build_site() fails to render some examples
spsanderson opened this issue · comments
I am not entirely certain as to why this is occurring but when I run the build_site() function on my package some examples no longer render. For example:
library(recipes)
library(dplyr)
library(tidyr)
df <- Titanic |>
as_tibble() |>
uncount(n) |>
mutate(across(everything(), as.factor))
rec_obj <- recipe(Survived ~ ., data = df)
fct_tbl <- fast_classification(
.data = df,
.rec_obj = rec_obj,
.parsnip_eng = c("glm"))
fct_tblGives the following error:
#> Error in dplyr::mutate(mod_fitted_tbl, pred_wflw = internal_make_wflw_predictions(mod_fitted_tbl, splits_obj)): ℹ In argument: `pred_wflw =
#> internal_make_wflw_predictions(mod_fitted_tbl, splits_obj)`.
#> Caused by error in `map2()`:
#> ℹ In index: 1.
#> Caused by error in `dplyr::filter()`:
#> ! object 'rec_obj' not found
fct_tbl
#> Error in eval(expr, envir, enclos): object 'fct_tbl' not foundBut I can run it fine in the terminal like so:
> library(recipes)
> library(dplyr)
> library(tidyr)
>
> df <- Titanic |>
+ as_tibble() |>
+ uncount(n) |>
+ mutate(across(everything(), as.factor))
>
> rec_obj <- recipe(Survived ~ ., data = df)
>
> fct_tbl <- fast_classification(
+ .data = df,
+ .rec_obj = rec_obj,
+ .parsnip_eng = c("glm"))
>
> fct_tbl
# A tibble: 1 × 8
.model_id .parsnip_engine .parsnip_mode .parsnip_fns model_spec wflw fitted_wflw pred_wflw
<int> <chr> <chr> <chr> <list> <list> <list> <list>
1 1 glm classification logistic_reg <spec[+]> <workflow> <workflow> <tibble> 
Your reprex doesn't run as-is. Where does fast_classification() come from?
library(recipes)
#> Loading required package: dplyr
#>
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#>
#> filter, lag
#> The following objects are masked from 'package:base':
#>
#> intersect, setdiff, setequal, union
#>
#> Attaching package: 'recipes'
#> The following object is masked from 'package:stats':
#>
#> step
library(dplyr)
library(tidyr)
df <- Titanic |>
as_tibble() |>
uncount(n) |>
mutate(across(everything(), as.factor))
rec_obj <- recipe(Survived ~ ., data = df)
fct_tbl <- fast_classification(
.data = df,
.rec_obj = rec_obj,
.parsnip_eng = c("glm"))
#> Error in fast_classification(.data = df, .rec_obj = rec_obj, .parsnip_eng = c("glm")): could not find function "fast_classification"
fct_tbl
#> Error in eval(expr, envir, enclos): object 'fct_tbl' not foundCreated on 2023-12-20 with reprex v2.0.2
This question was also posted here: https://stackoverflow.com/q/77689110.
You need to tell R that, by using library(tidyAML) before using it, or tidyAML::fast_classification when you use it. That's why your example doesn't run.
No, you shouldn't need to specify the package if you have it installed, and I think build_site() does an install in any case. I just cloned the tidyAML master branch and didn't get any error from build_site(). Have you made changes relative to the master branch? If so, I'd try installing first to make sure there are no errors.
Whoops, now I look at the output, I do see the error you reported.
I tried to take a closer look, but I couldn't follow the code you are using. You're doing a lot of tricky things -- suppressing errors, etc. I'd suggest adding a lot of message statements to the problematic function, which appears to be internal_make_wflw_predictions. Get them to print information about the intermediate values, or check that the assumptions you're making are true and print results from them.
I tried a bit of this, but as I said, I couldn't follow your code.
You can save some time by only running
pkgdown::build_reference(topics="fast_classification")
so only that page gets built.
I added this to the function and now it is working:
rec_obj <- workflows::extract_preprocessor(fitted_wflw)