ml_gbt_classifier: Spark ML -- Gradient Boosted Trees

Description

Perform binary classification and regression using gradient boosted trees. Multiclass classification is not supported yet.

Usage

ml_gbt_classifier(
  x,
  formula = NULL,
  max_iter = 20,
  max_depth = 5,
  step_size = 0.1,
  subsampling_rate = 1,
  feature_subset_strategy = "auto",
  min_instances_per_node = 1L,
  max_bins = 32,
  min_info_gain = 0,
  loss_type = "logistic",
  seed = NULL,
  thresholds = NULL,
  checkpoint_interval = 10,
  cache_node_ids = FALSE,
  max_memory_in_mb = 256,
  features_col = "features",
  label_col = "label",
  prediction_col = "prediction",
  probability_col = "probability",
  raw_prediction_col = "rawPrediction",
  uid = random_string("gbt_classifier_"),
  ...
)
ml_gradient_boosted_trees(
  x,
  formula = NULL,
  type = c("auto", "regression", "classification"),
  features_col = "features",
  label_col = "label",
  prediction_col = "prediction",
  probability_col = "probability",
  raw_prediction_col = "rawPrediction",
  checkpoint_interval = 10,
  loss_type = c("auto", "logistic", "squared", "absolute"),
  max_bins = 32,
  max_depth = 5,
  max_iter = 20L,
  min_info_gain = 0,
  min_instances_per_node = 1,
  step_size = 0.1,
  subsampling_rate = 1,
  feature_subset_strategy = "auto",
  seed = NULL,
  thresholds = NULL,
  cache_node_ids = FALSE,
  max_memory_in_mb = 256,
  uid = random_string("gradient_boosted_trees_"),
  response = NULL,
  features = NULL,
  ...
)
ml_gbt_regressor(
  x,
  formula = NULL,
  max_iter = 20,
  max_depth = 5,
  step_size = 0.1,
  subsampling_rate = 1,
  feature_subset_strategy = "auto",
  min_instances_per_node = 1,
  max_bins = 32,
  min_info_gain = 0,
  loss_type = "squared",
  seed = NULL,
  checkpoint_interval = 10,
  cache_node_ids = FALSE,
  max_memory_in_mb = 256,
  features_col = "features",
  label_col = "label",
  prediction_col = "prediction",
  uid = random_string("gbt_regressor_"),
  ...
)

Value

The object returned depends on the class of x.

spark_connection: When x is a spark_connection, the function returns an instance of a ml_estimator object. The object contains a pointer to a Spark Predictor object and can be used to compose Pipeline objects.
ml_pipeline: When x is a ml_pipeline, the function returns a ml_pipeline with the predictor appended to the pipeline.
tbl_spark: When x is a tbl_spark, a predictor is constructed then immediately fit with the input tbl_spark, returning a prediction model.
tbl_spark, with formula: specified When formula is specified, the input tbl_spark is first transformed using a RFormula transformer before being fit by the predictor. The object returned in this case is a ml_model which is a wrapper of a ml_pipeline_model.

