GH-15780 set weak learner parameter API

h2o-algos/src/main/java/hex/schemas/AdaBoostV3.java

@@ -23,6 +23,7 @@ public static final class AdaBoostParametersV3 extends ModelParametersSchemaV3<A
                 "nlearners",
                 "weak_learner",
                 "learn_rate",
+                "weak_learner_params",
                 "seed",
         };
 
@@ -35,6 +36,9 @@ public static final class AdaBoostParametersV3 extends ModelParametersSchemaV3<A
         @API(help="Learning rate (from 0.0 to 1.0)", gridable = true)
         public double learn_rate;
 
+        @API(help = "Customized parameters for the weak_learner algorithm.", gridable=true)
+        public String weak_learner_params;
+
         @API(help = "Seed for pseudo random number generator (if applicable)", gridable = true)
         public long seed;
     }

h2o-algos/src/test/java/hex/adaboost/AdaBoostTest.java

@@ -4,6 +4,7 @@
 import hex.deeplearning.DeepLearningModel;
 import hex.genmodel.algos.tree.SharedTreeSubgraph;
 import hex.glm.GLMModel;
+import hex.tree.SharedTreeModel;
 import hex.tree.drf.DRFModel;
 import hex.tree.gbm.GBMModel;
 import org.junit.Before;
@@ -629,7 +630,7 @@ public void testTrainWithCustomWeakLearnersGBM() {
             p._nlearners = nlearners;
             p._weak_learner = AdaBoostModel.Algorithm.GBM;
             p._response_column = response;
-            p._weak_learner_params = "{ntrees:3}";
+            p._weak_learner_params = "{ntrees:3, 'histogram_type': 'UniformAdaptive'}";
 
             AdaBoost adaBoost = new AdaBoost(p);
             AdaBoostModel adaBoostModel = adaBoost.trainModel().get();
@@ -640,6 +641,7 @@ public void testTrainWithCustomWeakLearnersGBM() {
                 System.out.println("GBM model = " + i);
                 GBMModel gbmModel = DKV.getGet(adaBoostModel._output.models[i]);
                 assertEquals(3, gbmModel._output._ntrees);
+                assertEquals(SharedTreeModel.SharedTreeParameters.HistogramType.UniformAdaptive, gbmModel._parms._histogram_type);
             }
             Frame score = adaBoostModel.score(train);
             Scope.track(score);

h2o-bindings/bin/custom/R/gen_adaboost.py

@@ -1,21 +1,33 @@
+def update_param(name, param):
+    if name == 'weak_learner_params':
+        param['default_value'] = None
+        return param
+    return None  # param untouched
+
 extensions = dict(
     skip_default_set_params_for=['training_frame', 'ignored_columns', 'response_column', 
-                                 'max_confusion_matrix_size', 'distribution', 'offset_column'],
+                                 'max_confusion_matrix_size', 'distribution', 'offset_column', 'weak_learner_params'],
     set_required_params="""
 parms$training_frame <- training_frame
 args <- .verify_dataxy(training_frame, x, y)
 parms$ignored_columns <- args$x_ignore
 parms$response_column <- args$y
 """,
+    set_params="""
+if (!missing(weak_learner_params))
+    parms$weak_learner_params <- as.character(toJSON(weak_learner_params, pretty = TRUE, auto_unbox = TRUE))
+"""
 )
 
-
 doc = dict(
     preamble="""
 Build an AdaBoost model
 
 Builds an AdaBoost model on an H2OFrame.
 """,
+    params=dict(
+        weak_learner_params="Customized parameters for the weak_learner algorithm. E.g list(ntrees=3, max_depth=2, histogram_type='UniformAdaptive'))",
+    ),
     returns="""
 Creates a \linkS4class{H2OModel} object of the right type.
 """,

h2o-bindings/bin/custom/python/gen_adaboost.py

@@ -1,8 +1,59 @@
-options = dict(
+def update_param(name, param):
+    if name == 'weak_learner_params':
+        param['type'] = 'KeyValue'
+        param['default_value'] = None
+        return param
+    return None  # param untouched
+
+extensions = dict(
+    __imports__="""
+import ast
+import json
+from h2o.estimators.estimator_base import H2OEstimator
+from h2o.exceptions import H2OValueError
+from h2o.frame import H2OFrame
+from h2o.utils.typechecks import assert_is_type, Enum, numeric
+""",
 )
 
