predict, simulate and genFLQuant

[colinpmillar]

There are some nice things in the package that are not exported, also with some overlap. Some effort could be put into reducing the code replication and simplifying the use interface.

example of hidden functionality, the function FLa4a:::predict.stkpars could be exported as a method on a4aStkParams class.

fit0 <- sca(ple4, ple4.index)
stk_submodel <- stkmodel(fit)

FLa4a:::predict.stkpars(stk_submodel)$ny1

An object of class "FLQuant"
, , unit = unique, season = all, area = unique

    year
age  1957
  1      NA
  2  277659
  3  510240
  4  807496
  5  946944
  6  738763
  7  369066
  8  122659
  9   30208
  10   6407

stk_submodel_sim <- simulate(stk_submodel, nsim = 100)
FLa4a:::predict.stkpars(stk_submodel_sim)$ny1

An object of class "FLQuant"
iters:  100

, , unit = unique, season = all, area = unique

    year
age  1957
  1      NA(    NA)
  2  277463(  8331)
  3  510797( 44575)
  4  811887(110382)
  5  964181(160899)
  6  742286(124788)
  7  360544( 74975)
  8  115474( 27940)
  9   27037(  9828)
  10   5529(  2749)

units:  1000

[colinpmillar]

Continuing this, the function genFLQuant has some similarities to predict.stkpars, and doing some code cleaning would make things more streamlined / cleaner

[colinpmillar]

Further notes: genFLQuant, combines simulate and predict.

These three things provide the same result

res <- microbenchmark::microbenchmark(
  FLa4a:::predict.stkpars(simulate(stkmodel(fit0), nsim = 100, seed = 12342))$harvest[, "2017"],
  harvest(simulate(fit0, nsim = 100, seed = 12342))[, "2017"],
  genFLQuant(stkmodel(fit0), nsim = 100, seed = 12342, type = "response")$harvest[, "2017"], times <- 1000L
)

ggplot2::autoplot(res)

But, i think genFLQuant returns also stock.n, and catch.n, as does simulate(fit). So perhaps makes sense. Anyway, would be good to clean this up a bit.