Using deSolve
, we can replicate Erlang
distribution and exponential distribution for
testing
library(denim)
library(deSolve)
## Warning: package 'deSolve' was built under R version 4.3.1
# --- Transition def for denim
<- list(
transitions "S -> I" = d_exponential(0.2),
"I -> R" = d_gamma(3, 2)
)<- c(rate = 0.2, scale = 3, shape=2)
parameters <- c(S = 999, I = 1, I1 = 1, I2=0, R=0)
initialValues
# --- Transition def for deSolve
<- function(t, state, param){
transition_func with(as.list( c(state, param) ), {
= 1/scale
gamma_rate = -rate*S
dS # apply linear chain trick
= rate*S - gamma_rate*I1
dI1 = gamma_rate*I1 - gamma_rate*I2
dI2 = dI1 + dI2
dI = gamma_rate*I2
dR list(c(dS, dI, dI1, dI2, dR))
})
}
# --- Timestep definition
<- 20
simulationDuration <- 0.001 # small timestep required for comparison timestep
denim_vs_deSolve.R
<- Sys.time()
denim_start <- sim(transitions = transitions, initialValues = initialValues, parameters, simulationDuration = simulationDuration, timeStep = timestep)
mod <- Sys.time()
denim_end
# --- show output
head(mod[mod$Time %in% 1:simulationDuration,])
## Time S I R
## 1001 1 817.9120 179.0627 3.025308
## 2001 2 669.6497 310.8811 19.469173
## 3001 3 548.2628 398.4336 53.303539
## 4001 4 448.8796 448.0932 103.027204
## 5001 5 367.5116 467.6504 164.838060
## 6001 6 300.8930 464.7307 234.376244
denim_vs_deSolve.R
<- seq(0, simulationDuration, timestep)
times
<- Sys.time()
desolve_start <- ode(y = initialValues, times = times, parms = parameters, func = transition_func)
ode_mod <- Sys.time()
desolve_end
# --- show output
<- as.data.frame(ode_mod)
ode_mod head(ode_mod[ode_mod$time %in% 1:simulationDuration, c("time", "S", "I", "R")])
## time S I R
## 1001 1 817.9120 179.0585 3.029466
## 2001 2 669.6497 310.8686 19.481654
## 3001 3 548.2628 398.4125 53.324630
## 4001 4 448.8796 448.0650 103.055392
## 5001 5 367.5116 467.6172 164.871207
## 6001 6 300.8930 464.6948 234.412204
denim_vs_deSolve.R
denim
takes approximately 16.38 times as long as
deSolve
to compute the result with the given specifications
.
This significant difference can be attributed to the difference in
approaches: deSolve
solves a system of ODEs while
denim
iterates through each timestep and updates the
population in each compartment
While the approach in denim
allow more flexibility in
types of dwell time distributions, the computation time scales up as
timestep grows smaller (in O(n)
time complexity).
# increase timestep before plotting
<- mod[mod$Time %in% seq(0, simulationDuration, 0.2),]
mod <- ode_mod[ode_mod$time %in% seq(0, simulationDuration, 0.2),] ode_mod
denim_vs_deSolve.R
# ---- Plot S compartment
plot(x = mod$Time, y = mod$S,xlab = "Time", ylab = "Count", main="S compartment",
col = "#4876ff", type="l", lwd=3)
lines(ode_mod$time, ode_mod$S, lwd=3, lty=3)
legend(x = 15, y = 900,legend=c("denim", "deSolve"), col = c("#4876ff", "black"), lty=c(1,3))
# ---- Plot I compartment
plot(x = mod$Time, y = mod$I, xlab = "Time", ylab = "Count", main="I compartment",
col = "#4876ff", type="l", lwd=2)
lines(ode_mod$time, ode_mod$I, lwd=3, lty=3)
legend(x = 15, y = 350,legend=c("denim", "deSolve"), col = c("#4876ff", "black"), lty=c(1,3))
# ---- Plot R compartment
plot(x = mod$Time, y = mod$R, xlab = "Time", ylab = "Count", main="R compartment",
col = "#4876ff", type="l", lwd=2)
lines(ode_mod$time, ode_mod$R, lwd=3, lty=3)
legend(x = 15, y = 300,legend=c("denim", "deSolve"), col = c("#4876ff", "black"), lty=c(1,3))
denim_vs_deSolve.R