 doc = dict(
     __class__="""
 Builds an AdaBoost model
 """
 )
+
+overrides = dict(
+    weak_learner_params=dict(
+        getter="""
+if self._parms.get("{sname}") != None:
+    return json.loads(self._parms.get("{sname}"))
+else:
+    self._parms["{sname}"] = None
+        """,
+        setter="""
+assert_is_type({pname}, None, {ptype})
+if {pname} is not None and {pname} != "":
+    for k in {pname}:
+        weak_learner_params[k] = weak_learner_params[k]
+    self._parms["{sname}"] = str(json.dumps({pname}))
+else:
+    self._parms["{sname}"] = None
+"""
+    )
+)
+
+examples = dict(
+    weak_learner_params="""
+>>> prostate_hex = h2o.import_file("http://s3.amazonaws.com/h2o-public-test-data/smalldata/prostate/prostate.csv")
+>>> prostate_hex["CAPSULE"] = prostate_hex["CAPSULE"].asfactor()
+>>> response = "CAPSULE"
+>>> seed = 42
+>>> adaboost_model = H2OAdaBoostEstimator(seed=seed,
+...                                       weak_learner="DRF",
+...                                       weak_learner_params={'ntrees':1,'max_depth':3})
+>>> adaboost_model.train(y=response,
+...                 ignored_columns=["ID"],
+...                 training_frame=prostate_hex)
+>>> print(adaboost_model)
+""",
+)

h2o-py/h2o/estimators/adaboost.py

@@ -5,6 +5,12 @@
 # Copyright 2016 H2O.ai;  Apache License Version 2.0 (see LICENSE for details)
 #
 
+import ast
+import json
+from h2o.estimators.estimator_base import H2OEstimator
+from h2o.exceptions import H2OValueError
+from h2o.frame import H2OFrame
+from h2o.utils.typechecks import assert_is_type, Enum, numeric
 from h2o.estimators.estimator_base import H2OEstimator
 from h2o.exceptions import H2OValueError
 from h2o.frame import H2OFrame
@@ -31,6 +37,7 @@ def __init__(self,
                  nlearners=50,  # type: int
                  weak_learner="auto",  # type: Literal["auto", "drf", "glm", "gbm", "deep_learning"]
                  learn_rate=0.5,  # type: float
+                 weak_learner_params=None,  # type: Optional[dict]
                  seed=-1,  # type: int
                  ):
         """
@@ -68,6 +75,9 @@ def __init__(self,
         :param learn_rate: Learning rate (from 0.0 to 1.0)
                Defaults to ``0.5``.
         :type learn_rate: float
+        :param weak_learner_params: Customized parameters for the weak_learner algorithm.
+               Defaults to ``None``.
+        :type weak_learner_params: dict, optional
         :param seed: Seed for pseudo random number generator (if applicable)
                Defaults to ``-1``.
         :type seed: int
@@ -83,6 +93,7 @@ def __init__(self,
         self.nlearners = nlearners
         self.weak_learner = weak_learner
         self.learn_rate = learn_rate
+        self.weak_learner_params = weak_learner_params
         self.seed = seed
 
     @property
@@ -203,6 +214,42 @@ def learn_rate(self, learn_rate):
         assert_is_type(learn_rate, None, numeric)
         self._parms["learn_rate"] = learn_rate
 
