mamut86 / diffusion Goto Github PK

is.diffusion <- function(x){
return(inherits(x,"diffusion"))
}

Include an option for the user to select manually an optimisation algorithm. This goes together with Issue #4

Various time series where NM ends up in a singular matrix

Add package documentation

Maybe we should have a look at: Venkatesan, R., T. V. Krishnan, V. Kumar. 2004. Evolutionary estimation of macro-level diffusion models using genetic algorithms: An alternative to nonlinear least squares.

There is a neat plot in case of type="bass" with innovators and imitators, but it seems that this can also be done for other curves. Here's an example with Shifted Gompertz (not compound with Gamma):

a <- 1000
b <- 0.4
c <- 1
p <- exp(-c)
q <- (b-p)/b
t <- c(0:100)/10

plot(t,a*b*exp(-b*t-c*exp(-b*t))*(1+c*(1-exp(-b*t))),type="l",ylab="y(t)",lwd=2)
# Innovators
lines(t,p*a*b*exp(-b*t-c*exp(-b*t))*(1+c*(1-exp(-b*t))),col="darkgreen",lwd=2)
# Imitators
lines(t,q*a*b*exp(-b*t-c*exp(-b*t))*(1+c*(1-exp(-b*t))),col="darkblue",lwd=2)

Is it possible to do similar plots for other functions?

If forecasts are produced these are unequal to the length requested in the predict.diffusion() function.

Update the (w) parameter to allow for fixed parameters and others to be estimated.

Develop a function (or wrapper) that allows to draw diffusion curve for any manual given paramters

A potential list:

• NIU - Easingwood et al. 1983. A Nonuniform Influence Innovation Diffusion Model.
• Logistic model
• Weibull model - done
• Gamma model
• GMB - Bass, F.M et al. 1994. Why the Bass Model Fits without Decision Variables.
• Trapezoids and Triangles - Hu et al. 2018. Forecasting New Product Life Cycle Curves: Practical Approach and Empirical Analysis

compute standard errors of the parameter estimates

incorporate the ability to use named vector w rather than fixed order and include error handling

add a wrapper, use all NM, BFGS and BOBYQA (to name a few) and pick the one with the lowest MSE

change variables names which some of them are violating good R coding practice, i.e. t and c used

TS object compatibility would ensure that the plots and outputs also handle date information. There is also a bug that actuals are not shown in the plot when entered as TS object

dataSafari <- ts(tsSafari[,-1],frequency=12,start=c(2009,1));
safari6 <- dataSafari[,8];
safari6 <- window(safari6,start=time(safari6)[which(safari6!=0)[1]]);
plot(diffusion(safari6))
plot(diffusion(dataSafari[,8]))

Should go into seqdiffusion() using all parameters at once

Replace sequential if statements with switch()

add documentation for the elements used in diffusionPlot.