Optimal strategy for reactor problem does not seem to be right

Create issue
Issue #241 closed
Iñigo Bermejo created an issue

Optimal strategy shown includes the decision to take a test (which has a cost) but then the strategy does not depend on the tests' result.


Comments (2)

  1. Manuel Luque

    The network attached is not the correct version of the "Reactor problem". It is also incorrect in the repository networks. For example, it can be seen as the "Result of test" in the attached network does not give any information about "Advanced reactor reliability", because the conditional probability table, in numbers, is uniform. Thus, the inference algorithm selects to build a reactor, although the result of the test is completely irrelevant, just because the expected utility of building the conventional test is higher than no performing the test (and thus not building any reactor).

  2. Log in to comment