Release patch release 3.9.1 (#520)

* Release patch release 3.9.1

Apply patch: Fix loading from XGBoost 2.0 dev (#509)

* Removed deprecated parameter size_leaf_vector

* Handle iteration_indptr field in JSON

* Additional fix

* More fixes

* [CI] Use XGBoost 2.0

* Formatting check

.readthedocs.yaml

@@ -28,4 +28,3 @@ sphinx:
 python:
   install:
    - requirements: docs/requirements.txt
-  system_packages: true

CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_policy(SET CMP0091 NEW)
 set(CMAKE_FIND_NO_INSTALL_PREFIX TRUE FORCE)
 cmake_minimum_required (VERSION 3.16)
-project(treelite LANGUAGES CXX C VERSION 3.9.0)
+project(treelite LANGUAGES CXX C VERSION 3.9.1)
 
 # check MSVC version
 if(MSVC)

ops/conda_env/dev.yml

@@ -18,10 +18,10 @@ dependencies:
 - llvm-openmp
 - cython
 - lightgbm
-- xgboost
 - cpplint
 - pylint
 - awscli
 - pip
 - pip:
   - cibuildwheel
+  - xgboost>=2.0

ops/cpp-python-coverage.sh

@@ -9,15 +9,15 @@ echo "##[section]Building Treelite..."
 mkdir build/
 cd build/
 cmake .. -DTEST_COVERAGE=ON -DCMAKE_BUILD_TYPE=Debug -DBUILD_CPP_TEST=ON -GNinja
-ninja
+ninja install -v
 cd ..
 
 echo "##[section]Running Google C++ tests..."
 ./build/treelite_cpp_test
 
 echo "##[section]Build Cython extension..."
 cd tests/cython
-python setup.py build_ext --inplace
+pip install -vvv .
 cd ../..
 
 echo "##[section]Running Python integration tests..."

python/treelite/VERSION

@@ -1 +1 @@
-3.9.0
+3.9.1

runtime/java/treelite4j/pom.xml

@@ -5,7 +5,7 @@
     <modelVersion>4.0.0</modelVersion>
     <groupId>ml.dmlc</groupId>
     <artifactId>treelite4j</artifactId>
-    <version>3.9.0</version>
+    <version>3.9.1</version>
     <packaging>jar</packaging>
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>

runtime/python/treelite_runtime/VERSION

@@ -1 +1 @@
-3.9.0
+3.9.1

src/frontend/xgboost.cc

@@ -296,7 +296,7 @@ class XGBTree {
     nodes[nodes[nid].cleft() ].set_parent(nid, true);
     nodes[nodes[nid].cright()].set_parent(nid, false);
   }
-  inline void Load(PeekableInputStream* fi) {
+  inline void Load(PeekableInputStream* fi, LearnerModelParam const& mparam) {
     TREELITE_CHECK_EQ(fi->Read(&param, sizeof(TreeParam)), sizeof(TreeParam))
      << "Ill-formed XGBoost model file: can't read TreeParam";
     TREELITE_CHECK_GT(param.num_nodes, 0)
@@ -309,13 +309,17 @@ class XGBTree {
     TREELITE_CHECK_EQ(fi->Read(stats.data(), sizeof(NodeStat) * stats.size()),
              sizeof(NodeStat) * stats.size())
      << "Ill-formed XGBoost model file: cannot read specified number of nodes";
-    if (param.size_leaf_vector != 0) {
+    if (param.size_leaf_vector != 0 && mparam.major_version < 2) {
       uint64_t len;
       TREELITE_CHECK_EQ(fi->Read(&len, sizeof(len)), sizeof(len))
        << "Ill-formed XGBoost model file";
       if (len > 0) {
         CONSUME_BYTES(fi, sizeof(bst_float) * len);
       }
+    } else if (mparam.major_version == 2) {
+      TREELITE_CHECK_EQ(param.size_leaf_vector, 1)
+          << "Multi-target models are not supported with binary serialization. "
+          << "Please save the XGBoost model using the JSON format.";
     }
     TREELITE_CHECK_EQ(param.num_roots, 1)
       << "Invalid XGBoost model file: treelite does not support trees "
@@ -378,7 +382,7 @@ inline std::unique_ptr<treelite::Model> ParseStream(std::istream& fi) {
     << "Invalid XGBoost model file: num_trees must be 0 or greater";
   for (int i = 0; i < gbm_param_.num_trees; ++i) {
     xgb_trees_.emplace_back();
-    xgb_trees_.back().Load(fp.get());
+    xgb_trees_.back().Load(fp.get(), mparam_);
   }
   if (mparam_.major_version < 1 || (mparam_.major_version == 1 && mparam_.minor_version < 6)) {
     // In XGBoost 1.6, num_roots is used as num_parallel_tree, so don't check

