Github user jkbradley commented on a diff in the pull request:
https://github.com/apache/spark/pull/5467#discussion_r28550207
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/feature/Word2Vec.scala ---
@@ -479,9 +492,16 @@ class Word2VecModel private[mllib] (
*/
def findSynonyms(vector: Vector, num: Int): Array[(String, Double)] = {
require(num > 0, "Number of similar words should > 0")
- // TODO: optimize top-k
- val fVector = vector.toArray.map(_.toFloat)
- model.mapValues(vec => cosineSimilarity(fVector, vec))
+
+ val numWords = wordVectors.numRows
+ val cosineVec = Vectors.zeros(numWords).asInstanceOf[DenseVector]
+ BLAS.gemv(1.0, wordVectors, vector.asInstanceOf[DenseVector], 0.0,
cosineVec)
+
+ // Need not divide with the norm of the given vector since it is
constant.
+ val updatedCosines = indexedModel.map { case (_, ind) =>
--- End diff --
Here's where you're assuming something about the ordering of Map items.
Instead of this, I'd recommend just operating on cosineVec and wordVecNorms.
After finding the top word indices, you can look them up in the wordIndex
(indexedModel). I'd also use a while loop instead of a map since it should be
faster.
I also wonder if we should store the word index in both directions (since
we need both in the 2 findSynonyms methods):
* ```wordList: Array[String]``` (words ordered by indices)
* ```wordIndex: Map[String, Int]``` (map from word to index)
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
