How do i add blueprint into workflow_set in tidymodels - sparse-matrix

Question

I tried to follow the examples in the
Link 1 - Sparse Matrix
https://www.tidyverse.org/blog/2020/11/tidymodels-sparse-support/
Link 2 - Workflow_sets
https://www.tmwr.org/workflow-sets.html
I had trouble including the blue print into the workflow sets.
In the examples where workflow_set is defined in link 2
no_pre_proc <-
workflow_set(
preproc = list(simple = model_vars),
models = list(MARS = mars_spec, CART = cart_spec, CART_bagged = bag_cart_spec,
RF = rf_spec, boosting = xgb_spec, Cubist = cubist_spec)
)
and the way we add blue print into the workflow in link 1
wf_sparse <-
workflow() %>%
add_recipe(text_rec, blueprint = sparse_bp) %>%
add_model(lasso_spec)
wf_default <-
workflow() %>%
add_recipe(text_rec) %>%
add_model(lasso_spec)
Where and how do I add the "blueprint = sparse_bp" option in the workflow_set above?
My attempts were
no_pre_proc <-
workflow_set(
preproc = list(simple = model_vars),
models = list(MARS = mars_spec, CART = cart_spec, CART_bagged = bag_cart_spec,
RF = rf_spec, boosting = xgb_spec, Cubist = cubist_spec)) %>%
option_add(update_blueprint(blueprint = sparse_bp))
Running the racing tune gave me this error
Error: Problem with `mutate()` column `option`.
i `option = purrr::map(option, append_options, dots)`.
x All options should be named.
Run `rlang::last_error()` to see where the error occurred
<error/rlang_error>
There were 9 workflows that had no results.
Backtrace:
1. ggplot2::autoplot(...)
2. workflowsets:::autoplot.workflow_set(...)
3. workflowsets:::rank_plot(...)
4. workflowsets:::pick_metric(object, rank_metric, metric)
6. workflowsets:::collect_metrics.workflow_set(x)
7. workflowsets:::check_incompete(x, fail = TRUE)
8. workflowsets:::halt(msg)
Run `rlang::last_trace()` to see the full context.
> rlang::last_trace()
<error/rlang_error>
There were 9 workflows that had no results.
Backtrace:
x
1. +-ggplot2::autoplot(...)
2. \-workflowsets:::autoplot.workflow_set(...)
3. \-workflowsets:::rank_plot(...)
4. \-workflowsets:::pick_metric(object, rank_metric, metric)
5. +-tune::collect_metrics(x)
6. \-workflowsets:::collect_metrics.workflow_set(x)
7. \-workflowsets:::check_incompete(x, fail = TRUE)
8. \-workflowsets:::halt(msg)
>
thanks,

Answer 1

Thank you for asking this question; we definitely are not supporting this use case (passing non-default arguments to the recipe or model) very well right now. We've opened an issue here where you can track our work on this.
In the meantime, you could try a bit of a hacky workaround by manually using update_recipe() on the workflow you are interested in:
library(tidymodels)
#> Registered S3 method overwritten by 'tune':
#> method from
#> required_pkgs.model_spec parsnip
data(parabolic)
set.seed(1)
split <- initial_split(parabolic)
train_set <- training(split)
test_set <- testing(split)
glmnet_spec <-
logistic_reg(penalty = 0.1, mixture = 0) %>%
set_engine("glmnet")
rec <-
recipe(class ~ ., data = train_set) %>%
step_YeoJohnson(all_numeric_predictors())
sparse_bp <- hardhat::default_recipe_blueprint(composition = "dgCMatrix")
wfs_orig <-
workflow_set(
preproc = list(yj = rec,
norm = rec %>% step_normalize(all_numeric_predictors())),
models = list(regularized = glmnet_spec)
)
new_wf <-
wfs_orig %>%
extract_workflow("yj_regularized") %>%
update_recipe(rec, blueprint = sparse_bp)
Created on 2021-12-09 by the reprex package (v2.0.1)
Then (I know this feels hacky for now) manually take this new_wf and stick it in to the wfs_orig$info[[1]]$workflow slot to replace what is there.