Spark ML -- Random Forest

Perform classification and regression using random forests.

ml_random_forest_classifier(x, formula = NULL, num_trees = 20L,
  subsampling_rate = 1, max_depth = 5L, min_instances_per_node = 1L,
  feature_subset_strategy = "auto", impurity = "gini", min_info_gain = 0,
  max_bins = 32L, seed = NULL, thresholds = NULL,
  checkpoint_interval = 10L, cache_node_ids = FALSE,
  max_memory_in_mb = 256L, features_col = "features", label_col = "label",
  prediction_col = "prediction", probability_col = "probability",
  raw_prediction_col = "rawPrediction",
  uid = random_string("random_forest_classifier_"), ...)

ml_random_forest(x, formula = NULL, type = c("auto", "regression",
  "classification"), features_col = "features", label_col = "label",
  prediction_col = "prediction", probability_col = "probability",
  raw_prediction_col = "rawPrediction", feature_subset_strategy = "auto",
  impurity = "auto", checkpoint_interval = 10L, max_bins = 32L,
  max_depth = 5L, num_trees = 20L, min_info_gain = 0,
  min_instances_per_node = 1L, subsampling_rate = 1, seed = NULL,
  thresholds = NULL, cache_node_ids = FALSE, max_memory_in_mb = 256L,
  uid = random_string("random_forest_"), response = NULL, features = NULL,
  ...)

ml_random_forest_regressor(x, formula = NULL, num_trees = 20L,
  subsampling_rate = 1, max_depth = 5L, min_instances_per_node = 1L,
  feature_subset_strategy = "auto", impurity = "variance",
  min_info_gain = 0, max_bins = 32L, seed = NULL,
  checkpoint_interval = 10L, cache_node_ids = FALSE,
  max_memory_in_mb = 256L, features_col = "features", label_col = "label",
  prediction_col = "prediction",
  uid = random_string("random_forest_regressor_"), ...)

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.

num_trees

Number of trees to train (>= 1). If 1, then no bootstrapping is used. If > 1, then bootstrapping is done.

subsampling_rate

Fraction of the training data used for learning each decision tree, in range (0, 1]. (default = 1.0)

max_depth

Maximum depth of the tree (>= 0); that is, the maximum number of nodes separating any leaves from the root of the tree.

min_instances_per_node

Minimum number of instances each child must have after split.

feature_subset_strategy

The number of features to consider for splits at each tree node. See details for options.

impurity

Criterion used for information gain calculation. Supported: "entropy" and "gini" (default) for classification and "variance" (default) for regression. For ml_decision_tree, setting "auto" will default to the appropriate criterion based on model type.

min_info_gain

Minimum information gain for a split to be considered at a tree node. Should be >= 0, defaults to 0.

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.

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; currently unused.

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.

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_predictor 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.

Details

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_random_forest is a wrapper around ml_random_forest_regressor.tbl_spark and ml_random_forest_classifier.tbl_spark and calls the appropriate method based on model type.

See also

See http://spark.apache.org/docs/latest/ml-classification-regression.html for more information on the set of supervised learning algorithms.

Other ml algorithms: ml_aft_survival_regression, ml_decision_tree_classifier, ml_gbt_classifier, ml_generalized_linear_regression, ml_isotonic_regression, ml_linear_regression, ml_linear_svc, ml_logistic_regression, ml_multilayer_perceptron_classifier, ml_naive_bayes, ml_one_vs_rest