+    @property
+    def weak_learner_params(self):
+        """
+        Customized parameters for the weak_learner algorithm.
+
+        Type: ``dict``.
+
+        :examples:
+
+        >>> prostate_hex = h2o.import_file("http://s3.amazonaws.com/h2o-public-test-data/smalldata/prostate/prostate.csv")
+        >>> prostate_hex["CAPSULE"] = prostate_hex["CAPSULE"].asfactor()
+        >>> response = "CAPSULE"
+        >>> seed = 42
+        >>> adaboost_model = H2OAdaBoostEstimator(seed=seed,
+        ...                                       weak_learner="DRF",
+        ...                                       weak_learner_params={'ntrees':1,'max_depth':3})
+        >>> adaboost_model.train(y=response,
+        ...                 ignored_columns=["ID"],
+        ...                 training_frame=prostate_hex)
+        >>> print(adaboost_model)
+        """
+        if self._parms.get("weak_learner_params") != None:
+            return json.loads(self._parms.get("weak_learner_params"))
+        else:
+            self._parms["weak_learner_params"] = None
+
+    @weak_learner_params.setter
+    def weak_learner_params(self, weak_learner_params):
+        assert_is_type(weak_learner_params, None, dict)
+        if weak_learner_params is not None and weak_learner_params != "":
+            for k in weak_learner_params:
+                weak_learner_params[k] = weak_learner_params[k]
+            self._parms["weak_learner_params"] = str(json.dumps(weak_learner_params))
+        else:
+            self._parms["weak_learner_params"] = None
+
     @property
     def seed(self):
         """

h2o-py/tests/testdir_algos/adaboost/pyunit_adaboost_weak_learner_params_smoke.py

@@ -0,0 +1,69 @@
+import sys, os
+sys.path.insert(1, os.path.join("..","..",".."))
+import h2o
+from tests import pyunit_utils
+from h2o.estimators import H2OAdaBoostEstimator
+
+
+def adaboost():
+    print("AdaBoost Weak Learner Params Smoke Test - test only that parameters are correctly passed to backend")
+
+    train = h2o.import_file(pyunit_utils.locate("smalldata/prostate/prostate.csv"))
+    train["CAPSULE"] = train["CAPSULE"].asfactor()
+
+    common_adaboost_def = {"nlearners": 10, "seed": 0xBEEF, "learn_rate": 0.6}
+    common_adaboost_train = {"training_frame": train, "y": "CAPSULE"}
+
+    adaboost_model = H2OAdaBoostEstimator(
+        weak_learner="DRF",
+        weak_learner_params={
+            'ntrees': 10,
+            'histogram_type': "UniformAdaptive"
+        },
+        **common_adaboost_def
+    )
+    assert isinstance(adaboost_model.weak_learner_params, dict)
+    adaboost_model.train(**common_adaboost_train)
+    assert adaboost_model._model_json is not None
+
+    adaboost_model = H2OAdaBoostEstimator(
+        weak_learner="GBM",
+        weak_learner_params={
+          'ntrees': 10,
+          'histogram_type': "UniformAdaptive",
+          "learn_rate": 0.1
+        },
+        **common_adaboost_def
+    )
+    assert isinstance(adaboost_model.weak_learner_params, dict)
+    adaboost_model.train(**common_adaboost_train)
+    assert adaboost_model._model_json is not None
+
+    adaboost_model = H2OAdaBoostEstimator(
+        weak_learner="GLM",
+        weak_learner_params={
+            'max_iterations': 10
+        },
+        **common_adaboost_def
+    )
+    assert isinstance(adaboost_model.weak_learner_params, dict)
+    adaboost_model.train(**common_adaboost_train)
+    assert adaboost_model._model_json is not None
+
+    adaboost_model = H2OAdaBoostEstimator(
+        weak_learner="DEEP_LEARNING",
+        weak_learner_params={
+            'nepochs': 10,
+            'hidden': [2, 2, 4]
+        },
+        **common_adaboost_def
+    )
+    assert isinstance(adaboost_model.weak_learner_params, dict)
+    adaboost_model.train(**common_adaboost_train)
+    assert adaboost_model._model_json is not None
+
+
+if __name__ == "__main__":
+    pyunit_utils.standalone_test(adaboost)
+else:
+    adaboost()

