We have something of a test suite for mixed behavior profiles (see src/pygambit/tests/test_behav.py). It is at the moment written in unittest style. Further, it is not ideal as the separate tests are not isolated (they all depend on strategy profiles which are defined at the class level).
#. Rewrite the tests to be in pytest style, with individual functions exercising specific features
#. Construct a parameterisation of the tests to consider several interesting games, and test edge cases. This requires a bit of experience with game theory to be aware of where calculations with mixed behavior profiles are likely to have problems (specifically with parts of the tree that are reached with zero probability!)
This could be considered in conjunction with #374 as well. The issues with the existing test suite are slightly different (the mixed behavior tests do consider a nontrivial game unlike the mixed strategy tests). However, more robust testing of the mapping between strategy and behavior profiles and the application of Kuhn's Theorem (and testing what happens for games without perfect recall!!!) could make it work someone attacking both issues in tandem.
We have something of a test suite for mixed behavior profiles (see
src/pygambit/tests/test_behav.py). It is at the moment written inunitteststyle. Further, it is not ideal as the separate tests are not isolated (they all depend on strategy profiles which are defined at the class level).#. Rewrite the tests to be in
pyteststyle, with individual functions exercising specific features#. Construct a parameterisation of the tests to consider several interesting games, and test edge cases. This requires a bit of experience with game theory to be aware of where calculations with mixed behavior profiles are likely to have problems (specifically with parts of the tree that are reached with zero probability!)
This could be considered in conjunction with #374 as well. The issues with the existing test suite are slightly different (the mixed behavior tests do consider a nontrivial game unlike the mixed strategy tests). However, more robust testing of the mapping between strategy and behavior profiles and the application of Kuhn's Theorem (and testing what happens for games without perfect recall!!!) could make it work someone attacking both issues in tandem.