Merge remote-tracking branch 'origin/rel-3.46.0'

h2o-ops · h2o-ops · commit b28b04957f1b · 2024-05-24T14:56:19.000-07:00
diff --git a/h2o-docs/src/product/api-changes.rst b/h2o-docs/src/product/api-changes.rst
@@ -1,89 +1,93 @@
-API-Related Changes
--------------------
+API-related changes
+===================
 
 H2O-3 does its best to keep backwards compatibility between major versions, but sometimes breaking changes are needed in order to improve code quality and to address issues. This section provides a list of current breaking changes between specific releases.
 
 From 3.32.0.1
-~~~~~~~~~~~~~
+-------------
 
 Modules
-'''''''
+~~~~~~~
 
 The deprecated ``h2o-scala`` module has been removed.
 
 Target Encoding
-'''''''''''''''
+~~~~~~~~~~~~~~~
 
-The Target Encoder API has been clarified and its consistency across clients has been improved:
+The Target Encoder API has been clarified and its consistency across clients has been improved. The following parameters are now deprecated in all clients and officially replaced by their new alternative:
 
-The following parameters are now deprecated in all clients and officially replaced by their new alternative:
-- ``k`` -> ``inflection_point``
-- ``f`` -> ``smoothing``
-- ``noise_level`` -> ``noise``
-- ``use_blending`` (R only) -> ``blending``
+- ``k`` :math:`\to` ``inflection_point``
+- ``f`` :math:`\to` ``smoothing``
+- ``noise_level`` :math:`\to` ``noise``
+- ``use_blending`` (R only) :math:`\to` ``blending``
 
-Legacy client code using the deprecated parameters should expect a deprecation warning when using them, they are strongly encouraged to update their code to use the new naming.
+Legacy client code using the deprecated parameters should expect a deprecation warning when using them. You are strongly encouraged to update your code to use the new naming.
+
+``transform`` parameter updates
+'''''''''''''''''''''''''''''''
+
+In an objective of performance optimization on the backend, and of simplification of the API, the ``transform`` method used to apply target encoding was modified as follows:
 
-In an objective of performance optimization on the backend, and of simplification of the API, the ``transform`` method used to apply target encoding was modified as follow.
 - The R ``h2o.transform`` function (accepting a target encoder model as the first argument) and the Python ``H2OTargetEncoderEstimator.transform`` methods are now fully compatible: they accept the same parameters and work consistently.
-- The parameters ``data_leakage_handling``, ``seed`` are now ignored on those methods: ``transform`` will use by default the corresponding values defined when building the TargetEncoder model.
+- The parameters ``data_leakage_handling``, ``seed`` are now ignored on those methods: by default, ``transform`` will use the corresponding values defined when building the TargetEncoder model.
 - The other regularization parameters on these ``transform`` methods (e.g. ``noise``, ``blending``, ``inflection_point``, ``smoothing``), always default to the value defined on the TargetEncoder model.
 - A new ``as_training`` parameter has been introduced to simplify and enforce a correct usage of target encoding:
 
-  - When transforming a training dataset, user should use (R) ``h2o.transform(te_model, train_dataset, as_training=TRUE)`` or (Python) ``te_model.transform(train_dataset, as_training=True)``.
-  - When transforming any other dataset (validation, test, ...), user can just use (R) ``h2o.transform(te_model, train_dataset)`` or (Python) ``te_model.transform(train_dataset)``.
+  - When transforming a training dataset, you should use (R) ``h2o.transform(te_model, train_dataset, as_training=TRUE)`` or (Python) ``te_model.transform(train_dataset, as_training=True)``.
+  - When transforming any other dataset (validation, test, ...), you can just use (R) ``h2o.transform(te_model, train_dataset)`` or (Python) ``te_model.transform(train_dataset)``.
   - Legacy code using for example ``h2o.transform(te_model, train_dataset, data_leakage_handling="KFold")`` will now be translated internally to ``h2o.transform(te_model, train_dataset, as_training=TRUE)``.
 
