Repository navigation
[feat]: More fims plots runs test, follows #469 - #473
ericward-noaa wants to merge 7 commits into
Conversation
convert_output() passed FIMS output through unchanged, so `estimate` and `year` were empty and no stockplotr plot could use it. The new conout_fims() reshapes `FIMS::get_estimates()` output (or a FIMSFit object, or an .rds file with either): - Each data point (index, landings, age and length compositions) is split into "*_observed" and "*_predicted" rows, with the observation error and negative log-likelihood on the observed row, as for SS - `estimated` becomes `estimate`, and the new `start_year` argument converts FIMS year indices to years; `era` is "time", or "fore" after the last year of data. - fims_var_names.csv maps current FIMS labels (e.g., expected_recruitment, numbers_at_age, log_devs) to the standard labels. Also fixes: - convert_output() ignored a user-supplied `model` when `file` is an object. - plot_index() matched every label containing "index" (e.g., index_log_q in Rceattle output or index numbers at age in FIMS output); it now plots only index_observed and index_predicted. - plot_index() failed for a single index fleet, because the fleet facet was added back after process_data() dropped the column.
these tests were failing, should pass now
Residuals of the fit to each index over time, with three-sigma control limits and a runs test for non-random residuals (as in ss3diags/JABBA). Works on convert_output() output for SS3, BAM, and FIMS.
ericward-noaa
changed the title
ericward-noaa
changed the title
|
@Schiano-NOAA @sbreitbart-NOAA ready for review |
There was a problem hiding this comment.
was devtools::document run when this was made? I don't see new documentation.
| #' in the file path provided. | ||
| #' | ||
| #' Default: NULL | ||
| #' @param start_year Number. First year of the model, for FIMS output only. |
There was a problem hiding this comment.
Is it possible so FIMS outputs years too or instead of this?
There was a problem hiding this comment.
The background here is that FIMS::get_estimates returns year_i, starting at 1. The FIMSFit object doesn't keep the data's FIMSFrame (where the start year lives -- can be extracted with FIMS::get_start_year() )
I think there's a few options,
(1) Keep start_year as is
(2) Ask FIMS to add years to output
(3) let the start_year argument also be a FIMSFrame, and then the function could call FIMS::get_start_year()
Others?
There was a problem hiding this comment.
I'm chatting with Kelli on thursday to align naming and output from stockplotr and FIMS. Ideally it should be a 1:1 so the conout_fims would be moot at some point. I will bring up the years in FIMS to her then
|
|
Schiano-NOAA
left a comment
Schiano-NOAA
left a comment
There was a problem hiding this comment.
I think this is almost good to go. It works for all of the example data and is a good way to incorporate the FIMS output into the workflow.
Could you please add the key quantities that are generated into the key_quantity_template.csv found in the inst/resources folder. In other plots that are exported to an rda, the key quantities get filled in to the sheet to provide a record of these quantities for future referece. To see what happens run the following example:
plot_biomass(stockplotr::example_data, module = "TIME_SERIES", make_rda = TRUE)
To understand how it works see here:
Line 354 in 42c6255
and
Line 401 in 42c6255
Once that is done, ping me and I will merge 😄
|
Thanks @Schiano-NOAA , I missed the key quantities stuff before. Fixed now. To clarify, I think these should be merged in first: |
|
@ericward-noaa thanks! I appreciate the update and the suggested merge order. I am working on the others now |
What is the feature?
stockplotr can show the fit to an index (plot_index()), but it has no standard diagnostic for whether the residuals of that fit are random. The residual runs test is a standard diagnostic in ss3diags (SSplotRunstest()) and JABBA, and it’s useful for FIMS models, which can include several survey and fishery indices. This PR adds plot_runs_test(), which works on any convert_output() output (SS3, BAM, FIMS) and on the package’s example data.
Follows PR #469
How have you implemented the solution?
plot_runs_test():
It plots log residuals, log(observed) − log(predicted), over time for each index.
A shaded band shows the three-sigma control limits, with sigma from the average moving range.
The band is green when a one-sided runs test passes (p ≥ 0.05) and red when it fails. Residuals outside the limits are filled red.
It has fleet, module, ncol, interactive, make_rda and figures_dir arguments, consistent with the other plot functions.
Does the PR impact any other area of the project, maybe another repo?
save_all_plots():
It runs every exported plot_* function, so it now also saves runs_test_figure.rda.
test-save_all_plots.R adds that file to the expected list.
Captions: captions_alt_text_template.csv gains one row (runs_test). Existing rows are unchanged.
PR contribution checklist
Thank you for contributing to stockplotr! Please check the boxes below to confirm this pull request (PR) is ready for review. Edit this PR description to add the release indicator and any explanations.