Make my model smaller

My Ranger model trained with Tidymodels is like 6.73GB. Butcher only reduced like 300MB for me. Is there still room to make it smaller? I only needs it to predict things. Is the data of case weight required?

> butcher::weigh(short_10$model)
# A tibble: 50,954,937 × 2
   object                                  size
   <chr>                                  <dbl>
 1 pre.actions.case_weights.col           1577.
 2 pre.actions.recipe.recipe.steps.terms   788.
 3 pre.mold.blueprint.recipe.steps.terms   788.
 4 pre.actions.recipe.recipe.steps.terms   631.
 5 pre.mold.blueprint.recipe.steps.terms   631.
 6 pre.actions.recipe.recipe.steps.terms   473.
 7 pre.mold.blueprint.recipe.steps.terms   473.
 8 pre.actions.recipe.recipe.steps.terms1  315.
 9 pre.actions.recipe.recipe.steps.terms2  315.
10 pre.mold.blueprint.recipe.steps.terms1  315.
# ℹ 50,954,927 more rows
# ℹ Use `print(n = ...)` to see more rows
> model <- butcher::butcher(short_10$model)
> lobstr::obj_size(short_10$model)
6.73 GB
> lobstr::obj_size(model)
6.47 GB
> butcher::weigh(model)
# A tibble: 50,954,875 × 2
   object                                            size
   <chr>                                            <dbl>
 1 pre.actions.case_weights.col                  1577.   
 2 pre.actions.recipe.recipe.steps.terms1         315.   
 3 pre.actions.recipe.recipe.steps.terms2         315.   
 4 pre.mold.blueprint.recipe.steps.terms1         315.   
 5 pre.mold.blueprint.recipe.steps.terms2         315.   
 6 pre.mold.extras.roles.case_weights.importance    1.53 
 7 pre.case_weights                                 1.53 
 8 fit.fit.fit.forest.child.nodeIDs931              0.245
 9 fit.fit.fit.forest.child.nodeIDs932              0.245
10 fit.fit.fit.forest.split.varIDs466               0.245
# ℹ 50,954,865 more rows
# ℹ Use `print(n = ...)` to see more rows

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.