Simplify epred

[seabbs]

This PR is unlinked to an issue. In reviewing the current functionality needed for a family I slightly simplified the current code.

Unrelated to this PR there are lots of ways we can streamline adding families and sharing support across models and across families (for models) it looks like which is great.

[seabbs]

Hmm interesting. The with approch is directly based on the lognormal code from brms so curious why this fails tests (but not otherwise?)

[athowes]

Hmm interesting. The with approch is directly based on the lognormal code from brms so curious why this fails tests (but not otherwise?)

Yeah not sure, I guess just look at prep$dpar and see if it looks as expected. Maybe it doesn't for some reason

[athowes]

I'll go in and look at this.

[athowes]

> str(prep$dpars)
List of 2
 $ mu   :List of 10
  ..$ family:List of 16
  .. ..$ family           : chr "custom"
  .. ..$ link             : chr "identity"
  .. ..$ name             : chr "latent_lognormal"
  .. ..$ dpars            : chr [1:2] "mu" "sigma"
  .. ..$ lb               :List of 2
  .. .. ..$ mu   : chr NA
  .. .. ..$ sigma: chr "0"
  .. ..$ ub               :List of 2
  .. .. ..$ mu   : chr NA
  .. .. ..$ sigma: chr NA
  .. ..$ type             : chr "real"
  .. ..$ vars             : chr [1:3] "pwindow" "swindow" "vreal1"
  .. ..$ loop             : logi FALSE
  .. ..$ specials         : chr(0) 
  .. ..$ log_lik          : NULL
  .. ..$ posterior_predict: NULL
  .. ..$ posterior_epred  : NULL
  .. ..$ env              :<environment: 0x12f030b38> 
  .. ..$ normalized       : chr ""
  .. ..$ link_sigma       : chr "log"
  .. ..- attr(*, "class")= chr [1:3] "customfamily" "brmsfamily" "family"
  ..$ ndraws: int 4000
  ..$ nobs  : int 500
  ..$ fe    :List of 2
  .. ..$ X: num [1:500, 1] 1 1 1 1 1 1 1 1 1 1 ...
  .. .. ..- attr(*, "dimnames")=List of 2
  .. .. .. ..$ : chr [1:500] "1" "2" "3" "4" ...
  .. .. .. ..$ : chr "Intercept"
  .. .. ..- attr(*, "assign")= int 0
  .. ..$ b: num [1:4000, 1] 1.75 1.77 1.73 1.76 1.76 ...
  .. .. ..- attr(*, "dimnames")=List of 2
  .. .. .. ..$ draw    : chr [1:4000] "1" "2" "3" "4" ...
  .. .. .. ..$ variable: chr "b_Intercept"
  .. .. ..- attr(*, "nchains")= int 4
  ..$ sp    : list()
  ..$ cs    : list()
  ..$ sm    : list()
  ..$ gp    : list()
  ..$ re    : list()
  ..$ ac    : list()
  ..- attr(*, "class")= chr "bprepl"
 $ sigma: num [1:4000] 0.498 0.502 0.486 0.469 0.488 ...

[athowes]

I think because it's a custom family it's more complicated, so reverting back to the get_dpars version (and keeping the line saving in the gamma case).

[seabbs]

Oh thats very interesting. Good detective work