discretize_xgb.md
In embed: Extra Recipes for Encoding Predictors

run_xgboost for classification

Code
  xgboost
Output
  ##### xgb.Booster
  raw: 74.2 Kb
  call:
  xgboost::xgb.train(params = .params, data = .train, nrounds = 100,
  watchlist = list(train = .train, test = .test), verbose = 0,
  early_stopping_rounds = 10, tree_method = "hist", objective = .objective,
  nthread = 1)
  params (as set within xgb.train):
  eta = "0.3", max_bin = "10", max_depth = "1", min_child_weight = "5", tree_method = "hist", objective = "binary:logistic", nthread = "1", validate_parameters = "TRUE"
  xgb.attributes:
  best_iteration, best_msg, best_ntreelimit, best_score, niter
  callbacks:
  cb.evaluation.log()
  cb.early.stop(stopping_rounds = early_stopping_rounds, maximize = maximize,
  verbose = verbose)
  # of features: 13
  niter: 96
  best_iteration : 86
  best_ntreelimit : 86
  best_score : 0.4421503
  best_msg : [86]   train-logloss:0.417583  test-logloss:0.442150
  nfeatures : 13
  evaluation_log:
  iter train_logloss test_logloss
  <num>         <num>        <num>
  1     0.6279229    0.6303495
  2     0.5869984    0.5894989
  ---           ---          ---
  95     0.4157892    0.4425857
  96     0.4156102    0.4432699

run_xgboost for multi-classification

Code
  xgboost
Output
  ##### xgb.Booster
  raw: 149.7 Kb
  call:
  xgboost::xgb.train(params = .params, data = .train, nrounds = 100,
  watchlist = list(train = .train, test = .test), verbose = 0,
  early_stopping_rounds = 10, tree_method = "hist", objective = .objective,
  nthread = 1)
  params (as set within xgb.train):
  eta = "0.3", max_bin = "10", max_depth = "1", min_child_weight = "5", num_class = "6", tree_method = "hist", objective = "multi:softprob", nthread = "1", validate_parameters = "TRUE"
  xgb.attributes:
  best_iteration, best_msg, best_ntreelimit, best_score, niter
  callbacks:
  cb.evaluation.log()
  cb.early.stop(stopping_rounds = early_stopping_rounds, maximize = maximize,
  verbose = verbose)
  # of features: 30
  niter: 33
  best_iteration : 23
  best_ntreelimit : 23
  best_score : 1.246428
  best_msg : [23]   train-mlogloss:1.178121 test-mlogloss:1.246428
  nfeatures : 30
  evaluation_log:
  iter train_mlogloss test_mlogloss
  <num>          <num>         <num>
  1       1.623174      1.631783
  2       1.515108      1.531188
  ---            ---           ---
  32       1.159813      1.249701
  33       1.158088      1.250462

run_xgboost for regression

Code
  xgboost
Output
  ##### xgb.Booster
  raw: 40.2 Kb
  call:
  xgboost::xgb.train(params = .params, data = .train, nrounds = 100,
  watchlist = list(train = .train, test = .test), verbose = 0,
  early_stopping_rounds = 10, tree_method = "hist", objective = .objective,
  nthread = 1)
  params (as set within xgb.train):
  eta = "0.3", max_bin = "10", max_depth = "1", min_child_weight = "5", tree_method = "hist", objective = "reg:squarederror", nthread = "1", validate_parameters = "TRUE"
  xgb.attributes:
  best_iteration, best_msg, best_ntreelimit, best_score, niter
  callbacks:
  cb.evaluation.log()
  cb.early.stop(stopping_rounds = early_stopping_rounds, maximize = maximize,
  verbose = verbose)
  # of features: 73
  niter: 50
  best_iteration : 40
  best_ntreelimit : 40
  best_score : 0.1165337
  best_msg : [40]   train-rmse:0.064010 test-rmse:0.116534
  nfeatures : 73
  evaluation_log:
  iter train_rmse test_rmse
  <num>      <num>     <num>
  1 3.31007782 3.3068878
  2 2.31969213 2.3262197
  ---        ---       ---
  49 0.06207940 0.1175223
  50 0.06191289 0.1188113

xgb_binning for classification

Code
  xgb_binning
Output
  [1]  1  2  3  5  6  9 12 15 20

xgb_binning for multi-classification

Code
  xgb_binning
Output
  [1] 26 31 35 38

xgb_binning for regression

Code
  xgb_binning
Output
  [1] 42.01972 42.02510 42.03122 42.03462 42.03840 42.04638 42.05236 42.05917

