Implemented Yamachi, k64r from Axelrod's Second.

[gaffney2010]

Fingerprints below.

The refactoring is somewhat straight-forward with the exception of /~https://github.com/Axelrod-Python/TourExec/blob/v0.3.0/src/strategies/k64r.f#L30. Here, I refactored as
P = |E-F|, but E+F = M-1, so that P = |2E+1-M|. Then 10*P < M iff P/(2M) < 0.05. Equivalently, |(E+0.5)/M - 0.5| < 0.05. I called (E+0.5)/M as portion_defect and bounded by 0.45 and 0.55.
Also a little different is I update history before deciding on a move, so to emulate the Fortan (which does after), I wait until the third turn to start (instead of the second) and look back one extra turn.

I think count_them_us_them was intended to be indexed by them_three_ago, us_two_ago, them_last, so that it predicts the opponent's next turn. It kind of predicts the opponents last move instead... [If you look at lines 22 and 23 of the Fortan, it's moving A(X,Y) based on J, but J can't be in response to Y, our own last move.]

[drvinceknight]

This all looks good to me! Thanks @gaffney2010.

FYI, I've worked through the logic and also run the axelrod_fortran test suite with this strategy.