details_boost_tree_xgboost: Boosted trees via xgboost

In parsnip: A Common API to Modeling and Analysis Functions

Boosted trees via xgboost

Description

xgboost::xgb.train() creates a series of decision trees forming an ensemble. Each tree depends on the results of previous trees. All trees in the ensemble are combined to produce a final prediction.

Details

For this engine, there are multiple modes: classification and regression

Tuning Parameters

This model has 8 tuning parameters:

• tree_depth: Tree Depth (type: integer, default: 6L)

• trees: # Trees (type: integer, default: 15L)

• learn_rate: Learning Rate (type: double, default: 0.3)

• mtry: # Randomly Selected Predictors (type: integer, default: see below)

• min_n: Minimal Node Size (type: integer, default: 1L)

• loss_reduction: Minimum Loss Reduction (type: double, default: 0.0)

• sample_size: Proportion Observations Sampled (type: double, default: 1.0)

• stop_iter: # Iterations Before Stopping (type: integer, default: Inf)

For mtry, the default value of NULL translates to using all available columns.

Translation from parsnip to the original package (regression)

boost_tree(
  mtry = integer(), trees = integer(), min_n = integer(), tree_depth = integer(),
  learn_rate = numeric(), loss_reduction = numeric(), sample_size = numeric(),
  stop_iter = integer()
) %>%
  set_engine("xgboost") %>%
  set_mode("regression") %>%
  translate()

## Boosted Tree Model Specification (regression)
## 
## Main Arguments:
##   mtry = integer()
##   trees = integer()
##   min_n = integer()
##   tree_depth = integer()
##   learn_rate = numeric()
##   loss_reduction = numeric()
##   sample_size = numeric()
##   stop_iter = integer()
## 
## Computational engine: xgboost 
## 
## Model fit template:
## parsnip::xgb_train(x = missing_arg(), y = missing_arg(), weights = missing_arg(), 
##     colsample_bynode = integer(), nrounds = integer(), min_child_weight = integer(), 
##     max_depth = integer(), eta = numeric(), gamma = numeric(), 
##     subsample = numeric(), early_stop = integer(), nthread = 1, 
##     verbose = 0)

Translation from parsnip to the original package (classification)

boost_tree(
  mtry = integer(), trees = integer(), min_n = integer(), tree_depth = integer(),
  learn_rate = numeric(), loss_reduction = numeric(), sample_size = numeric(),
  stop_iter = integer()
) %>% 
  set_engine("xgboost") %>% 
  set_mode("classification") %>% 
  translate()

## Boosted Tree Model Specification (classification)
## 
## Main Arguments:
##   mtry = integer()
##   trees = integer()
##   min_n = integer()
##   tree_depth = integer()
##   learn_rate = numeric()
##   loss_reduction = numeric()
##   sample_size = numeric()
##   stop_iter = integer()
## 
## Computational engine: xgboost 
## 
## Model fit template:
## parsnip::xgb_train(x = missing_arg(), y = missing_arg(), weights = missing_arg(), 
##     colsample_bynode = integer(), nrounds = integer(), min_child_weight = integer(), 
##     max_depth = integer(), eta = numeric(), gamma = numeric(), 
##     subsample = numeric(), early_stop = integer(), nthread = 1, 
##     verbose = 0)

xgb_train() is a wrapper around xgboost::xgb.train() (and other functions) that makes it easier to run this model.

Preprocessing requirements

xgboost does not have a means to translate factor predictors to grouped splits. Factor/categorical predictors need to be converted to numeric values (e.g., dummy or indicator variables) for this engine. When using the formula method via fit.model_spec(), parsnip will convert factor columns to indicators using a one-hot encoding.

For classification, non-numeric outcomes (i.e., factors) are internally converted to numeric. For binary classification, the event_level argument of set_engine() can be set to either "first" or "second" to specify which level should be used as the event. This can be helpful when a watchlist is used to monitor performance from with the xgboost training process.