 
-Finally the following APIs, deprecated since 3.28, have been fully removed:
+Finally the following APIs (deprecated since 3.28) have been fully removed:
+
 - Python: ``h2o.targetencoder`` module.
 - R: ``h2o.target_encode_fit`` and ``h2o.target_encode_transform`` functions.
 
 Parameters
-''''''''''
-The ``max_hit_ratio_k`` param has been removed.
+~~~~~~~~~~
+
+The ``max_hit_ratio_k`` parameter has been removed.
 
 From 3.30.1.2
-~~~~~~~~~~~~~
+-------------
 
-The ``max_hit_ratio_k`` param is deprecated in version 3.30.1.2 and will be completely removed in the next major version, 3.32.0.1.
+The ``max_hit_ratio_k`` parameter is deprecated in version 3.30.1.2 and will be completely removed in the next major version, 3.32.0.1.
 
 From 3.30.1.1
-~~~~~~~~~~~~~
+-------------
 
 The deprecated ``h2o-scala`` module has been removed.
 
 
 From 3.30.0.5
-~~~~~~~~~~~~~
+-------------
 
 The ``h2o-scala`` module is deprecated in version 3.30.0.5 and will be completely removed in the next major version, 3.30.1.1.
 
 
 From 3.30.0.4
-~~~~~~~~~~~~~
+-------------
 
 The following options are no longer supported by native `XGBoost <https://xgboost.readthedocs.io/en/latest/parameter.html>`__ and have been removed.
 
 - ``min_sum_hessian_in_leaf``
 - ``min_data_in_leaf``
 
 
-From 3.28 or Below to 3.30
-~~~~~~~~~~~~~~~~~~~~~~~~~~~
+From 3.28 or below to 3.30
+--------------------------
 
 Java API
-''''''''
+~~~~~~~~
 
-``hex.grid.HyperSpaceWalker`` and ``hex.grid.HyperspaceWalker.HyperSpaceIterator`` interfaces have been simplified.
-Users implementing those interfaces directly, for example to create a custom grid search exploration algorithm, may want to look at the default implementations in **h2o-core/src/main/java/hex/grid/HyperSpaceWalker.java** if they are facing any issue when compiling against the new interfaces.
+The ``hex.grid.HyperSpaceWalker`` and ``hex.grid.HyperspaceWalker.HyperSpaceIterator`` interfaces have been simplified. Users implementing those interfaces directly, for example to create a custom grid search exploration algorithm, may want to look at the default implementations in ``h2o-core/src/main/java/hex/grid/HyperSpaceWalker.java`` if they are facing any issue when compiling against the new interfaces.
 
 
-From 3.26 or Below to 3.28
-~~~~~~~~~~~~~~~~~~~~~~~~~~~
+From 3.26 or below to 3.28
+--------------------------
 
 Java API
-''''''''
+~~~~~~~~
 
 The following classes were moved:
 
@@ -96,11 +100,11 @@ The following classes were moved:
 =================================================   =========================================
 
 
-From 3.22 or Below to 3.24
-~~~~~~~~~~~~~~~~~~~~~~~~~~~
+From 3.22 or below to 3.24
+--------------------------
 
 Java API
-''''''''
+~~~~~~~~
 
 The following classes were moved and/or renamed:
 
@@ -112,4 +116,4 @@ The following classes were moved and/or renamed:
 ``ai.h2o.automl.AutoML.algo``                       ``ai.h2o.automl.Algo``
 =================================================   ======================================
 
-Some internal methods of ``StackedEnsemble`` and ``StackedEnsembleModel`` are no longer public, but this should not impact users.
+Some internal methods of ``StackedEnsemble`` and ``StackedEnsembleModel`` are no longer public, but this should not impact you.
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_equality_constraints_only_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_equality_constraints_only_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_constraints_binomial():
+def test_equality_constraints_only_binomial():
     '''
     This test checks and make sure the equality constraints work with binomial family.  Coefficients are initialized
     with glm coefficients built without contraints, default coefficients and random coefficients.
@@ -129,6 +129,6 @@ def test_constraints_binomial():
  
