On Thu, Aug 5, 2010 at 2:38 PM, Sebastian Schelter <[email protected]>wrote:
> Hi Jake, > > the patch attached at MAHOUT-456 should fix that. > > The problem is that RowSimilarityJob (which is used internally by the > RecommenderJob) produces SequentialAccessSparseVectors with cardinality > Integer.MAX_VALUE. When assign(Vector) is invoked on those later, the > inner loop in AbstractVector.assign() runs from 0 to Integer.MAX_VALUE... > > I made RowSimilarityJob produce RandomAccessSparseVectors to prevent this. > Would it not be better to fix that inner loop of AbstractVector.assign(), to zero out the current values, followed by using iterateNonZero() on the argument? -jake