Other details

Interfacing with the params argument

The xgboost function that parsnip indirectly wraps, xgboost::xgb.train(), takes most arguments via the params list argument. To supply engine-specific arguments that are documented in xgboost::xgb.train() as arguments to be passed via params, supply the list elements directly as named arguments to set_engine() rather than as elements in params. For example, pass a non-default evaluation metric like this:

# good
boost_tree() %>%
  set_engine("xgboost", eval_metric = "mae")

## Boosted Tree Model Specification (unknown mode)
## 
## Engine-Specific Arguments:
##   eval_metric = mae
## 
## Computational engine: xgboost

…rather than this:

# bad
boost_tree() %>%
  set_engine("xgboost", params = list(eval_metric = "mae"))

## Boosted Tree Model Specification (unknown mode)
## 
## Engine-Specific Arguments:
##   params = list(eval_metric = "mae")
## 
## Computational engine: xgboost

parsnip will then route arguments as needed. In the case that arguments are passed to params via set_engine(), parsnip will warn and re-route the arguments as needed. Note, though, that arguments passed to params cannot be tuned.

Sparse matrices

xgboost requires the data to be in a sparse format. If your predictor data are already in this format, then use fit_xy.model_spec() to pass it to the model function. Otherwise, parsnip converts the data to this format.

Parallel processing

By default, the model is trained without parallel processing. This can be change by passing the nthread parameter to set_engine(). However, it is unwise to combine this with external parallel processing when using the package.

Interpreting mtry

The mtry argument denotes the number of predictors that will be randomly sampled at each split when creating tree models.

Some engines, such as "xgboost", "xrf", and "lightgbm", interpret their analogue to the mtry argument as the proportion of predictors that will be randomly sampled at each split rather than the count. In some settings, such as when tuning over preprocessors that influence the number of predictors, this parameterization is quite helpful—interpreting mtry as a proportion means that ⁠[0, 1]⁠ is always a valid range for that parameter, regardless of input data.

parsnip and its extensions accommodate this parameterization using the counts argument: a logical indicating whether mtry should be interpreted as the number of predictors that will be randomly sampled at each split. TRUE indicates that mtry will be interpreted in its sense as a count, FALSE indicates that the argument will be interpreted in its sense as a proportion.

mtry is a main model argument for boost_tree() and rand_forest(), and thus should not have an engine-specific interface. So, regardless of engine, counts defaults to TRUE. For engines that support the proportion interpretation (currently "xgboost" and "xrf", via the rules package, and "lightgbm" via the bonsai package) the user can pass the counts = FALSE argument to set_engine() to supply mtry values within ⁠[0, 1]⁠.

Early stopping

The stop_iter() argument allows the model to prematurely stop training if the objective function does not improve within early_stop iterations.

The best way to use this feature is in conjunction with an internal validation set. To do this, pass the validation parameter of xgb_train() via the parsnip set_engine() function. This is the proportion of the training set that should be reserved for measuring performance (and stopping early).

If the model specification has early_stop >= trees, early_stop is converted to trees - 1 and a warning is issued.

Note that, since the validation argument provides an alternative interface to watchlist, the watchlist argument is guarded by parsnip and will be ignored (with a warning) if passed.

Objective function

parsnip chooses the objective function based on the characteristics of the outcome. To use a different loss, pass the objective argument to set_engine() directly.

Saving fitted model objects

This model object contains data that are not required to make predictions. When saving the model for the purpose of prediction, the size of the saved object might be substantially reduced by using functions from the butcher package.

Models fitted with this engine may require native serialization methods to be properly saved and/or passed between R sessions. To learn more about preparing fitted models for serialization, see the bundle package.

Examples

The “Fitting and Predicting with parsnip” article contains examples for boost_tree() with the "xgboost" engine.

References

• XGBoost: A Scalable Tree Boosting System

• Kuhn, M, and K Johnson. 2013. Applied Predictive Modeling. Springer.

parsnip documentation built on June 24, 2024, 5:14 p.m.