rahil-c commented on code in PR #18432: URL: https://github.com/apache/hudi/pull/18432#discussion_r3024393540
########## hudi-spark-datasource/hudi-spark/src/test/scala/org/apache/hudi/functional/TestHoodieVectorSearchFunction.scala: ########## @@ -0,0 +1,1481 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.hudi.functional + +import org.apache.hudi.DataSourceWriteOptions._ +import org.apache.hudi.common.schema.HoodieSchema +import org.apache.hudi.testutils.HoodieSparkClientTestBase + +import org.apache.spark.sql.{Row, SaveMode, SparkSession} +import org.apache.spark.sql.types._ +import org.junit.jupiter.api.{AfterEach, BeforeEach, Test} +import org.junit.jupiter.api.Assertions._ + +/** + * End-to-end tests for the hudi_vector_search table-valued function. + * Tests both single-query and batch-query modes with Spark SQL and DataFrame API. + */ +class TestHoodieVectorSearchFunction extends HoodieSparkClientTestBase { + + var spark: SparkSession = null + private val corpusPath = "corpus" + private val corpusViewName = "corpus_view" + + // Test corpus: 5 unit-ish vectors in 3D for easy manual verification + // doc_1: [1, 0, 0] - x-axis + // doc_2: [0, 1, 0] - y-axis + // doc_3: [0, 0, 1] - z-axis + // doc_4: [0.707, 0.707, 0] - 45 degrees in xy-plane (normalized) + // doc_5: [0.577, 0.577, 0.577] - equal in all 3 dims (normalized) + private val corpusData = Seq( + ("doc_1", Seq(1.0f, 0.0f, 0.0f), "x-axis"), + ("doc_2", Seq(0.0f, 1.0f, 0.0f), "y-axis"), + ("doc_3", Seq(0.0f, 0.0f, 1.0f), "z-axis"), + ("doc_4", Seq(0.70710678f, 0.70710678f, 0.0f), "xy-diagonal"), + ("doc_5", Seq(0.57735027f, 0.57735027f, 0.57735027f), "xyz-diagonal") + ) + + @BeforeEach override def setUp(): Unit = { + initPath() + initSparkContexts() + spark = sqlContext.sparkSession + initTestDataGenerator() + initHoodieStorage() + createCorpusTable() + } + + @AfterEach override def tearDown(): Unit = { + spark.catalog.dropTempView(corpusViewName) + cleanupSparkContexts() + cleanupTestDataGenerator() + cleanupFileSystem() + } + + private def createCorpusTable(): Unit = { + val metadata = new MetadataBuilder() + .putString(HoodieSchema.TYPE_METADATA_FIELD, "VECTOR(3)") + .build() + + val schema = StructType(Seq( + StructField("id", StringType, nullable = false), + StructField("embedding", ArrayType(FloatType, containsNull = false), + nullable = false, metadata), + StructField("label", StringType, nullable = true) + )) + + val rows = corpusData.map { case (id, emb, label) => + Row(id, emb, label) + } + + val df = spark.createDataFrame( + spark.sparkContext.parallelize(rows), + schema + ) + + df.write.format("hudi") + .option(RECORDKEY_FIELD.key, "id") + .option(PRECOMBINE_FIELD.key, "id") + .option(TABLE_NAME.key, "vector_search_corpus") + .option(TABLE_TYPE.key, "COPY_ON_WRITE") + .mode(SaveMode.Overwrite) + .save(basePath + "/" + corpusPath) + + spark.read.format("hudi").load(basePath + "/" + corpusPath) + .createOrReplaceTempView(corpusViewName) + } + + @Test + def testSingleQueryCosineDistance(): Unit = { + // Query vector [1, 0, 0] should be closest to doc_1, then doc_4, then doc_5 + val result = spark.sql( + s""" + |SELECT id, label, _hudi_distance + |FROM hudi_vector_search( + | '$corpusViewName', + | 'embedding', + | ARRAY(1.0, 0.0, 0.0), + | 3, + | 'cosine' + |) + |ORDER BY _hudi_distance + |""".stripMargin + ).collect() + + assertEquals(3, result.length) + + // doc_1 [1,0,0]: cosine distance to [1,0,0] = 1 - 1.0 = 0.0 + assertEquals("doc_1", result(0).getAs[String]("id")) + assertEquals(0.0, result(0).getAs[Double]("_hudi_distance"), 1e-5) + + // doc_4 [0.707,0.707,0]: cosine distance to [1,0,0] = 1 - 0.707 ~= 0.293 + assertEquals("doc_4", result(1).getAs[String]("id")) + assertEquals(1.0 - 0.70710678, result(1).getAs[Double]("_hudi_distance"), 1e-4) Review Comment: is there a cleaner way we can do these assertions? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
