Inconsistency with "embedded" CPT

Originally reported on Google Code with ID 11

Consider the following CPT example in Leapp;

x = ?:(1/3)
y = ?:(1/4)
z1 = ?! ( ~x & ~y -> ?:(1/5),
          ~x &  y -> ?:(1/7),
           x & ~y -> ?:(1/2),
           x &  y -> ?:(1/2))

Notice that z1 is independent of y when x is true (i.e. "context-specific Independence").

Then, in the following refactoring

z0 = ?! ( ~y -> ?:(1/5),
           y -> ?:(1/7))
z3 = ?! ( ~x -> z0,
          x -> ?:(1/2))

z3 is expected to be equivalent to z1.

However the two distributions unexpectedly differ:

lea> :@ z1 ! y
11/42
lea> :@ z3 ! y
8/21

Reported by pde@n-side.com on 2015-03-14 13:53:52

Comments (6)