Player, NormalFormGame: Make payoff_array's C contiguous

[oyamad]

Make payoff_array's C contiguous when when a payoff_profile_array is passed to NormalFormGame, in order for Numba not to complain.

[oyamad]

The errors occur with Numba 0.29.0, but not with 0.28.1 or 0.26.0.

[sglyon]

Sorry, I didn't quite follow.

Does this change-set fix the errors on all versions of numba, or do we still have issues on 0.29?

[oyamad]

Sorry for not being clear.

• The errors from Travis CI are not related to this PR. They are from random/utilities.py and lss.py. Those errors only occur with the latest Numba.
• What this PR does is to make all the NormalFormGame.players[i].payoff_array's C contiguous, where there is no point of having a non-contiguous numpy array, and there is indeed a case (in some code I am currently writing) that passing a tuple of a C-contiguous array and a F-contiguous array does a bad thing making Numba complain (which has nothing to do with the errors here).

[oyamad]

For the error from random/utilities.py, I filed an issue numba/numba#2159. I haven't investigated for the one from lss.py.

[oyamad]

Opened numba/numba#2169 for the error from lss.py.

In fact, this error can be avoided by removing A = np.asarray(A) assuming that the input A is already an np.array, which allows us to have nopython=True for @jit.

[oyamad]

I changed the patch (and made a forced push), where order='C' is now applied when instantiating a Player.

[oyamad]

Let me bring the previous commit back...

[coveralls]

Coverage increased (+0.03%) to 86.674% when pulling e4c633b on nfg_fix into 6da6bfe on master.

[oyamad]

Rebased on master. This is ready to merge.

[mmcky]

Thanks @oyamad merging this into master now.