src/frontend/xgboost_json.cc

@@ -366,7 +366,8 @@ bool GBTreeModelHandler::StartArray() {
   return (push_key_handler<ArrayHandler<treelite::Tree<float, float>, RegTreeHandler>,
                            std::vector<treelite::Tree<float, float>>>(
                                "trees", output.model->trees) ||
-          push_key_handler<ArrayHandler<int>, std::vector<int>>("tree_info", output.tree_info));
+          push_key_handler<ArrayHandler<int>, std::vector<int>>("tree_info", output.tree_info) ||
+          push_key_handler<IgnoreHandler>("iteration_indptr"));
 }
 
 bool GBTreeModelHandler::StartObject() {
@@ -377,7 +378,8 @@ bool GBTreeModelHandler::StartObject() {
 }
 
 bool GBTreeModelHandler::is_recognized_key(const std::string& key) {
-  return (key == "trees" || key == "tree_info" || key == "gbtree_model_param");
+  return (key == "trees" || key == "tree_info" || key == "gbtree_model_param"
+          || key == "iteration_indptr");
 }
 
 /******************************************************************************
@@ -460,7 +462,8 @@ bool ObjectiveHandler::StartObject() {
           push_key_handler<IgnoreHandler>("lambda_rank_param") ||
           push_key_handler<IgnoreHandler>("aft_loss_param") ||
           push_key_handler<IgnoreHandler>("pseduo_huber_param") ||
-          push_key_handler<IgnoreHandler>("pseudo_huber_param"));
+          push_key_handler<IgnoreHandler>("pseudo_huber_param") ||
+          push_key_handler<IgnoreHandler>("lambdarank_param"));
 }
 
 bool ObjectiveHandler::String(const char *str, std::size_t length, bool) {
@@ -474,7 +477,8 @@ bool ObjectiveHandler::is_recognized_key(const std::string& key) {
   return (key == "reg_loss_param" || key == "poisson_regression_param"
           || key == "tweedie_regression_param" || key == "softmax_multiclass_param"
           || key == "lambda_rank_param" || key == "aft_loss_param"
-          || key == "pseduo_huber_param" || key == "pseudo_huber_param" || key == "name");
+          || key == "pseduo_huber_param" || key == "pseudo_huber_param"
+          || key == "lambdarank_param" || key == "name");
 }
 
