How to deal with empty Terms and/or Zero/One BDDs

[michaelhglass]

We have a rich processing chain that directly derives reliability functions from terms and/or BDDs. However, if we pass e.g. a zero or a one BDD, the functions will typically observe a system that fails at time 0 or will never fail. Both situations are against the assumption of a reliability function to be 1 at time 0 and 0 at infinity. We need to think about a concept to resolve this throughout JReliability.

[felixreimann]

Would specific degenerated reliability functions do the job? https://en.wikipedia.org/wiki/Degenerate_distribution
A reliability function has to be monotonically decreasing. Wouldn't this also hold for the constant R(x)=1 and the constant R(x)=0 function?
What else would break if we just implement it like this (distributions, the gui, ...)?

[michaelhglass]

This I am not sure. By definition, a reliability function should be 1 at time 0 and 0 at infinity. So both degenerated function either violate the condition at 1 or at infinity. Consider for example the evaluator for the first moment and a constant 1 function (not to mix with constant failure rate). The area below the curve is then infinity. I think we should check what happens to our framework. One may be able to tolerate degenerated functions, giving the user respective strange results. I am just unsure whether all our approximations etc. terminate. Should we go ahead and enable degenerated functions? Then we should possibly create a branch and check all functions.
Follow-up question would then be: Should we create a marker for ReliabilityFunctions that decay to a degenerated function? Something like "isDegenerated?"? This could ease the consideration also for tests. Several functions can be easily configured to become degenerated (Linear term with empty lhs).

[felixreimann]

Another possibility would be to throw an Exception (not a RuntimeException) as soon as you cannot handle it: public ReliabilityFunction convert() throws DegeneratedBDDException;

…

Am 07.07.2017 um 10:16 schrieb Michael Glass ***@***.***>: This I am not sure. By definition, a reliability function should be 1 at time 0 and 0 at infinity. So both degenerated function either violate the condition at 1 or at infinity. Consider for example the evaluator for the first moment and a constant 1 function (not to mix with constant failure rate). The area below the curve is then infinity. I think we should check what happens to our framework. One may be able to tolerate degenerated functions, giving the user respective strange results. I am just unsure whether all our approximations etc. terminate. Should we go ahead and enable degenerated functions? Then we should possibly create a branch and check all functions. Follow-up question would then be: Should we create a marker for ReliabilityFunctions that decay to a degenerated function? Something like "isDegenerated?"? This could ease the consideration also for tests. Several functions can be easily configured to become degenerated (Linear term with empty lhs). — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub, or mute the thread.

[michaelhglass]

Hmmmm. So this basically means we would only capture this when converting something to ReliabilityFunction. But we could also just create a new ReliabilityFunction. :D Also, for example the Moment Evaluator would benefit from being able to ask the Function whether the assumptions will hold; otherwise, we would have to integrate this check in every class. Lots of options.