def non_test_goofspiel():
g = Goofspiel(4, scoring=Goofspiel.Scoring.ZEROSUM)
mc = OutcomeMCCFR(g, seed=42)
for s in [10, 100, 1000]:
mc.compute(s)
br = BestResponse(g, 0, [None, mc])
print(
"Exploit after", s,
np.mean([
g.play_strategies([br, mc], seed=i)[-1].values()[0]
for i in range(1000)
]))
vs = GoofSpielCardsValueStore(g)
val = SparseStochasticValueLearning(g, vs, seed=43)
for alpha in [0.1, 0.01, 0.01, 0.001, 0.0001]:
print(alpha)
val.compute([mc, mc], 200, alpha)
def test_goofspeil_rewards():
g = Goofspiel(2, Goofspiel.Scoring.ZEROSUM, rewards=[100, 11])
for _ in range(10):
history = g.play_strategies([UniformStrategy(), UniformStrategy()])
t = history[-1]
assert t.values() in ([0, 0], [-89, 89], [89, -89])