Skip to content

create WMH only once and broadcast #194

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Feb 7, 2019
Merged

create WMH only once and broadcast #194

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion src/main/java/tech/sourced/gemini/WeightedMinHash.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,8 @@
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.io.Serializable;

import static java.lang.Math.floor;
import static java.lang.Math.log;

Expand All @@ -15,7 +17,7 @@
* https://github.com/ekzhu/datasketch/blob/master/datasketch/weighted_minhash.py
* https://github.com/src-d/go-license-detector/blob/master/licensedb/internal/wmh/wmh.go
*/
public class WeightedMinHash {
public class WeightedMinHash implements Serializable {
private static final Logger log = LoggerFactory.getLogger(WeightedMinHash.class);

protected int dim;
Expand Down
25 changes: 21 additions & 4 deletions src/main/scala/tech/sourced/gemini/Hash.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -196,6 +196,8 @@ class Hash(session: SparkSession,
): Dataset[RDDHash] = {
log.warn("hashing features")

val wmh = makeBroadcastedWmh(docFreq.tokens.size, sampleSize)

val tf = features.rdd
.map { case Row(feature: String, doc: String, weight: Long) => (RDDFeatureKey(feature, doc), weight) }
.reduceByKey(_ + _)
Expand All @@ -204,9 +206,8 @@ class Hash(session: SparkSession,
.map(row => (row._1.doc, Feature(row._1.token, row._2)))
.groupByKey(session.sparkContext.defaultParallelism)
.mapPartitions { partIter =>
val wmh = FeaturesHash.initWmh(docFreq.tokens.size, sampleSize) // ~1.6 Gb (for 1 PGA bucket)
partIter.map { case (doc, features) =>
RDDHash(doc, wmh.hash(FeaturesHash.toBagOfFeatures(features.iterator, docFreq)))
RDDHash(doc, wmh.value.hash(FeaturesHash.toBagOfFeatures(features.iterator, docFreq)))
}
}
tfIdf.toDS()
Expand All @@ -222,17 +223,33 @@ class Hash(session: SparkSession,
): Dataset[RDDHash] = {
log.warn("hashing features")

val wmh = makeBroadcastedWmh(docFreq.tokens.size, sampleSize)
val tf = features.groupBy("feature", "doc").sum("weight").alias("weight")
val tfIdf = tf
.map { case Row(token: String, doc: String, weight: Long) => (doc, Feature(token, weight)) }
.groupByKey { case (doc, _) => doc }
.mapGroups { (doc, features) =>
val wmh = FeaturesHash.initWmh(docFreq.tokens.size, sampleSize) // ~1.6 Gb RAM (for 1 PGA bucket)
RDDHash(doc, wmh.hash(FeaturesHash.toBagOfFeatures(features.map(_._2), docFreq)))
RDDHash(doc, wmh.value.hash(FeaturesHash.toBagOfFeatures(features.map(_._2), docFreq)))
}
tfIdf
}

/**
* Create WeightedMinHash instance and broadcasts it
*
* create it only once and keep on node
* because the instance is relatively huge (2 * N of features * sampleSize(160 or 256 depends on mode) * 8)
* According to tests ~1.6 Gb per 1 PGA bucket (but really depends on bucket)
*
* @param tokens number of features
* @param sampleSize depends on hashing mode and threshold
* @return
*/
def makeBroadcastedWmh(tokens: Int, sampleSize: Int): Broadcast[WeightedMinHash] = {
val wmh = FeaturesHash.initWmh(tokens, sampleSize)
session.sparkContext.broadcast(wmh)
}

protected def saveDocFreqToDB(docFreq: OrderedDocFreq, keyspace: String, tables: Tables): Unit = {
log.warn(s"save document frequencies to DB")

Expand Down