Message2265

Author malte
Recipients malte, mikko
Date 2012-06-16.23:57:58
Content
If I understand the issue correctly, the problem is that the causal graph
ordering procedure gives logically different results for the two problem
formulations.

As long as the procedure works as advertised (i.e. as described in the Fast
Downward paper) in all cases, I wouldn't classify this as a bug or as something
that ought to be changed. In general it cannot be avoided that heuristic
decisions are influenced by innocuous changes in the problem formulation.

For example, if you duplicate certain operators by introducting
identically-behaving operators with different names, this will change the weight
assigned to the induced arcs in the causal graph and hence affect the causal
graph sorting criterion. Similarly, breaking one logical operator with many
parameters into two parts with fewer parameters can dramatically reduce the
weight of a variable in the causal graph sorting.

For what it's worth, our handling of axioms is generally quite poor at the
moment, and if there's a way to avoid them, it's usually a good idea. (See issue98.)

I'll mark this as resolved, but if I'm misunderstanding something about the
nature of the issue, feel free to reopen.
History
Date User Action Args
2012-06-16 23:57:58maltesetmessageid: <1339883878.79.0.810681001579.issue341@gmail.com>
2012-06-16 23:57:58maltesetrecipients: + malte, mikko
2012-06-16 23:57:58maltelinkissue341 messages
2012-06-16 23:57:58maltecreate