 /******************************************************************************

tests/cython/CMakeLists.txt

@@ -0,0 +1,40 @@
+cmake_minimum_required(VERSION 3.18)
+
+project(treelite_serializer_ext LANGUAGES CXX)
+
+find_package(
+  Python
+  COMPONENTS Interpreter Development.Module
+  REQUIRED)
+
+find_program(CYTHON "cython")
+
+find_package(Treelite REQUIRED)
+
+add_custom_command(
+  OUTPUT serializer.cpp
+  DEPENDS serializer.pyx
+  VERBATIM
+  COMMAND "${CYTHON}" "${PROJECT_SOURCE_DIR}/serializer.pyx" --output-file
+          "${PROJECT_BINARY_DIR}/serializer.cpp")
+
+if(DEFINED ENV{CONDA_PREFIX})
+  set(CMAKE_PREFIX_PATH "$ENV{CONDA_PREFIX};${CMAKE_PREFIX_PATH}")
+  message(STATUS "Detected Conda environment, CMAKE_PREFIX_PATH set to: ${CMAKE_PREFIX_PATH}")
+  if(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
+    message(STATUS "No CMAKE_INSTALL_PREFIX argument detected, setting to: $ENV{CONDA_PREFIX}")
+    set(CMAKE_INSTALL_PREFIX $ENV{CONDA_PREFIX})
+  endif()
+else()
+  message(STATUS "No Conda environment detected")
+endif()
+
+python_add_library(serializer MODULE "${PROJECT_BINARY_DIR}/serializer.cpp" WITH_SOABI)
+target_link_libraries(serializer PRIVATE treelite::treelite)
+set_target_properties(serializer
+  PROPERTIES
+  POSITION_INDEPENDENT_CODE ON
+  CXX_STANDARD 17
+  CXX_STANDARD_REQUIRED ON)
+
+install(TARGETS serializer DESTINATION "${PROJECT_SOURCE_DIR}")

tests/cython/pyproject.toml

@@ -0,0 +1,7 @@
+[build-system]
+requires = ["scikit-build-core", "cython"]
+build-backend = "scikit_build_core.build"
+
+[project]
+name = "example"
+version = "0.0.1"

tests/python/test_gtil.py

@@ -8,7 +8,7 @@
 import numpy as np
 import pytest
 import scipy
-from hypothesis import assume, given, settings
+from hypothesis import given, settings
 from hypothesis.strategies import data as hypothesis_callback
 from hypothesis.strategies import floats, integers, just, sampled_from
 from sklearn.datasets import load_svmlight_file
@@ -198,16 +198,15 @@ def test_skl_hist_gradient_boosting_with_categorical():
         treelite.sklearn.import_model(clf)
 
 
+@pytest.mark.parametrize("objective",
+    [
+        "reg:linear",
+        "reg:squarederror",
+        "reg:squaredlogerror",
+        "reg:pseudohubererror",
+    ])
 @given(
     dataset=standard_regression_datasets(),
-    objective=sampled_from(
-        [
-            "reg:linear",
-            "reg:squarederror",
-            "reg:squaredlogerror",
-            "reg:pseudohubererror",
-        ]
-    ),
     model_format=sampled_from(["binary", "json"]),
     num_boost_round=integers(min_value=5, max_value=50),
     num_parallel_tree=integers(min_value=1, max_value=5),
@@ -218,9 +217,14 @@ def test_xgb_regression(
 ):
     # pylint: disable=too-many-locals
     """Test XGBoost with regression data"""
+
+    # See https://github.com/dmlc/xgboost/pull/9574
+    if objective == "reg:pseudohubererror":
+        pytest.xfail("XGBoost 2.0 has a bug in the serialization of Pseudo-Huber error")
+
     X, y = dataset
     if objective == "reg:squaredlogerror":
-        assume(np.all(y > -1))
+        y = np.where(y <= -1, -0.9, y)
     X_train, X_test, y_train, y_test = train_test_split(
         X, y, test_size=0.2, shuffle=False
     )
@@ -330,7 +334,7 @@ def test_xgb_multiclass_classifier(
             ("count:poisson", 4),
             ("rank:pairwise", 5),
             ("rank:ndcg", 5),
-            ("rank:map", 5),
+            ("rank:map", 2),
         ],
     ),
     model_format=sampled_from(["binary", "json"]),

tests/python/test_xgboost_integration.py

@@ -7,7 +7,7 @@
 
 import numpy as np
 import pytest
-from hypothesis import assume, given, settings
+from hypothesis import given, settings
 from hypothesis.strategies import integers, lists, sampled_from
 from sklearn.datasets import load_iris
 from sklearn.model_selection import train_test_split
@@ -27,16 +27,15 @@
     pytest.skip("XGBoost not installed; skipping", allow_module_level=True)
 
