4. How It Works

How It Works

The regression and classification functions work in identical fashions so we will only pick on the regression function, that is to say, how does the fast_regression() function work. While, it does work for many, it does fail for some, which means the design is flawed and it is possible at this point that it is a fundamental flaw (which I think it is). First I will post the code, and then we will see what the output of each step looks like.

The Function

Here is the full function:

fast_regression <- function(.data, .rec_obj, .parsnip_fns = "all",
                            .parsnip_eng = "all", .split_type = "initial_split",
                            .split_args = NULL){

  # Tidy Eval ----
  call <- list(.parsnip_fns) %>%
    purrr::flatten_chr()
  engine <- list(.parsnip_eng) %>%
    purrr::flatten_chr()

  rec_obj <- .rec_obj
  split_type <- .split_type
  split_args <- .split_args

  # Checks ----

  # Get data splits
  df <- dplyr::as_tibble(.data)
  splits_obj <- create_splits(
    .data = df,
    .split_type = split_type,
    .split_args = split_args
  )

  # Generate Model Spec Tbl
  mod_spec_tbl <- fast_regression_parsnip_spec_tbl(
    .parsnip_fns = call,
    .parsnip_eng = engine
  )

  # Generate Workflow object
  mod_tbl <- mod_spec_tbl %>%
    dplyr::mutate(
      wflw = internal_make_wflw(mod_spec_tbl, .rec_obj = rec_obj)
    )

  mod_fitted_tbl <- mod_tbl %>%
    dplyr::mutate(
      fitted_wflw = internal_make_fitted_wflw(mod_tbl, splits_obj)
    )

  mod_pred_tbl <- mod_fitted_tbl %>%
    dplyr::mutate(
      pred_wflw = internal_make_wflw_predictions(mod_fitted_tbl, splits_obj)
    )


  # Return ----
  class(mod_tbl) <- c("fst_reg_tbl", class(mod_tbl))
  attr(mod_tbl, ".parsnip_engines") <- .parsnip_eng
  attr(mod_tbl, ".parsnip_functions") <- .parsnip_fns
  attr(mod_tbl, ".split_type") <- .split_type
  attr(mod_tbl, ".split_args") <- .split_args

  return(mod_pred_tbl)
}

We see that the function indeed is broken up into steps basically by building off of the previous output. So what is happening?

What's going on?

This function, named fast_regression, is designed to generate model specifications for regression using the parsnip package in R. Let's break down the key components of the code:

1. Function Signature:

   fast_regression <- function(.data, .rec_obj, .parsnip_fns = "all",
                               .parsnip_eng = "all", .split_type = "initial_split",
                               .split_args = NULL)

   This function takes several parameters:

   • .data: The data for the regression problem.
   • .rec_obj: A recipe object.
   • .parsnip_fns: The parsnip model functions to use. Default is "all."
   • .parsnip_eng: The parsnip model engines to use. Default is "all."
   • .split_type: The type of data split to use (e.g., initial split). Default is "initial_split."
   • .split_args: Additional arguments for data splitting. Default is NULL.

2. Tidy Eval and Parameter Handling:

   call <- list(.parsnip_fns) %>%
     purrr::flatten_chr()
   engine <- list(.parsnip_eng) %>%
     purrr::flatten_chr()
   
   rec_obj <- .rec_obj
   split_type <- .split_type
   split_args <- .split_args

   Here, the function uses tidy evaluation to handle the parameters related to parsnip functions and engines.

3. Data Splitting:

   df <- dplyr::as_tibble(.data)
   splits_obj <- create_splits(
     .data = df,
     .split_type = split_type,
     .split_args = split_args
   )

   The function converts the input data to a tibble and then uses the create_splits function to generate data splits based on the specified split type and arguments.

4. Model Specification Table Generation:

   mod_spec_tbl <- fast_regression_parsnip_spec_tbl(
     .parsnip_fns = call,
     .parsnip_eng = engine
   )

   The fast_regression_parsnip_spec_tbl function is called to generate a table of parsnip model specifications based on the specified parsnip functions and engines.

5. Workflow and Fitted Model Generation:

   mod_tbl <- mod_spec_tbl %>%
     dplyr::mutate(
       wflw = internal_make_wflw(mod_spec_tbl, .rec_obj = rec_obj)
     )
   
   mod_fitted_tbl <- mod_tbl %>%
     dplyr::mutate(
       fitted_wflw = internal_make_fitted_wflw(mod_tbl, splits_obj)
     )

   The function creates a workflow (wflw) and a fitted workflow (fitted_wflw) using internal functions (internal_make_wflw and internal_make_fitted_wflw) based on the generated model specifications and data splits.

6. Prediction Generation:

   mod_pred_tbl <- mod_fitted_tbl %>%
     dplyr::mutate(
       pred_wflw = internal_make_wflw_predictions(mod_fitted_tbl, splits_obj)
     )

   Finally, the function generates predictions using the fitted workflow and the data splits.

7. Return:

   return(mod_pred_tbl)

   The function returns a table (mod_pred_tbl) containing information about the generated model specifications, workflows, fitted models, and predictions.