Rework xgboost jvm packages for spark

- Remove xgboost4j-gpu and move its functionality to xgboost4j-spark-gpu - Remove any soft links in xgboost4j-spark-gpu - Abstract an XGBoost Estimator which handles the common functionality for all xgboost estimators. - Support XGBoostRanker - Remove any unnecessary ETL in XGBoost - Fix the missing value usage - Support uber jar for xgboost4j-spark - Support uber jar for xgboost4j-spark-gpu - Rework GPU plugin - More sannity test to ensure the training/transform results are same between CPU and GPU.
dmlc · Jul 22, 2024 · 82c06da · 82c06da
1 parent 6d9fcb7
commit 82c06da
Show file tree

Hide file tree

Showing 122 changed files with 6,128 additions and 8,743 deletions.
diff --git a/dev/change_scala_version.py b/dev/change_scala_version.py
@@ -62,6 +62,17 @@ def main(args):
                     )
                     if nsubs > 0:
                         replaced_scala_binver = True
+                # Replace the final name of shaded jar
+                if "<finalName>" in line:
+                    for artifact in [
+                        "xgboost-spark",
+                        "xgboost-spark-gpu",
+                    ]:
+                        line = re.sub(
+                            f"<finalName>{artifact}_[0-9\\.]*",
+                            f"<finalName>{artifact}_{scala_ver}",
+                            line,
+                        )
                 f.write(line)
 
 

diff --git a/include/xgboost/c_api.h b/include/xgboost/c_api.h
@@ -402,13 +402,15 @@ XGB_EXTERN_C typedef int XGBCallbackDataIterNext(  // NOLINT(*)
  * \param data_handle The handle to the data.
  * \param callback The callback to get the data.
  * \param cache_info Additional information about cache file, can be null.
+ * \param missing Which value to represent missing value.
  * \param out The created DMatrix
  * \return 0 when success, -1 when failure happens.
  */
 XGB_DLL int XGDMatrixCreateFromDataIter(
     DataIterHandle data_handle,
     XGBCallbackDataIterNext* callback,
     const char* cache_info,
+    float missing,
     DMatrixHandle *out);
 
 /**

diff --git a/jvm-packages/.gitignore b/jvm-packages/.gitignore
@@ -1,4 +1,5 @@
-tracker.py
 build.sh
 xgboost4j-tester/pom.xml
 xgboost4j-tester/iris.csv
+dependency-reduced-pom.xml
+surefire-reports
diff --git a/jvm-packages/CMakeLists.txt b/jvm-packages/CMakeLists.txt
@@ -2,11 +2,11 @@ find_package(JNI REQUIRED)
 
 list(APPEND JVM_SOURCES
   ${PROJECT_SOURCE_DIR}/jvm-packages/xgboost4j/src/native/xgboost4j.cpp
-  ${PROJECT_SOURCE_DIR}/jvm-packages/xgboost4j-gpu/src/native/xgboost4j-gpu.cpp)
+  ${PROJECT_SOURCE_DIR}/jvm-packages/xgboost4j/src/native/xgboost4j-gpu.cpp)
 
 if(USE_CUDA)
   list(APPEND JVM_SOURCES
-    ${PROJECT_SOURCE_DIR}/jvm-packages/xgboost4j-gpu/src/native/xgboost4j-gpu.cu)
+    ${PROJECT_SOURCE_DIR}/jvm-packages/xgboost4j/src/native/xgboost4j-gpu.cu)
 endif()
 
 add_library(xgboost4j SHARED ${JVM_SOURCES} ${XGBOOST_OBJ_SOURCES})

