implement partitioned

python/sedona/utils/adapter.py

@@ -210,3 +210,20 @@ def toDf(
             return Adapter.toDf(srdd, fieldNames, spark)
         else:
             return Adapter.toDf(srdd, spark)
+
+    @classmethod
+    def toDfPartitioned(cls, spatialRDD: SpatialRDD, sparkSession: SparkSession) -> DataFrame:
+        """
+
+        :param spatialRDD:
+        :param sparkSession:
+        :return:
+        """
+        sc = spatialRDD._sc
+        jvm = sc._jvm
+
+        jdf = jvm.Adapter.toDfPartitioned(spatialRDD._srdd, sparkSession._jsparkSession)
+
+        df = Adapter._create_dataframe(jdf, sparkSession)
+
+        return df

spark/common/src/main/scala/org/apache/sedona/sql/utils/Adapter.scala

@@ -162,14 +162,10 @@ object Adapter {
       spatialRDD: SpatialRDD[T],
       schema: StructType,
       sparkSession: SparkSession): DataFrame = {
-    val rdd = spatialRDD.rawSpatialRDD.rdd.mapPartitions(
-      iter => {
-        iter.map[Row](geom => {
-          val stringRow = extractUserData(geom)
-          castRowToSchema(stringRow = stringRow, schema = schema)
-        })
-      },
-      true)
+    val rdd = spatialRDD.rawSpatialRDD.rdd.map[Row](geom => {
+      val stringRow = extractUserData(geom)
+      castRowToSchema(stringRow = stringRow, schema = schema)
+    })
 
     sparkSession.sqlContext.createDataFrame(rdd, schema)
   }
@@ -239,6 +235,46 @@ object Adapter {
     sparkSession.sqlContext.createDataFrame(rdd, schema)
   }
 
+  /**
+   * Convert a spatial RDD to DataFrame with a given schema keeping spatial partitioning
+   *
+   * Note that spatial partitioning methods that introduce duplicates will result in an output
+   * data frame with duplicate features. This property is essential for implementing correct
+   * joins; however, may introduce surprising results.
+   *
+   * @param spatialRDD
+   *   Spatial RDD
+   * @param fieldNames
+   *   Desired field names
+   * @param sparkSession
+   *   Spark Session
+   * @tparam T
+   *   Geometry
+   * @return
+   *   DataFrame with the specified field names with spatial partitioning preserved
+   */
+  def toDfPartitioned[T <: Geometry](
+      spatialRDD: SpatialRDD[T],
+      fieldNames: Seq[String],
+      sparkSession: SparkSession): DataFrame = {
+    val rowRdd = spatialRDD.spatialPartitionedRDD.map[Row](geom => {
+      val stringRow = extractUserData(geom)
+      Row.fromSeq(stringRow)
+    })
+    var cols: Seq[StructField] = Seq(StructField("geometry", GeometryUDT))
+    if (fieldNames != null && fieldNames.nonEmpty) {
+      cols = cols ++ fieldNames.map(f => StructField(f, StringType))
+    }
+    val schema = StructType(cols)
+    sparkSession.createDataFrame(rowRdd, schema)
+  }
+
+  def toDfPartitioned[T <: Geometry](
+      spatialRDD: SpatialRDD[T],
+      sparkSession: SparkSession): DataFrame = {
+    toDfPartitioned(spatialRDD, null, sparkSession)
+  }
+
   /**
    * Extract user data from a geometry.
    *