h2o_tree_convertR_via_mojo: Convert h2o gbm or drf to tabular structure via mojo

In richardangell/h2oTreeHelpR: h2o Tree Helper

Description

Takes a h2o tree based model (gbm or drf) and returns a list of data.frames representing each tree in the model in tabular structure. In order to access the tree internals this functions exports the h2o model to a mojo .zip and reads the .gv files from within.

Usage

h2o_tree_convertR_via_mojo(h2o_model, output_subdir = getwd(),
  get_internal_predictions = FALSE, delete_intermediate_files = TRUE)

Arguments

h2o_model	gbm or drf h2o model.
output_subdir	directory to output intermediate files (mojo .zip, .gv and model.ini files) to. Default is location is current working dir. Files are put into a sudir with date-time in name to avoid conflicts.
get_internal_predictions	logical default = FALSE, should predictions for internal (non-terminal nodes) be extracted?
delete_intermediate_files	should intermediate files output in processing be deleted? Default = TRUE.

Value

returns a list containing a data.frame for each tree containing the tree structure. Tree structure data.frames contain the following columns;

• "node" name of the node in the tree

• "node_text" complete text associated with the node

• "predictions" predicted values for terminal nodes or NA

• "left_split" left child node or NA if a terminal node

• "right_split" right child node or NA if a terminal node

• "left_split_levels" levels of the split variable that are sent to the left child node from the current node, separated by |

• "right_split_levels" levels of the split variable that are sent to the left child node from the current node, separated by |

• "NA_direction" the direction missing values are sent from the current node

• "node_text_label" the label arg of node_text, specifically the split condition for numeric split variables or the variable name for categorical variables

• "node_variable_type" split variable type for node, either categorical or numeric

• "split_column" the name of the split column for the current node

• "node_split_point" the split value for the current node if numeric, otherwise NA

Examples

library(h2o)

h2o.init()

prostate.hex = h2o.uploadFile(path = system.file("extdata",
                                                 "prostate.csv",
                                                 package = "h2o"),
                              destination_frame = "prostate.hex")

prostate.hex["RACE"] = as.factor(prostate.hex["RACE"])

prostate.hex["DPROS"] = as.factor(prostate.hex["DPROS"])

expl_cols <- c("AGE", "RACE", "DPROS", "DCAPS", "PSA", "VOL", "GLEASON")

prostate.gbm = h2o.gbm(x = expl_cols,
                       y = "CAPSULE",
                       training_frame = prostate.hex,
                       ntrees = 5,
                       max_depth = 5,
                       learn_rate = 0.1)

h2o_trees <- h2o_tree_convertR_via_mojo(h2o_model = prostate.gbm)

richardangell/h2oTreeHelpR documentation built on July 3, 2019, 7:35 a.m.