diff --git a/jvm-packages/aggregator-gpu/pom.xml b/jvm-packages/aggregator-gpu/pom.xml
@@ -0,0 +1,67 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project xmlns="http://maven.apache.org/POM/4.0.0"
+         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
+  <modelVersion>4.0.0</modelVersion>
+  <parent>
+    <groupId>ml.dmlc</groupId>
+    <artifactId>xgboost-jvm_2.12</artifactId>
+    <version>2.2.0-SNAPSHOT</version>
+  </parent>
+
+  <name>xgboost-spark-gpu</name>
+  <artifactId>xgboost-spark-gpu_2.12</artifactId>
+
+  <dependencies>
+    <dependency>
+      <groupId>ml.dmlc</groupId>
+      <artifactId>xgboost4j_2.12</artifactId>
+      <version>${project.version}</version>
+      <scope>compile</scope>
+    </dependency>
+
+    <dependency>
+      <groupId>ml.dmlc</groupId>
+      <artifactId>xgboost4j-spark_2.12</artifactId>
+      <version>${project.version}</version>
+      <scope>compile</scope>
+    </dependency>
+
+    <dependency>
+      <groupId>ml.dmlc</groupId>
+      <artifactId>xgboost4j-spark-gpu_2.12</artifactId>
+      <version>${project.version}</version>
+      <scope>compile</scope>
+    </dependency>
+  </dependencies>
+
+  <build>
+    <finalName>xgboost-spark-gpu_2.12-${project.version}</finalName>
+    <plugins>
+      <plugin>
+        <groupId>org.apache.maven.plugins</groupId>
+        <artifactId>maven-shade-plugin</artifactId>
+        <configuration>
+          <shadedArtifactAttached>false</shadedArtifactAttached>
+          <createDependencyReducedPom>true</createDependencyReducedPom>
+          <artifactSet>
+            <includes>
+              <include>ml.dmlc:xgboost4j_2.12</include>
+              <include>ml.dmlc:xgboost4j-spark_2.12</include>
+              <include>ml.dmlc:xgboost4j-spark-gpu_2.12</include>
+            </includes>
+          </artifactSet>
+        </configuration>
+        <executions>
+          <execution>
+            <phase>package</phase>
+            <goals>
+              <goal>shade</goal>
+            </goals>
+          </execution>
+        </executions>
+      </plugin>
+    </plugins>
+  </build>
+
+</project>
diff --git a/jvm-packages/aggregator/pom.xml b/jvm-packages/aggregator/pom.xml
@@ -0,0 +1,59 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project xmlns="http://maven.apache.org/POM/4.0.0"
+         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
+  <modelVersion>4.0.0</modelVersion>
+  <parent>
+    <groupId>ml.dmlc</groupId>
+    <artifactId>xgboost-jvm_2.12</artifactId>
+    <version>2.2.0-SNAPSHOT</version>
+  </parent>
+
+  <name>xgboost-spark</name>
+  <artifactId>xgboost-spark_2.12</artifactId>
+
+  <dependencies>
+    <dependency>
+      <groupId>ml.dmlc</groupId>
+      <artifactId>xgboost4j_2.12</artifactId>
+      <version>${project.version}</version>
+      <scope>compile</scope>
+    </dependency>
+
+    <dependency>
+      <groupId>ml.dmlc</groupId>
+      <artifactId>xgboost4j-spark_2.12</artifactId>
+      <version>${project.version}</version>
+      <scope>compile</scope>
+    </dependency>
+  </dependencies>
+
+  <build>
+    <finalName>xgboost-spark_2.12-${project.version}</finalName>
+    <plugins>
+      <plugin>
+        <groupId>org.apache.maven.plugins</groupId>
+        <artifactId>maven-shade-plugin</artifactId>
+        <configuration>
+          <shadedArtifactAttached>false</shadedArtifactAttached>
+          <createDependencyReducedPom>true</createDependencyReducedPom>
+          <artifactSet>
+            <includes>
+              <include>ml.dmlc:xgboost4j_2.12</include>
+              <include>ml.dmlc:xgboost4j-spark_2.12</include>
+            </includes>
+          </artifactSet>
+        </configuration>
+        <executions>
+          <execution>
+            <phase>package</phase>
+            <goals>
+              <goal>shade</goal>
+            </goals>
+          </execution>
+        </executions>
+      </plugin>
+    </plugins>
+  </build>
+
+</project>
diff --git a/jvm-packages/checkstyle.xml b/jvm-packages/checkstyle.xml
@@ -133,7 +133,7 @@
             <property name="braceAdjustment" value="0"/>
             <property name="caseIndent" value="2"/>
             <property name="throwsIndent" value="4"/>
-            <property name="lineWrappingIndentation" value="4"/>
+            <property name="lineWrappingIndentation" value="2"/>
             <property name="arrayInitIndent" value="2"/>
         </module>
         <module name="ImportOrder">

diff --git a/jvm-packages/create_jni.py b/jvm-packages/create_jni.py
@@ -131,14 +131,6 @@ def native_build(args):
                 run("cmake .. " + " ".join(args))
             run("cmake --build . --config Release" + maybe_parallel_build)
 
-        with cd("demo/CLI/regression"):
-            run(f'"{sys.executable}" mapfeat.py')
-            run(f'"{sys.executable}" mknfold.py machine.txt 1')
-
-    xgboost4j = "xgboost4j-gpu" if cli_args.use_cuda == "ON" else "xgboost4j"
-    xgboost4j_spark = (
-        "xgboost4j-spark-gpu" if cli_args.use_cuda == "ON" else "xgboost4j-spark"
-    )
 
     print("copying native library")
     library_name, os_folder = {
@@ -155,26 +147,34 @@ def native_build(args):
         "arm64": "aarch64",  # on macOS & Windows ARM 64-bit
         "aarch64": "aarch64",
     }[platform.machine().lower()]
-    output_folder = "{}/src/main/resources/lib/{}/{}".format(
-        xgboost4j, os_folder, arch_folder
+    output_folder = "xgboost4j/src/main/resources/lib/{}/{}".format(
+        os_folder, arch_folder
     )
     maybe_makedirs(output_folder)
     cp("../lib/" + library_name, output_folder)
 
     print("copying train/test files")
-    maybe_makedirs("{}/src/test/resources".format(xgboost4j_spark))
+
+    # for xgboost4j
+    maybe_makedirs("xgboost4j/src/test/resources")
+    for file in glob.glob("../demo/data/agaricus.*"):
+        cp(file, "xgboost4j/src/test/resources")
+
+    # for xgboost4j-spark
+    maybe_makedirs("xgboost4j-spark/src/test/resources")
     with cd("../demo/CLI/regression"):
         run(f'"{sys.executable}" mapfeat.py')
         run(f'"{sys.executable}" mknfold.py machine.txt 1')
-
     for file in glob.glob("../demo/CLI/regression/machine.txt.t*"):
-        cp(file, "{}/src/test/resources".format(xgboost4j_spark))
+        cp(file, "xgboost4j-spark/src/test/resources")
     for file in glob.glob("../demo/data/agaricus.*"):
-        cp(file, "{}/src/test/resources".format(xgboost4j_spark))
+        cp(file, "xgboost4j-spark/src/test/resources")
 
-    maybe_makedirs("{}/src/test/resources".format(xgboost4j))
-    for file in glob.glob("../demo/data/agaricus.*"):
-        cp(file, "{}/src/test/resources".format(xgboost4j))
+    # for xgboost4j-spark-gpu
+    if cli_args.use_cuda == "ON":
+        maybe_makedirs("xgboost4j-spark-gpu/src/test/resources")
+        for file in glob.glob("../demo/data/veterans_lung_cancer.csv"):
+            cp(file, "xgboost4j-spark-gpu/src/test/resources")
 
 
 if __name__ == "__main__":