phymbert / spark-search   0.2.0

import org.apache.spark.search.rdd._ // to implicitly enhance RDD with search features

// Load some Amazon computer user reviews
val computersReviews: RDD[Review] = loadReviews("**/*/reviews_Computers.json.gz") 
    // Number of partition is the number of Lucene index which will be created across your cluster
    .repartition(4)

// Count positive review: indexation + count matched doc with fuzzy matching
computersReviews.count("reviewText:happy OR reviewText:best OR reviewText:good OR reviewText:\"sounds great\"~")

// Search for key words
computersReviews.searchList("reviewText:\"World of Warcraft\" OR reviewText:\"Civilization IV\"",
 topK = 100, minScore = 10)
  .foreach(println)

// /!\ Important lucene indexation is done each time a SearchRDD is computed,
// if you do multiple operations on the same parent RDD, you might have a variable in the driver:
val computersReviewsSearchRDD: SearchRDD[Review] = computersReviewsRDD.searchRDD(
  SearchOptions.builder[Review]() // See all other options SearchOptions, IndexationOptions and ReaderOptions
    .read((r: ReaderOptions.Builder[Review]) => r.defaultFieldName("reviewText"))
    .analyzer(classOf[EnglishAnalyzer])
    .build())

// Boolean queries and boosting examples returning RDD
computersReviewsSearchRDD.search("(RAM OR memory) AND (CPU OR processor~)^4", 15)
        .collect()
        .foreach(println)

// Fuzzy matching
computersReviews.searchList("(reviewerName:Mikey~0.8) OR (reviewerName:Wiliam~0.4) OR (reviewerName:jonh~0.2)",
                                      topKByPartition = 10)
                        .map(doc => s"${doc.source.reviewerName}=${doc.score}")
                        .foreach(println)

// RDD full text joining - example here searches for persons
// who did both computer and software reviews with fuzzy matching on reviewer name
val softwareReviews: RDD[Review] = loadReviews("**/*/reviews_Software_10.json.gz")
val matchesReviewers: RDD[(Review, Array[SearchRecord[Review]])] = computersReviews.matches(
                             softwareReviewsRDD.filter(_.reviewerName != null).map(sr => (sr.asin, sr)),
                            (sr: Review) => "reviewerName:\"" + sr.reviewerName + "\"~0.4",
                             topK = 10)
                            .values
matchesReviewersRDD
  .filter(_._2.nonEmpty)
  .map(m => (s"Reviewer ${m._1.reviewerName} reviews computer ${m._1.asin} but also on software:",
          m._2.map(h => s"${h.source.reviewerName}=${h.score}=${h.source.asin}").toList))
  .collect()
  .foreach(println)

// Drop duplicates
println("Dropping duplicated reviewers:")
val distinctReviewers: RDD[String] = computersReviews.searchDropDuplicates[Int, Review](
 queryBuilder = queryStringBuilder(sr => "reviewerName:\"" + sr.reviewerName.replace('"', ' ') + "\"~0.4")
).map(sr => sr.reviewerName)
distinctReviewers.collect().foreach(println)

// Save then restore onto hdfs
matchesReviewersRDD.save("/tmp/hdfs-pathname")
val restoredSearchRDD: SearchRDD[Review] = SearchRDD.load[Review](sc, "/tmp/hdfs-pathname")

// Restored index can be used as classical rdd
val topReviewer = restoredSearchRDD.map(r => (r.reviewerID, 1))
        .reduceByKey(_ + _)
        .sortBy(_._2, ascending = false)
        .take(1).head
println(s"${topReviewer._1} has submitted ${topReviewer._2} reviews")

import org.apache.spark.search.sql._

val sentences = spark.read.csv("...")
sentences.count("sentence:happy OR sentence:best OR sentence:good")

// coming soon: SearchSparkStrategy/LogicPlan & column enhanced with search
sentences.where($"sentence".matches($"searchKeyword" ))