Turn regression rule models into tidy tibbles

# S3 method for cubist
tidy(x, ...)

# S3 method for xrf
tidy(x, penalty = NULL, unit = c("rules", "columns"), ...)

Arguments

x	A `Cubist` or `xrf` object.
...	Not currently used.
penalty	A single numeric value for the `lambda` penalty value.
unit	What data should be returned? For `unit = 'rules'`, each row corresponds to a rule. For `unit = 'columns'`, each row is a predictor column. The latter can be helpful when determining variable importance.

Value

The Cubist method has columns committee, rule_num, rule, estimate, and statistics. The latter two are nested tibbles. estimate contains the parameter estimates for each term in the regression model and statistics has statistics about the data selected by the rules and the model fit.

The xrf results has columns rule_id, rule, and estimate. The rule_id column has the rule identifier (e.g., "r0_21") or the feature column name when the column is added directly into the model. For multiclass models, a class column is included.

In each case, the rule column has a character string with the rule conditions. These can be converted to an R expression using rlang::parse_expr().

Examples

library(dplyr)
#> 
#> Attaching package: ‘dplyr’
#> The following objects are masked from ‘package:stats’:
#> 
#>     filter, lag
#> The following objects are masked from ‘package:base’:
#> 
#>     intersect, setdiff, setequal, union

data(ames, package = "modeldata")

ames <-
  ames %>%
  mutate(Sale_Price = log10(ames$Sale_Price),
         Gr_Liv_Area = log10(ames$Gr_Liv_Area))

# ------------------------------------------------------------------------------

# \donttest{
cb_fit <-
  cubist_rules(committees = 10) %>%
  set_engine("Cubist") %>%
  fit(Sale_Price ~ Neighborhood + Longitude + Latitude + Gr_Liv_Area + Central_Air,
      data = ames)

cb_res <- tidy(cb_fit)
cb_res
#> # A tibble: 157 x 5
#>    committee rule_num rule                               estimate    statistic  
#>        <int>    <int> <chr>                              <list>      <list>     
#>  1         1        1 ( Central_Air == 'N' ) & ( Gr_Liv… <tibble [4… <tibble [1…
#>  2         1        2 ( Gr_Liv_Area <= 3.0326188 ) & ( … <tibble [4… <tibble [1…
#>  3         1        3 ( Neighborhood  %in% c( 'Old_Town… <tibble [3… <tibble [1…
#>  4         1        4 ( Neighborhood  %in% c( 'Old_Town… <tibble [4… <tibble [1…
#>  5         1        5 ( Central_Air == 'N' ) & ( Gr_Liv… <tibble [4… <tibble [1…
#>  6         1        6 ( Longitude <= -93.652023 ) & ( N… <tibble [4… <tibble [1…
#>  7         1        7 ( Gr_Liv_Area > 3.2284005 ) & ( N… <tibble [4… <tibble [1…
#>  8         1        8 ( Neighborhood  %in% c( 'North_Am… <tibble [4… <tibble [1…
#>  9         1        9 ( Latitude <= 42.009399 ) & ( Nei… <tibble [3… <tibble [1…
#> 10         1       10 ( Neighborhood  %in% c( 'College_… <tibble [4… <tibble [1…
#> # … with 147 more rows

cb_res$estimate[[1]]
#> # A tibble: 4 x 2
#>   term        estimate
#>   <chr>          <dbl>
#> 1 (Intercept)  -408.  
#> 2 Longitude      -1.43
#> 3 Latitude        6.6 
#> 4 Gr_Liv_Area     0.7 
cb_res$statistic[[1]]
#> # A tibble: 1 x 6
#>   num_conditions coverage  mean   min   max  error
#>            <dbl>    <dbl> <dbl> <dbl> <dbl>  <dbl>
#> 1              2      154  4.94  4.11  5.31 0.0956
# }

# ------------------------------------------------------------------------------

# \donttest{
library(recipes)
#> 
#> Attaching package: ‘recipes’
#> The following object is masked from ‘package:stats’:
#> 
#>     step

xrf_reg_mod <-
  rule_fit(trees = 10, penalty = .001) %>%
  set_engine("xrf") %>%
  set_mode("regression")

# Make dummy variables since xgboost will not
ames_rec <-
  recipe(Sale_Price ~ Neighborhood + Longitude + Latitude +
         Gr_Liv_Area + Central_Air,
         data = ames) %>%
  step_dummy(Neighborhood, Central_Air) %>%
  step_zv(all_predictors())

ames_processed <- prep(ames_rec) %>% bake(new_data = NULL)

set.seed(1)
xrf_reg_fit <-
  xrf_reg_mod %>%
  fit(Sale_Price ~ ., data = ames_processed)
#> [16:18:45] WARNING: amalgamation/../src/objective/regression_obj.cu:170: reg:linear is now deprecated in favor of reg:squarederror.
#> [16:18:45] WARNING: amalgamation/../src/learner.cc:541: 
#> Parameters: { nrounds } might not be used.
#> 
#>   This may not be accurate due to some parameters are only used in language bindings but
#>   passed down to XGBoost core.  Or some parameters are not used but slip through this
#>   verification. Please open an issue if you find above cases.
#> 
#> 
#> New names:
#> * . -> ....1
#> * . -> ....2
#> * . -> ....3
#> * . -> ....4
#> * . -> ....5
#> * ...

xrf_rule_res <- tidy(xrf_reg_fit)
xrf_rule_res$rule[nrow(xrf_rule_res)] %>% rlang::parse_expr()
#> (Gr_Liv_Area < 3.30210185) & (Gr_Liv_Area < 3.38872266) & (Gr_Liv_Area >= 
#>     2.94571471) & (Gr_Liv_Area >= 3.24870872) & (Latitude < 42.0271072) & 
#>     (Neighborhood_Old_Town >= -9.53674316e-07)

xrf_col_res <- tidy(xrf_reg_fit, unit = "columns")
xrf_col_res
#> # A tibble: 149 x 3
#>    rule_id term           estimate
#>    <chr>   <chr>             <dbl>
#>  1 r0_1    Gr_Liv_Area   -1.27e- 2
#>  2 r2_4    Gr_Liv_Area   -3.70e-10
#>  3 r2_2    Gr_Liv_Area    7.59e- 3
#>  4 r2_4    Central_Air_Y -3.70e-10
#>  5 r3_5    Longitude      1.06e- 1
#>  6 r3_6    Longitude      2.65e- 2
#>  7 r3_5    Latitude       1.06e- 1
#>  8 r3_6    Latitude       2.65e- 2
#>  9 r3_5    Longitude      1.06e- 1
#> 10 r3_6    Longitude      2.65e- 2
#> # … with 139 more rows
# }