report mode when simulation error occurs

[rtesra]

This PR updates the extract_indicators() and output_package() function to return mode indicator values when simulating indicators fails. Changes made:

• Added a class to the object returned by extract_indicators() to indicate if the mean or mode has been returned
• Added an additional step in output_package() that will remove sample if simulation error occurs and generate indicators with only mean values. This will also output a naomi warning that will be displayed to users on the web app to indicate that uncertainty was not generated.

@jeffeaton At the moment this does not output the original warning from extract_indicators() indicating which indicator caused the simulation fail. Options are:

• Output as a warning in naomi scripts - this is useful for debugging purpose
• Output as warning on web app - less useful as it does not give users much of an indication of the source of the problem

[jeffeaton]

Code for producing example:

devtools::load_all()

source("~/Downloads/naomi-v2.6.9/tests/testthat/setup-model-frame.R")

a_fit_bad_sample <- a_fit_sample
a_fit_bad_sample$sample$anc_alpha_t2_out[1,] <- NA_real_

out_bad <- output_package(a_fit_bad_sample, a_naomi_data) out_good <-
output_package(a_fit_sample, a_naomi_data)

# For sample with NAs, mode is returned and mean is NA
expect_true("mode" %in% class(out_bad$indicators))
expect_true(all(is.na(out$indicators$mean)))

# For sample without NAs, mode and mode returned expect_true("mean" %in%
class(out_good$indicators))
expect_true(all(!is.na(out_good$indicators$mean)))

[jeffeaton]

Thanks for this, this looks generally good to me.

A comment on the structure: I think that returning mode field if no uncertainty generated and mean field if is generated will be an undue headache for the front end. In principle, I like keeping this separate, but I suspect it will be far less hassle for @r-ash if we copy the mode field into the mean field for display purposes.

Let's discuss with @r-ash + also confirm with him that missing values for lower / upper doesn't create any issues for plotting, etc.

Note to update NEWS and bump version number.

[jeffeaton]

@rtesra -- regarding this comment:

@jeffeaton At the moment this does not output the original warning from extract_indicators() indicating which indicator caused the simulation fail. Options are:

Output as a warning in naomi scripts - this is useful for debugging purpose
Output as warning on web app - less useful as it does not give users much of an indication of the source of the problem

I not completely sure what you are referring to with naomi scripts? Can you clarify? But I agree it's not so useful to return to the user at the moment.

[jeffeaton]

Thanks, this looks good to me. Thanks for example on the warnings generation.

I've approved this; I'm holding merging it because I'm not sure if @r-ash is ready for it yet.

[rtesra]

hintr build hivtools/hintr#404

[r-ash]

Run up the app and still some issues in calibrate so this is on hold