draw fails for models with no smoothing

[Nelson-Gon]

Hi,

Awesome package, continuing from my discussion from Twitter. I have tried the following and found that draw fails

library(gratia)
library(mgcv)
dat <- data_sim("eg4", n = 400, seed = 42)
b <- gam(y ~ x0 + x1 + x2 , data = dat, method = "REML")
appraise(b)
draw(b, parametric = TRUE)

It fails with

Unable to draw any of the model terms.

I am using gratia ‘0.6.0’.

EDIT

I have upgraded to 0.7.0 and now have the following error

Error: Problem with mutate() column data.
i data = list(add_confint(.data$data, coverage = ci_level)).
x Column data not found in .data

A traceback

+-<error/dplyr:::mutate_error>
| Problem with `mutate()` column `data`.
| i `data = list(add_confint(.data$data, coverage = ci_level))`.
| x Column `data` not found in `.data`
\-<error/rlang_error_data_pronoun_not_found>
  Column `data` not found in `.data`
Backtrace:
  1. gratia::draw(b, parametric = TRUE)
 14. rlang:::abort_data_pronoun(x)

Thank you,

NelsonGon

[gavinsimpson]

Confirmed; thanks @Nelson-Gon. I'll see where the issue is...

[gavinsimpson]

This was caused by a couple of issues with the ordering of code in draw.gam() and a potential problem in predict.gam() where if one passes exclude = character(0) (which arises when smooths() returns such a thing for a model with no smooths...), predict.gam() doesn't return any columns in the fit or se.fit arrays.

The ordering issues are fixed and the exclude problem is now worked around...