Supporting a grid of models

[cimentadaj]

Is there room for creating a new type of object that allows you to make a grid of model objects? This grid can be added to the workflow and it will only add it to the model slot, while fit will have a method for running all three models

• Check if they've done this in parsnip or tune

• Think about how to extract the results of some many models

• How would you select the best model? This is a big chunk of stuff as it will return all models for all tuning params for all metrics. How can users explore this? All of this is important since in the end we want a single final model.

An idea:

       model_grid(
         linear_reg() %>% set_engine("glmnet") %>% grid_regular(levels = 10),
         rand_forest(mtry = tune()) %>% set_engine("ranger") %>% grid_regular(levels = 100),
         svm_rbf(cost = tune()) %>% set_engine("kernlab") %>% grid_random(size = 10)
       )

[gutama]

Hi Jorge,

In tidymodels, they have workflowsets

I still wonder as I mentioned in issue #31 while working with DALEXtra
the relationship of tidyflow and workflow, why It can't be like tibble and data.frame relationship,

tibble is also a dataframe but It can convert back and forth using convenience methods

It will be easier if tidyflow has method
as_tidyflow(aWorkflow)
as.workflow(aTidtflow

Thank you