 
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_constraints_binomial)
+    pyunit_utils.standalone_test(test_equality_constraints_only_binomial)
 else:
-    test_constraints_binomial()
+    test_equality_constraints_only_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_equality_loose_lessthan_linear_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_equality_loose_lessthan_linear_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_constraints_binomial():
+def test_equality_linear_constraints_binomial():
     '''
     This test checks and make sure the equality constraints work with binomial family.  Coefficients are initialized
     with glm coefficients built without constraints, default coefficients and random coefficients.  Note in this case,
@@ -168,6 +168,6 @@ def test_constraints_binomial():
 
 
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_constraints_binomial)
+    pyunit_utils.standalone_test(test_equality_linear_constraints_binomial)
 else:
-    test_constraints_binomial()
+    test_equality_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_light_tight_beta_equality_lessthan_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_light_tight_beta_equality_lessthan_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_light_tight_linear_constraints_only_gaussian():
+def test_light_tight_linear_constraints_binomial():
     '''
     Test constrained GLM with beta, equality and less than and equal to constraints.  The constraints are not very 
     tight.  However, coefficients from GLM built without constraints won't be able to satisfied the constraints.
@@ -207,6 +207,6 @@ def test_light_tight_linear_constraints_only_gaussian():
                                      "not.".format(logloss, random_init_logloss)
     
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_gaussian)
+    pyunit_utils.standalone_test(test_light_tight_linear_constraints_binomial)
 else:
-    test_light_tight_linear_constraints_only_gaussian()
+    test_light_tight_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_light_tight_equality_lessthan_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_light_tight_equality_lessthan_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_light_tight_linear_constraints_only_gaussian():
+def test_light_tight_linear_constraints_only_binomial():
     '''
     Test constrained GLM with equality and less than and equal to constraints.  The constraints are not very 
     tight.  However, coefficients from GLM built without constraints won't be able to satisfied the constraints.
@@ -185,6 +185,6 @@ def test_light_tight_linear_constraints_only_gaussian():
                                      "not.".format(logloss, random_init_logloss)
     
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_gaussian)
+    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_binomial)
 else:
-    test_light_tight_linear_constraints_only_gaussian()
+    test_light_tight_linear_constraints_only_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_light_tight_linear_constraints_only_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_light_tight_linear_constraints_only_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_light_tight_linear_constraints_only_gaussian():
+def test_light_tight_linear_constraints_only_binomial():
     '''
     Test constrained GLM with less than and equal to constraints.  The constraints are not very 
     tight.  However, coefficients from GLM built without constraints won't be able to satisfied the constraints.
@@ -201,6 +201,6 @@ def test_light_tight_linear_constraints_only_gaussian():
                                      "not.".format(logloss, random_init_logloss)
     
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_gaussian)
+    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_binomial)
 else:
-    test_light_tight_linear_constraints_only_gaussian()
+    test_light_tight_linear_constraints_only_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_loose_beta_linear_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_loose_beta_linear_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_constraints_binomial():
+def test_loose_beta_linear_constraints_binomial():
     '''
     check and make sure coefficients close to GLM built without constraints are generated with loose constraints
     that are satisfied with coefficients from GLM without constraints.  Only beta and less than and equal to
@@ -157,6 +157,6 @@ def test_constraints_binomial():
 
 
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_constraints_binomial)
+    pyunit_utils.standalone_test(test_loose_beta_linear_constraints_binomial)
 else:
-    test_constraints_binomial()
+    test_loose_beta_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_loose_only_linear_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_loose_only_linear_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_constraints_binomial():
+def test_loose_linear_constraints_binomial():
     '''
     check and make sure coefficients close to GLM built without constraints are generated with loose constraints
     that are satisfied with coefficients from GLM without constraints.  Only less than and equal to
@@ -135,6 +135,6 @@ def test_constraints_binomial():
                                                        " but is not.".format(logloss, default_init_logloss)
 
