TupleBundles ( http://dl.acm.org/citation.cfm?id=1376686 ) are a tactic for
computing over probabilistic data.
TupleBundles ( http://dl.acm.org/citation.cfm?id=1376686 ) are a tactic for
computing over probabilistic data. Loosely put, the approach is to compile
the query to evaluate simultaneously in N possible worlds. The results can
then be aggregated to produce an assortment of statistics, etc...
This class actually wraps three different compilation strategies inspired
by tuple bundles, each handling parallelization in a slightly different way
* **Long**: Not technically "TupleBundles". This approach simply unions
* together a set of results, one per possible world sampled.
* **Flat**: Creates a wide result, splitting each non-deterministic column
into a set of columns, one per sample.
* **Array**: Like flat, but uses native array types to avoid overpopulating
the result schema.
At present, only 'Flat' is fully implemented, although a 'Long'-like approach
can be achieved by using convertFlatToLong.
TupleBundles ( http://dl.acm.org/citation.cfm?id=1376686 ) are a tactic for computing over probabilistic data. Loosely put, the approach is to compile the query to evaluate simultaneously in N possible worlds. The results can then be aggregated to produce an assortment of statistics, etc...
This class actually wraps three different compilation strategies inspired by tuple bundles, each handling parallelization in a slightly different way
* **Long**: Not technically "TupleBundles". This approach simply unions * together a set of results, one per possible world sampled. * **Flat**: Creates a wide result, splitting each non-deterministic column into a set of columns, one per sample. * **Array**: Like flat, but uses native array types to avoid overpopulating the result schema.
At present, only 'Flat' is fully implemented, although a 'Long'-like approach can be achieved by using convertFlatToLong.