Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

cache duplicated preprocessor fits #958

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

cache duplicated preprocessor fits #958

wants to merge 1 commit into from

Conversation

simonpcouch
Copy link
Contributor

Closes #955. A proof of concept for cutting out duplicated preprocessor fits for iterative tuning approaches. A bit of a contrived example, but on main:

library(tidymodels)
library(embed)

hotel_rates$arrival_date <- NULL
  
bench::mark(
  tune_bayes = tune_bayes(
    workflow(
      recipe(avg_price_per_room ~ ., hotel_rates) %>% 
        step_lencode_glm(all_nominal_predictors(), outcome = vars(avg_price_per_room)),
      linear_reg(engine = "glmnet", penalty = tune())
    ),
    vfold_cv(hotel_rates)
  )
)
#> ! No improvement for 10 iterations; returning current results.
#> ! No improvement for 10 iterations; returning current results.
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 1 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tune_bayes    2.16m    2.16m   0.00773      38GB    0.897

Created on 2024-11-01 with reprex v2.1.1

After this PR:

library(tidymodels)
library(embed)

hotel_rates$arrival_date <- NULL
  
bench::mark(
  tune_bayes = tune_bayes(
    workflow(
      recipe(avg_price_per_room ~ ., hotel_rates) %>% 
        step_lencode_glm(all_nominal_predictors(), outcome = vars(avg_price_per_room)),
      linear_reg(engine = "glmnet", penalty = tune())
    ),
    vfold_cv(hotel_rates)
  )
)
#> ! No improvement for 10 iterations; returning current results.
#> ! No improvement for 10 iterations; returning current results.
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 1 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tune_bayes      20s      20s    0.0500    5.16GB     2.05

Created on 2024-11-01 with reprex v2.1.1

  • I'm not fully convinced it's actually this simple. 1) Is the split_id and iter_msg_preprocessor enough to uniquely identify a unique preprocessor fit when tuning iteratively? 2) What are the implications with different types of parallelism? We haven't used the progress_env so far for parallelized computations, yet, but also the fact that has_cached_result() would only return TRUE for iterative searches might mean that the needed information is always available in the parent process.
  • This same machinery could be used to e.g. cache model fits if only tuning the postprocessor.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Fit preprocessor just once with tune_bayes
1 participant
@simonpcouch