 
+@pytest.mark.parametrize("objective",
+    [
+        "reg:linear",
+        "reg:squarederror",
+        "reg:squaredlogerror",
+        "reg:pseudohubererror",
+    ])
 @given(
     toolchain=sampled_from(os_compatible_toolchains()),
-    objective=sampled_from(
-        [
-            "reg:linear",
-            "reg:squarederror",
-            "reg:squaredlogerror",
-            "reg:pseudohubererror",
-        ]
-    ),
     model_format=sampled_from(["binary", "json"]),
     num_parallel_tree=integers(min_value=1, max_value=10),
     dataset=standard_regression_datasets(),
@@ -45,9 +44,14 @@
 def test_xgb_regression(toolchain, objective, model_format, num_parallel_tree, dataset):
     # pylint: disable=too-many-locals
     """Test a random regression dataset"""
+
+    # See https://github.com/dmlc/xgboost/pull/9574
+    if objective == "reg:pseudohubererror":
+        pytest.xfail("XGBoost 2.0 has a bug in the serialization of Pseudo-Huber error")
+
     X, y = dataset
     if objective == "reg:squaredlogerror":
-        assume(np.all(y > -1))
+        y = np.where(y <= -1, -0.9, y)
     X_train, X_test, y_train, y_test = train_test_split(
         X, y, test_size=0.2, shuffle=False
     )
@@ -59,6 +63,7 @@ def test_xgb_regression(toolchain, objective, model_format, num_parallel_tree, d
         "verbosity": 0,
         "objective": objective,
         "num_parallel_tree": num_parallel_tree,
+        "base_score": 0.0
     }
     num_round = 10
     bst = xgb.train(
@@ -96,7 +101,7 @@ def test_xgb_regression(toolchain, objective, model_format, num_parallel_tree, d
         assert predictor.num_feature == dtrain.num_col()
         assert predictor.num_class == 1
         assert predictor.pred_transform == "identity"
-        assert predictor.global_bias == 0.5
+        assert predictor.global_bias == 0.0
         assert predictor.sigmoid_alpha == 1.0
         dmat = treelite_runtime.DMatrix(X_test, dtype="float32")
         out_pred = predictor.predict(dmat)
@@ -184,7 +189,7 @@ def test_xgb_iris(
         ("count:poisson", 4, math.log(0.5)),
         ("rank:pairwise", 5, 0.5),
         ("rank:ndcg", 5, 0.5),
-        ("rank:map", 5, 0.5),
+        ("rank:map", 2, 0.5),
     ],
     ids=[
         "binary:logistic",
@@ -276,7 +281,8 @@ def test_xgb_deserializers(toolchain, dataset):
     )
     dtrain = xgb.DMatrix(X_train, label=y_train)
     dtest = xgb.DMatrix(X_test, label=y_test)
-    param = {"max_depth": 8, "eta": 1, "silent": 1, "objective": "reg:linear"}
+    param = {"max_depth": 8, "eta": 1, "silent": 1, "objective": "reg:linear",
+             "base_score": 0.5}
     num_round = 10
     bst = xgb.train(
         param,
@@ -417,7 +423,6 @@ def test_xgb_dart(tmpdir, toolchain, model_format):
     assert predictor.num_feature == dtrain.num_col()
     assert predictor.num_class == 1
     assert predictor.pred_transform == "sigmoid"
-    np.testing.assert_almost_equal(predictor.global_bias, 0, decimal=5)
     assert predictor.sigmoid_alpha == 1.0
     dmat = treelite_runtime.DMatrix(X, dtype="float32")
     out_pred = predictor.predict(dmat)