Arguments

x: A spark_connection, ml_pipeline, or a tbl_spark.
formula: Used when x is a tbl_spark. R formula as a character string or a formula. This is used to transform the input dataframe before fitting, see ft_r_formula for details.
max_iter: Maxmimum number of iterations.
max_depth: Maximum depth of the tree (>= 0); that is, the maximum number of nodes separating any leaves from the root of the tree.
step_size: Step size (a.k.a. learning rate) in interval (0, 1] for shrinking the contribution of each estimator. (default = 0.1)
subsampling_rate: Fraction of the training data used for learning each decision tree, in range (0, 1]. (default = 1.0)
feature_subset_strategy: The number of features to consider for splits at each tree node. See details for options.
min_instances_per_node: Minimum number of instances each child must have after split.
max_bins: The maximum number of bins used for discretizing continuous features and for choosing how to split on features at each node. More bins give higher granularity.
min_info_gain: Minimum information gain for a split to be considered at a tree node. Should be >= 0, defaults to 0.
loss_type: Loss function which GBT tries to minimize. Supported: "squared" (L2) and "absolute" (L1) (default = squared) for regression and "logistic" (default) for classification. For ml_gradient_boosted_trees, setting "auto" will default to the appropriate loss type based on model type.
seed: Seed for random numbers.
thresholds: Thresholds in multi-class classification to adjust the probability of predicting each class. Array must have length equal to the number of classes, with values > 0 excepting that at most one value may be 0. The class with largest value p/t is predicted, where p is the original probability of that class and t is the class's threshold.
checkpoint_interval: Set checkpoint interval (>= 1) or disable checkpoint (-1). E.g. 10 means that the cache will get checkpointed every 10 iterations, defaults to 10.
cache_node_ids: If FALSE, the algorithm will pass trees to executors to match instances with nodes. If TRUE, the algorithm will cache node IDs for each instance. Caching can speed up training of deeper trees. Defaults to FALSE.
max_memory_in_mb: Maximum memory in MB allocated to histogram aggregation. If too small, then 1 node will be split per iteration, and its aggregates may exceed this size. Defaults to 256.
features_col: Features column name, as a length-one character vector. The column should be single vector column of numeric values. Usually this column is output by ft_r_formula.
label_col: Label column name. The column should be a numeric column. Usually this column is output by ft_r_formula.
prediction_col: Prediction column name.
probability_col: Column name for predicted class conditional probabilities.
raw_prediction_col: Raw prediction (a.k.a. confidence) column name.
uid: A character string used to uniquely identify the ML estimator.
...: Optional arguments; see Details.
type: The type of model to fit. "regression" treats the response as a continuous variable, while "classification" treats the response as a categorical variable. When "auto" is used, the model type is inferred based on the response variable type -- if it is a numeric type, then regression is used; classification otherwise.
response: (Deprecated) The name of the response column (as a length-one character vector.)
features: (Deprecated) The name of features (terms) to use for the model fit.

Details

When x is a tbl_spark and formula (alternatively, response and features) is specified, the function returns a ml_model object wrapping a ml_pipeline_model which contains data pre-processing transformers, the ML predictor, and, for classification models, a post-processing transformer that converts predictions into class labels. For classification, an optional argument predicted_label_col (defaults to "predicted_label") can be used to specify the name of the predicted label column. In addition to the fitted ml_pipeline_model, ml_model objects also contain a ml_pipeline object where the ML predictor stage is an estimator ready to be fit against data. This is utilized by ml_save with type = "pipeline" to faciliate model refresh workflows.

The supported options for feature_subset_strategy are

"auto": Choose automatically for task: If num_trees == 1, set to "all". If num_trees > 1 (forest), set to "sqrt" for classification and to "onethird" for regression.
"all": use all features
"onethird": use 1/3 of the features
"sqrt": use use sqrt(number of features)
"log2": use log2(number of features)
"n": when n is in the range (0, 1.0], use n * number of features. When n is in the range (1, number of features), use n features. (default = "auto")

ml_gradient_boosted_trees is a wrapper around ml_gbt_regressor.tbl_spark and ml_gbt_classifier.tbl_spark and calls the appropriate method based on model type.

Examples

Run this code

if (FALSE) {
sc <- spark_connect(master = "local")
iris_tbl <- sdf_copy_to(sc, iris, name = "iris_tbl", overwrite = TRUE)

partitions <- iris_tbl %>%
  sdf_random_split(training = 0.7, test = 0.3, seed = 1111)

iris_training <- partitions$training
iris_test <- partitions$test

gbt_model <- iris_training %>%
  ml_gradient_boosted_trees(Sepal_Length ~ Petal_Length + Petal_Width)

pred <- ml_predict(gbt_model, iris_test)

ml_regression_evaluator(pred, label_col = "Sepal_Length")
}