step_discretize_xgb for classification

Code
  xgb_train_bins[1:10, ]
Output
  # A tibble: 10 x 3
     x               z               class
     <fct>           <fct>           <fct>
   1 [0.6808, Inf]   [0.4208,0.7999) a    
   2 [0.5749,0.6808) [0.4208,0.7999) b    
   3 [0.3687,0.5749) [-Inf,0.3327)   b    
   4 [0.5749,0.6808) [0.4208,0.7999) b    
   5 [0.6808, Inf]   [0.4208,0.7999) a    
   6 [0.5749,0.6808) [0.7999, Inf]   a    
   7 [0.5749,0.6808) [-Inf,0.3327)   b    
   8 [0.5749,0.6808) [-Inf,0.3327)   a    
   9 [0.2779,0.3687) [0.4208,0.7999) a    
  10 [0.3687,0.5749) [0.4208,0.7999) b

Code
  xgb_test_bins[1:10, ]
Output
  # A tibble: 10 x 3
     x               z               class
     <fct>           <fct>           <fct>
   1 [0.5749,0.6808) [-Inf,0.3327)   b    
   2 [0.5749,0.6808) [0.4208,0.7999) b    
   3 [0.3687,0.5749) [-Inf,0.3327)   b    
   4 [0.2779,0.3687) [0.4208,0.7999) b    
   5 [0.6808, Inf]   [0.7999, Inf]   a    
   6 [0.3687,0.5749) [-Inf,0.3327)   b    
   7 [0.2779,0.3687) [0.4208,0.7999) a    
   8 [0.6808, Inf]   [0.7999, Inf]   a    
   9 [0.3687,0.5749) [0.7999, Inf]   b    
  10 [0.3687,0.5749) [0.4208,0.7999) b

Code
  prep(step_discretize_xgb(recipe(class ~ ., data = sim_tr_cls[1:9, ]),
  all_predictors(), outcome = "class"))
Condition
  Error in `step_discretize_xgb()`:
  Caused by error in `prep()`:
  ! Too few observations in the early stopping validation set.
  i Consider increasing the `sample_val` parameter.

Code
  set.seed(1)
  prep(step_discretize_xgb(recipe(Status ~ ., data = credit_data_train), Time,
  outcome = "Status"), retain = TRUE)
Condition
  Warning:
  More than 20 unique training set values are required.
  i Predictors 'Time' were not processed; their original values will be used.
Message

  -- Recipe ----------------------------------------------------------------------

  -- Inputs 
  Number of variables by role
  outcome:    1
  predictor: 13

  -- Training information 
  Training data contained 3340 data points and 301 incomplete rows.

  -- Operations 
  * Discretizing variables using xgboost: <none> | Trained

step_discretize_xgb for multi-classification

Code
  xgb_train_bins[1:10, ]
Output
  # A tibble: 10 x 3
     x               z             class
     <fct>           <fct>         <fct>
   1 [0.6879, Inf]   [0.3274, Inf] c    
   2 [0.5863,0.6879) [0.3274, Inf] b    
   3 [0.3821,0.5863) [-Inf,0.3274) b    
   4 [0.5863,0.6879) [0.3274, Inf] b    
   5 [0.6879, Inf]   [0.3274, Inf] c    
   6 [0.5863,0.6879) [0.3274, Inf] c    
   7 [0.5863,0.6879) [-Inf,0.3274) b    
   8 [0.5863,0.6879) [-Inf,0.3274) c    
   9 [-Inf,0.2887)   [0.3274, Inf] a    
  10 [0.3821,0.5863) [0.3274, Inf] b

Code
  xgb_test_bins[1:10, ]
Output
  # A tibble: 10 x 3
     x               z             class
     <fct>           <fct>         <fct>
   1 [0.5863,0.6879) [-Inf,0.3274) b    
   2 [0.5863,0.6879) [0.3274, Inf] b    
   3 [0.3821,0.5863) [-Inf,0.3274) b    
   4 [0.2887,0.3821) [0.3274, Inf] b    
   5 [0.6879, Inf]   [0.3274, Inf] c    
   6 [0.3821,0.5863) [-Inf,0.3274) b    
   7 [0.2887,0.3821) [0.3274, Inf] a    
   8 [0.6879, Inf]   [0.3274, Inf] c    
   9 [0.3821,0.5863) [0.3274, Inf] b    
  10 [0.3821,0.5863) [0.3274, Inf] b