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_constraints_binomial)
+    pyunit_utils.standalone_test(test_loose_linear_constraints_binomial)
 else:
-    test_constraints_binomial()
+    test_loose_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_tight_beta_equality_linear_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_tight_beta_equality_linear_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_light_tight_linear_constraints_only_gaussian():
+def test_tight_beta_linear_constraints_binomial():
     '''
     Test constrained GLM with beta, equality and less than and equal to constraints.  The constraints are very 
     tight and coefficients from GLM built without constraints won't be able to satisfied the constraints.
@@ -174,42 +174,6 @@ def test_light_tight_linear_constraints_only_gaussian():
     types = "Equal"
     contraint_numbers = 5
     tight_constraints.append([name, values, types, contraint_numbers])
-
-    name = "C19"
-    values = 0.5
-    types = "Equal"
-    contraint_numbers = 4
-    tight_constraints.append([name, values, types, contraint_numbers])
-
-    name = "C10.1"
-    values = -0.3
-    types = "Equal"
-    contraint_numbers = 4
-    tight_constraints.append([name, values, types, contraint_numbers])
-
-    name = "constant"
-    values = -0.5
-    types = "Equal"
-    contraint_numbers = 4
-    tight_constraints.append([name, values, types, contraint_numbers])
-
-    name = "C18"
-    values = 0.75
-    types = "Equal"
-    contraint_numbers = 5
-    tight_constraints.append([name, values, types, contraint_numbers])
-
-    name = "C20"
-    values = -0.13
-    types = "Equal"
-    contraint_numbers = 5
-    tight_constraints.append([name, values, types, contraint_numbers])
-
-    name = "constant"
-    values = -3
-    types = "Equal"
-    contraint_numbers = 5
-    tight_constraints.append([name, values, types, contraint_numbers])
     
     linear_constraints2 = h2o.H2OFrame(tight_constraints)
     linear_constraints2.set_names(["names", "values", "types", "constraint_numbers"])
@@ -317,6 +281,6 @@ def test_light_tight_linear_constraints_only_gaussian():
                                      "not.".format(logloss, random_init_logloss)
     
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_gaussian)
+    pyunit_utils.standalone_test(test_tight_beta_linear_constraints_binomial)
 else:
-    test_light_tight_linear_constraints_only_gaussian()
+    test_tight_beta_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_tight_equality_linear_constraints_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_tight_equality_linear_constraints_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_light_tight_linear_constraints_only_gaussian():
+def test_tight_equality_linear_constraints_binomial():
     '''
     Test constrained GLM with equality and less than and equal to constraints.  The constraints are very 
     tight and coefficients from GLM built without constraints won't be able to satisfied the constraints.
@@ -237,6 +237,6 @@ def test_light_tight_linear_constraints_only_gaussian():
                                      "not.".format(logloss, random_init_logloss)
     
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_gaussian)
+    pyunit_utils.standalone_test(test_tight_equality_linear_constraints_binomial)
 else:
-    test_light_tight_linear_constraints_only_gaussian()
+    test_tight_equality_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_tight_linear_constraints_only_binomial.py b/h2o-py/tests/testdir_algos/glm/pyunit_GH_6722_tight_linear_constraints_only_binomial.py
@@ -3,7 +3,7 @@
 from tests import pyunit_utils
 from tests.pyunit_utils import utils_for_glm_tests
 
-def test_light_tight_linear_constraints_only_gaussian():
+def test_tight_linear_constraints_binomial():
     '''
     Test constrained GLM with less than and equal to constraints.  The constraints are very 
     tight and coefficients from GLM built without constraints won't be able to satisfied the constraints.
@@ -201,6 +201,6 @@ def test_light_tight_linear_constraints_only_gaussian():
                                      "not.".format(logloss, random_init_logloss)
     
 if __name__ == "__main__":
-    pyunit_utils.standalone_test(test_light_tight_linear_constraints_only_gaussian)
+    pyunit_utils.standalone_test(test_tight_linear_constraints_binomial)
 else:
-    test_light_tight_linear_constraints_only_gaussian()
+    test_tight_linear_constraints_binomial()
diff --git a/h2o-py/tests/testdir_algos/glm/pyunit_gh_16203_constrained_glm_example.py b/h2o-py/tests/testdir_algos/glm/pyunit_gh_16203_constrained_glm_example.py
