[SPARK-7615] MLLIB Word2Vec wordVectors divided by Euclidean Norm equals to zero

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 1.3.1
• Fix Version/s: 1.6.1, 2.0.0
• Component/s: MLlib
• Assignee: Sean R. Owen
• Reporter: Eric Li
• Created: 13/May/15 21:16
• Updated: 12/Jan/16 13:28
• Resolved: 12/Jan/16 13:28

Description

In Word2VecModel, wordVecNorms may contains Euclidean Norm equals to zero. This will cause incorrect calculation for cosine distance. when you do cosineVec(ind) / wordVecNorms(ind). Cosine distance should be equal to 0 for norm = 0.

Issue Links

• links to: [Github] Pull Request #6245 (ezli)
• links to: [Github] Pull Request #10696 (srowen)