[SPARK-9903] [MLLIB] skip local processing in PrefixSpan if there are no small prefixes

There exists a chance that the prefixes keep growing to the maximum pattern length. Then the final local processing step becomes unnecessary. feynmanliang

Author: Xiangrui Meng <meng@databricks.com>

Closes #8136 from mengxr/SPARK-9903.

mllib/src/main/scala/org/apache/spark/mllib/fpm/PrefixSpan.scala

@@ -282,25 +282,30 @@ object PrefixSpan extends Logging {
       largePrefixes = newLargePrefixes
     }
 
-    // Switch to local processing.
-    val bcSmallPrefixes = sc.broadcast(smallPrefixes)
-    val distributedFreqPattern = postfixes.flatMap { postfix =>
-      bcSmallPrefixes.value.values.map { prefix =>
-        (prefix.id, postfix.project(prefix).compressed)
-      }.filter(_._2.nonEmpty)
-    }.groupByKey().flatMap { case (id, projPostfixes) =>
-      val prefix = bcSmallPrefixes.value(id)
-      val localPrefixSpan = new LocalPrefixSpan(minCount, maxPatternLength - prefix.length)
-      // TODO: We collect projected postfixes into memory. We should also compare the performance
-      // TODO: of keeping them on shuffle files.
-      localPrefixSpan.run(projPostfixes.toArray).map { case (pattern, count) =>
-        (prefix.items ++ pattern, count)
+    var freqPatterns = sc.parallelize(localFreqPatterns, 1)
+
+    val numSmallPrefixes = smallPrefixes.size
+    logInfo(s"number of small prefixes for local processing: $numSmallPrefixes")
+    if (numSmallPrefixes > 0) {
+      // Switch to local processing.
+      val bcSmallPrefixes = sc.broadcast(smallPrefixes)
+      val distributedFreqPattern = postfixes.flatMap { postfix =>
+        bcSmallPrefixes.value.values.map { prefix =>
+          (prefix.id, postfix.project(prefix).compressed)
+        }.filter(_._2.nonEmpty)
+      }.groupByKey().flatMap { case (id, projPostfixes) =>
+        val prefix = bcSmallPrefixes.value(id)
+        val localPrefixSpan = new LocalPrefixSpan(minCount, maxPatternLength - prefix.length)
+        // TODO: We collect projected postfixes into memory. We should also compare the performance
+        // TODO: of keeping them on shuffle files.
+        localPrefixSpan.run(projPostfixes.toArray).map { case (pattern, count) =>
+          (prefix.items ++ pattern, count)
+        }
       }
+      // Union local frequent patterns and distributed ones.
+      freqPatterns = freqPatterns ++ distributedFreqPattern
     }
 
-    // Union local frequent patterns and distributed ones.
-    val freqPatterns = (sc.parallelize(localFreqPatterns, 1) ++ distributedFreqPattern)
-      .persist(StorageLevel.MEMORY_AND_DISK)
     freqPatterns
   }