom intercept
f(ts_weeks.1, #random walk + noise
constr=TRUE,
model="rw2",
hyper=pc.prior) +
f(ts_weeks.2, #extra variation outside of rw time and linear trends
constr=TRUE,
model="iid",
hyper=pc.prior) +
f(Region.1, #state-level variation
constr=TRUE,
model="iid",
hyper=pc.prior) +
ts_weeks.3 # linear trend
#run model
rw.mod = inla(form.rw, #formula
data = comb_data, #data
family = c("nbinomial"), #negative binomial
verbose = FALSE,
quantiles = c(0.05, 0.25, 0.5, 0.75, 0.95),
control.fixed = list(prec = 1,
prec.intercept = 1),
control.predictor = list(
compute = TRUE,
link = 1),
control.inla = list(strategy="adaptive",
int.strategy = "eb"),
control.compute=list(dic = F, cpo = F, waic = F))
```
## National Prediction
The bar chart indicates truth, solid line is the predicted 0.5 quantile, and shaded bands provide the 95 credible interval.
```{r fig.width=8, fig.height=6}
model_out <- rw.mod$summary.fitted.values[,c(3:7)]
names(model_out) <- c("q0.05", "q0.25", "q0.5", "q0.75", "q0.95")
comb_data_pred <- cbind(comb_data, model_out)
rw_natl <- comb_data_pred %>%
filter(is.na(Y) == TRUE) %>%
group_by(date) %>%
summarise(Q0.05 = sum(q0.05),
Q0.25 = sum(q0.25),
Q0.5 = sum(q0.5),
Q0.75 = sum(q0.75),
Q0.95 = sum(q0.95),
truth = sum(truth, na.rm=T))
ggplot(rw_natl, aes(date, truth)) +
geom_bar(stat="identity", fill="tan") +
geom_ribbon(aes(ymin=Q0.05, ymax=Q0.95),fill="steelblue", alpha = 0.3) +
geom_ribbon(aes(ymin=Q0.25, ymax=Q0.75),fill="steelblue", alpha = 0.5) +
geom_line(data=rw_natl,
aes(date, Q0.5)) +
scale_x_date(date_breaks = "1 month", date_labels = "%b-%Y") +
theme_classic() +
ylab("Hospitalizations") +
xlab(" ") +
theme_minimal() +
theme(plot.margin = unit(c(2,0.1,2,0.1), "cm"),
panel.grid.minor = element_line(color = "gray90", linewidth = 0.25, linetype = 1),
panel.grid.major = element_line(color = "gray60", linewidth = 0.5, linetype = 1),
panel.background = element_blank(),
plot.background = element_blank(),
strip.text = element_text(size=14, face="bold"),
strip.background = element_blank(),
legend.position="none",
legend.text = element_text(size=12, face="bold"),
legend.title = element_text(size=16, face="bold"),
axis.title.x = element_text(size=16, face="bold"),
axis.title.y = element_text(size=16, face="bold"),
axis.text.x = element_text(size=14, face="bold", angle=60, hjust=1),
axis.text.y = element_text(size=12, face="bold"),
plot.title = element_text(size=22, face="bold"))
```
## Random States
```{r fig.width=8, fig.height=8, warning=FALSE, message=FALSE}
set.seed(34)
random_states <- sample(comb_data_pred$abbreviation, size=4)
states_plot <- comb_data_pred %>%
filter(abbreviation %in% random_states,
is.na(Y) == TRUE)
ggplot(states_plot, aes(date, truth)) +
geom_bar(stat="identity", fill="tan") +
geom_ribbon(aes(ymin=q0.05, ymax=q0.95),fill="steelblue", alpha = 0.3) +
geom_ribbon(aes(ymin=q0.25, ymax=q0.75),fill="steelblue", alpha = 0.5) +
geom_line(data=states_plot,
aes(date, q0.5)) +
scale_x_date(date_breaks = "1 month", date_labels = "%b-%Y") +
facet_grid(rows = vars(location_name), scales = "free_y") +
theme_classic() +
ylab("Hospitalizations") +
xlab(" ") +
theme_minimal() +
theme(panel.grid.minor = element_line(color = "gray90", linewidth = 0.25, linetype = 1),
panel.grid.major = element_line(color = "gray60", linewidth = 0.5, linetype = 1),
panel.background = element_blank(),
plot.background = element_blank(),
strip.text = element_text(size=14, face="bold"),
strip.background = element_blank(),
legend.position="none",
legend.text = element_text(size=12, face="bold"),
legend.title = element_text(size=16, face="bold"),
axis.title.x = element_text(size=16, face="bold"),
axis.title.y = element_text(size=16, face="bold"),
axis.text.x = element_text(size=14, face="bold", angle=60, hjust=1),
axis.text.y = element_text(size=12, face="bold"),
plot.title = element_text(size=22, face="bold"))
```