Apache Mahout Incompatible Changes and Limitations
CDH 5.1 introduces the following incompatible changes.
Minor changes in behavior:
- MAHOUT-1368
The org.apache.mahout.math.stats.OnlineSummarizer algorithm has changed, potentially leading to different results.
-
The streaming KMeans implementation has changed in how it outputs centroids when run outside of a Hadoop cluster.
Major Developer API changes
- MAHOUT-1296:
The following implementations have been removed:
- Clustering:
- DirichletMeanShift
- MinHash
- Eigencuts in o.a.m.clustering.spectral.eigencuts
- Classification:
- WinnowPerceptron
- Frequent Pattern Mining
- Collaborative Filtering:
- All recommenders in o.a.m.cf.taste.impl.recommender.knn
- TreeClusteringRecommender in o.a.m.cf.taste.impl.recommender
- SlopeOne implementations in o.a.m..cf.taste.hadoop.slopeone and o.a.m.cf.taste.impl.recommender.slopeone
- Distributed pseudo recommender in o.a.m.cf.taste.hadoop.pseudo
- Clustering:
-
The examples/bin/build-reuters.sh script has been removed.
Minor Developer API changes
- MAHOUT-1280:
Some SSVD support code, such as UpperTriangularMatrix, has been moved to mahout-math.
- MAHOUT-1363:
Scala-related math code has been moved into an org.apache.mahout.math.scalabindings sub-package.
Minor packaging changes
- MAHOUT-1382:
Move to Guava 16.
- MAHOUT-1364:
Update to require Lucene 4.6.1.