In my mind the cost goes into applying the cleaning program (fast), computing the quality function (can be slow), and search. I imagine the following optimizations. Which are actually supported? Is this a complete list?
- parallelize
- incremental computation of quality function/constraints. e.g.,
ops = ops_prev + [new_op]
qfn(data, ops) == qfn'(qfn(data, ops_prev), new_op)
- caching intermediate results e.g.,
cache(qfn(data, ops))
- estimating effects of an operation e.g.,
est_qfn(data, ops) ~= qfn(data, ops) . # est_qfn is very fast
- statically pruning the cleaning program.
In my mind the cost goes into applying the cleaning program (fast), computing the quality function (can be slow), and search. I imagine the following optimizations. Which are actually supported? Is this a complete list?
cache(qfn(data, ops))