I try to estimate Bass Curves to analyse diffusion of innovation for different groups. Until now I use nlsLM()
of the minpack.lm
package to estimate the parameter of the curve/to fit the curve. I loop through different starting values to estimate the best fit using this command for the different starting values:
Bass.nls <- nlsLM(cumulative_y~ M * (((P + Q)^2/P) * exp(-(P + Q) * time))/(1 + (Q/P) * exp(-(P + Q) * time))^2
, start = list(M=m_start, P= p_start, Q=q_start)
, trace = F
, control = list(maxiter = 100, warnOnly = T) )
Since some groups have little data points many do not converge.
Venkatesan and Kumar (2002) suggest to use a Genetic Algorithm approach for bass model estimations when data is scarce (see also Venkatesan et al 2004). I have found some packages that implement GA in R (like GA
, genalg
, gafit
). However, since I am new to the field, I don't know which package to use and how to use the bass formula in the packages.
I hope below code can help you. I used "GA" package to utilize genetic algorithm.
x <- c(840,1470,2110,4000,7590,10950,10530,9470,
7790,5890)
t<- 1:length(x)
Horiz <- length(x)
fit <- function(p,q,m) {
res = x - (m*((exp((p+q)*t)*p*(p+q)^2) / (p*exp((p+q)*t)+q)^2))
-(sum(res**2)/Horiz)
}
GA <- ga(type = "real-valued",
fitness = function(x) fit(x[1],x[2],x[3]),
lower = c(0,0,0), upper = c(1,1,sum(x)*2),
popSize = 1000, maxiter = 1000 ,run = 500)
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With