h2o-r/h2o-package/R/adaboost.R

@@ -28,6 +28,7 @@
 #' @param weak_learner Choose a weak learner type. Defaults to AUTO, which means DRF. Must be one of: "AUTO", "DRF", "GLM", "GBM",
 #'        "DEEP_LEARNING". Defaults to AUTO.
 #' @param learn_rate Learning rate (from 0.0 to 1.0) Defaults to 0.5.
+#' @param weak_learner_params Customized parameters for the weak_learner algorithm. E.g list(ntrees=3, max_depth=2, histogram_type='UniformAdaptive'))
 #' @param seed Seed for random numbers (affects certain parts of the algo that are stochastic and those might or might not be enabled by default).
 #'        Defaults to -1 (time-based random number).
 #' @return Creates a \linkS4class{H2OModel} object of the right type.
@@ -60,6 +61,7 @@ h2o.adaBoost <- function(x,
                          nlearners = 50,
                          weak_learner = c("AUTO", "DRF", "GLM", "GBM", "DEEP_LEARNING"),
                          learn_rate = 0.5,
+                         weak_learner_params = NULL,
                          seed = -1)
 {
   # Validate required training_frame first and other frame args: should be a valid key or an H2OFrame object
@@ -99,6 +101,9 @@ h2o.adaBoost <- function(x,
   if (!missing(seed))
     parms$seed <- seed
 
+  if (!missing(weak_learner_params))
+      parms$weak_learner_params <- as.character(toJSON(weak_learner_params, pretty = TRUE, auto_unbox = TRUE))
+
   # Error check and build model
   model <- .h2o.modelJob('adaboost', parms, h2oRestApiVersion=3, verbose=FALSE)
   return(model)
@@ -112,6 +117,7 @@ h2o.adaBoost <- function(x,
                                          nlearners = 50,
                                          weak_learner = c("AUTO", "DRF", "GLM", "GBM", "DEEP_LEARNING"),
                                          learn_rate = 0.5,
+                                         weak_learner_params = NULL,
                                          seed = -1,
                                          segment_columns = NULL,
                                          segment_models_id = NULL,
@@ -156,6 +162,9 @@ h2o.adaBoost <- function(x,
   if (!missing(seed))
     parms$seed <- seed
 
+  if (!missing(weak_learner_params))
+      parms$weak_learner_params <- as.character(toJSON(weak_learner_params, pretty = TRUE, auto_unbox = TRUE))
+
   # Build segment-models specific parameters
   segment_parms <- list()
   if (!missing(segment_columns))

h2o-r/tests/testdir_algos/adaboost/runit_adaboost_weak_learner_smoke.R

@@ -0,0 +1,24 @@
+setwd(normalizePath(dirname(R.utils::commandArgs(asValues=TRUE)$"f")))
+source("../../../scripts/h2o-r-test-setup.R")
+
+
+test.adaBoost.smoke <- function() {
+    f <- "https://s3.amazonaws.com/h2o-public-test-data/smalldata/prostate/prostate.csv"
+    data <- h2o.importFile(f)
+
+    # Set predictors and response; set response as a factor
+    data["CAPSULE"] <- as.factor(data["CAPSULE"])
+    predictors <- c("AGE","RACE","DPROS","DCAPS","PSA","VOL","GLEASON")
+    response <- "CAPSULE"
+
+    h2o_adaboost <- h2o.adaBoost(nlearners = 5, x = predictors, y = response, training_frame = data, seed = 1234, weak_learner = "DRF", weak_learner_params = list(ntrees=3, max_depth=2, histogram_type="UniformAdaptive"))
+    expect_equal(is.null(h2o_adaboost), FALSE)
+    h2o_adaboost <- h2o.adaBoost(nlearners = 5, x = predictors, y = response, training_frame = data, seed = 1234, weak_learner = "GBM", weak_learner_params = list(ntrees=3, max_depth=2, histogram_type="UniformAdaptive"))
+    expect_equal(is.null(h2o_adaboost), FALSE)
+    h2o_adaboost <- h2o.adaBoost(nlearners = 5, x = predictors, y = response, training_frame = data, seed = 1234, weak_learner = "GLM", weak_learner_params = list(max_iterations=3))
+    expect_equal(is.null(h2o_adaboost), FALSE)
+    h2o_adaboost <- h2o.adaBoost(nlearners = 5, x = predictors, y = response, training_frame = data, seed = 1234, weak_learner = "DEEP_LEARNING", weak_learner_params = list(nepochs=3, hidden=list(2,1,2)))
+    expect_equal(is.null(h2o_adaboost), FALSE)
+}
+
+doTest("adaBoost: Smoke Test For Weak Learner Params - only that is pass through the API", test.adaBoost.smoke)