Vignette constructing model data from output

[lentinj]

The core of this is converting report array back into a data.frame. You can achieve this with canned_to_data below:

canned_to_data <- function (x, col = "number") {
  dn <- names(dim(x))
  out <- as.data.frame.table(
      g3_array_agg(x, c(dn[dn != "time"], "year", "step")),
      responseName = col )
  out$year <- as.numeric(out$year)
  out$step <- as.numeric(out$step)
  if ("stock" %in% names(out)) out$stock <- as.character(out$stock)
  if ("length" %in% names(out)) levels(out$length) <- sapply(
      strsplit(levels(out$length), ":"), function (x) paste0("[", x[[1]], ",", x[[2]], ")"))
  if ("age" %in% names(out)) out$age <- as.integer(gsub("^age", "", out$age))
  return(out)
}

ldist_f_surv <- canned_to_data(canned$r$cdist_sumofsquares_ldist_f_surv_model__num)
aldist_f_surv <- canned_to_data(canned$r$cdist_sumofsquares_aldist_f_surv_model__num)
matp_f_surv <- canned_to_data(canned$r$cdist_sumofsquares_matp_f_surv_model__num)

...but there's more impedance matching than I'd like. Get rid of as much as possible, hide the rest in a g3_array_tbl, and make a vignette demonstrating how to reverse engineer the parameters we started with.

[lentinj]

@willbutler42 this is probably very relevant to your interests in gadget-framework/gadgetutils#44 - would a g3_array_table() make your life easier? Do you have any opinions on what it would do? Is this something you've already written?

The above is something lifted out of my scrap-book, and only handles the cases I was interested in at the time. But I don't think it's too far off.

[willbutler42]

@willbutler42 this is probably very relevant to your interests in gadget-framework/gadgetutils#44 - would a g3_array_table() make your life easier? Do you have any opinions on what it would do? Is this something you've already written?

Yes, think it would make life easier. I don´t have a specific function written, but its something I was considering when doing the g3_fit re-write as it would basically be implemented throughout g3_fit (e.g. here). Regarding what it should do, at least, when processing catchdistributions, we check for columns (stock, stock_re and predator_length, so maybe an option to create and fill these columns with NAs could be useful? Though I´m not sure how useful this would be outside of g3_fit so maybe not necessary. Otherwise, it seems to cover the manipulations needed for g3_fit at least.

The above is something lifted out of my scrap-book, and only handles the cases I was interested in at the time. But I don't think it's too far off.

When I get back to g3_fit (probably next week) I can starting plugging it in to the individual functions and report back

[lentinj]

maybe an option to create and fill these columns with NAs could be useful?

So the main aim is to have a expected set of columns, rather than having to second guess all the time? It certainly could do, but I'm kinda tempted to say that it's a separate utility function and/or neat one-liner.

Somewhat related, I think stock_re is kinda useless now, as the "stock" column will match part names as well as whole. Assuming you'd have to work on models to use the new g3_fit anyway, it might be worth not supporting any more? That is, unless the models do something that the part-name syntax can't deal with.

I'm also not sure how useful this distinction is between low-level g3_array_* tools and high-level g3f_* stuff is. I'm somewhat worried we'll end up with a parallel-universe g3_fit. But the code above needs to live somewhere at least.