-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathREADME.Rmd
152 lines (118 loc) · 4.76 KB
/
README.Rmd
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
---
output: github_document
---
<!-- README.md is generated from README.Rmd. Please edit that file -->
```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>",
fig.path = "man/figures/README-",
out.width = "100%"
)
```
# idbrms: Infectious Disease Modelling using brms
[](https://www.tidyverse.org/lifecycle/#experimental)
[](https://codecov.io/gh/epiforecasts/brms.ide?branch=master)
Provides population-level infectious disease models as an extension of `brms`.
## Installation
You can install the unstable development version from [GitHub](https://github.com/) with:
```{r, eval = FALSE}
# install.packages("devtools")
devtools::install_github("epiforecasts/idbrms")
```
## Example
* Load `idbrms`, `brms`, `data.table` (used for manipulating data), and `ggplot2` for visualisation.
```{r packages, message = FALSE}
library(idbrms)
library(brms)
library(data.table)
library(ggplot2)
```
* Define simulated data containing cases of some infectious disease over time in a single region (with an initial growth of 20% a day followed by a decline of 10% a day). We then assume that deaths are a convolution of cases using a log normal distribution with a log mean of 1.6 (standard deviation 0.2) and a log standard deviation of 0.8 (standard deviation 0.2), and are then scaled by a fraction (here 0.4 (standard deviation 0.025)).
```{r toy-data}
# apply a convolution of a log normal to a vector of observations
weight_cmf <- function(x, ...) {
set.seed(x[1])
meanlog <- rnorm(1, 1.6, 0.1)
sdlog <- rnorm(1, 0.6, 0.025)
cmf <- (cumsum(dlnorm(1:length(x), meanlog, sdlog)) -
cumsum(dlnorm(0:(length(x) - 1), meanlog, sdlog)))
conv <- sum(x * rev(cmf), na.rm = TRUE)
conv <- rpois(1, round(conv, 0))
return(conv)
}
obs <- data.table(
region = "Glastonbury",
cases = as.integer(c(10 * exp(0.15 * 1:50),
10 * exp(0.15 * 50) * exp(-0.1 * 1:50))),
date = seq(as.Date("2020-10-01"), by = "days", length.out = 100))
# roll over observed cases to produce a convolution
obs <- obs[, deaths := frollapply(cases, 15, weight_cmf, align = "right")]
obs <- obs[!is.na(deaths)]
obs <- obs[, deaths := round(deaths * rnorm(.N, 0.25, 0.025), 0)]
obs <- obs[deaths < 0, deaths := 0]
```
* Visual simulated data (columns are cases and points are deaths).
```{r}
ggplot(obs) +
aes(x = date, y = cases) +
geom_col(fill = "lightgrey") +
geom_point(aes(y = deaths)) +
theme_minimal()
```
* Prepare the data to be fit using the convolution model.
```{r prep-data}
prep_obs <- prepare(obs, model = "convolution", location = "region",
primary = "cases", secondary = "deaths", max_convolution = 15)
head(prep_obs, 10)
```
* Fit the model assuming a Poisson observation model (*It is important to use an identity link here as `idbrm` provides its own link by default*).
```{r fit-model-eval, include = FALSE}
fit <- idbrm(data = prep_obs, family = poisson(link = "identity"))
```
```{r fit-model, eval = FALSE}
fit <- idbrm(data = prep_obs, family = poisson(link = "identity"))
```
* Summarise fit.
```{r summarise-fit}
summary(fit)
```
* Explore estimated effect sizes (we approximately should recover those used for simulation).
```{r}
exp(posterior_summary(fit, "scale_Intercept")) /
(1 + exp(posterior_summary(fit, "scale_Intercept")))
posterior_summary(fit, "cmean_Intercept")
exp(posterior_summary(fit, "lcsd_Intercept"))
```
* Expose model stan functions.
```{r expose-fns-eval, eval = FALSE}
expose_functions(fit)
```
```{r expose-fns, include = FALSE}
expose_functions(fit)
```
* Test central estimates returns something sensible compared to observed data.
```{r}
n_obs <- length(prep_obs$primary)
fixed <- summary(fit)$fixed
pt_ests <- fixed[, 1]
names(pt_ests) <- rownames(fixed)
p_primary <- with(prep_obs, idbrms_convolve(primary, rep(pt_ests["scale_Intercept"], n_obs),
rep(pt_ests["cmean_Intercept"], n_obs),
rep(pt_ests["lcsd_Intercept"], n_obs),
cmax, index, cstart, init_obs))
ggplot(prep_obs) +
aes(x = date, y = secondary) +
geom_col(fill = "lightgrey") +
geom_point(aes(y = p_primary)) +
theme_minimal()
```
* Posterior predictive check.
```{r pred-check}
pp_check(fit)
```
* Plot conditional effects (fails with due to mismatched vector sizes in the stan dot product + there are no variables in this model so should always fail).
```{r effect-plot, eval = FALSE}
plot(conditional_effects(fit), ask = FALSE)
```