What's the meaning of tmax of the control parameter in optim of R?

I'm new to R and data mining/ machine learning.

I'm trying to understand the use of optim with SANN method.

I found the documentation of the parameter tmax as follows:

tmax
    is the number of function evaluations at each temperature for the "SANN" method. Defaults to 10.

What does that suppose to mean ?

In my understanding of SANN, you just need to come up with one candidate solution at each temperature. So I don't kn开发者_C百科ow what does this tmax means. Does it mean you can try up totmaxcandidates and then choose the best one to proceed ?

Try making the temperature function listed in the documentation into an R function so you can experiment:

tf <- function(t,temp,tmax) temp / log(((t-1) %/% tmax)*tmax + exp(1))
curve(tf(x,temp=10,tmax=10),from=1,to=1000)
curve(tf(x,temp=10,tmax=100),col=2,add=TRUE)
curve(tf(x,temp=10,tmax=5),col=4,add=TRUE)

(Bottom line: yes, tmax keeps the temperature fixed at a specified temperature for longer. You don't choose the best option from each temperature -- rather, at each step you pick a candidate option and compare it to the previously kept option according to the simulated annealing (Metropolis) rule ...)