[R] Save a list of list and search for values

[Duncan Murdoch]

On 26/02/2015 9:27 AM, Alaios via R-help wrote:
> Dear all,in my code I am using the mix() function that returns results in a list. The result looks like
> List of 10
>   $ parameters  :'data.frame':   2 obs. of  3 variables:
>    ..$ pi   : num [1:2] 0.77 0.23
>    ..$ mu   : num [1:2] -7034 162783
>    ..$ sigma: num [1:2] 20235 95261
>   $ se          :'data.frame':   2 obs. of  3 variables:
>    ..$ pi.se   : num [1:2] 0.0423 0.0423
>    ..$ mu.se   : num [1:2] 177 12422
>    ..$ sigma.se: num [1:2] 1067 65551
>   $ distribution: chr "norm"
>   $ constraint  :List of 8
>    ..$ conpi   : chr "NONE"
>    ..$ conmu   : chr "NONE"
>    ..$ consigma: chr "NONE"
>    ..$ fixpi   : NULL
>    ..$ fixmu   : NULL
>    ..$ fixsigma: NULL
>    ..$ cov     : NULL
>    ..$ size    : NULL
>   $ chisq       : num 28
>   $ df          : num 5
>   $ P           : num 3.67e-05
>   $ vmat        : num [1:5, 1:5] 1.79e-03 -3.69e-01 -1.17e+02 2.95e+01 -2.63e+03 ...
>   $ mixdata     :Classes ‘mixdata’ and 'data.frame':     11 obs. of  2 variables:
>    ..$ X    : num [1:11] 1e+04 2e+04 3e+04 4e+04 5e+04 6e+04 7e+04 8e+04 9e+04 1e+05 ...
>    ..$ count: int [1:11] 993 137 82 30 21 5 7 14 21 2 ...
>   $ usecondit   : logi FALSE
>   - attr(*, "class")= chr "mix"
>
> In my code I am trying around 10.000 fit (and each of these fits returns the list above) and I want to keep those in a way that later on I would be able to search inside all the lists.For example I would like to find inside those 10.000 lists which one has the smallest $chisq value. What would be a suitable way to implement that in R? Luckily I am working in a computer with a lot of ram so storing 10.000 lists temporary in memory before saving to disk would not be a problem.
> What would you suggest me?

If all of the lists have the same components, then it would be 
convenient to convert them into a big matrix or dataframe, with one row 
per fit.  It would need to be a dataframe if you include character data 
along with the numbers, but a matrix would be faster, if it's only 
numbers that you need.  You'd use code like this to produce the matrix:

results <- matrix(NA_real_, 10000, ncols = .... however many you keep ....)
for (i in 1:10000) {
   fit <- .... code to get the fit object ....
   results[i,] <- with(fit, c(parameters$pi, parameters$mu, 
parameters$sigma,   ......  fill in the rest ......)
}

Duncan Murdoch