def test_rejection_predict(self):
learner = CorralLearner(
[FixedLearner([1 / 2, 1 / 2]),
FixedLearner([1 / 4, 3 / 4])],
eta=0.5,
mode="rejection")
predict = learner.predict(None, [1, 2])[0]
self.assertEqual(.375, predict[0])
self.assertEqual(.625, predict[1])
def test_off_policy_predict(self):
learner = CorralLearner(
[FixedLearner([1 / 2, 1 / 2]),
FixedLearner([1 / 4, 3 / 4])],
eta=0.5,
mode="off-policy")
predict = learner.predict(None, [1, 2])[0]
self.assertEqual(.375, predict[0])
self.assertEqual(.625, predict[1])
def test_importance_learn(self):
actions = [1, 2]
base1 = ReceivedLearnFixedLearner([1 / 2, 1 / 2])
base2 = ReceivedLearnFixedLearner([1 / 4, 3 / 4])
learner = CorralLearner([base1, base2], eta=0.5, mode="importance")
predict, info = learner.predict(None, actions)
action = actions[0]
probability = predict[0]
reward = 1 / 2
learner.learn(None, action, reward, probability, info)
self.assertEqual((None, 1, 1, 1 / 2), base1.received_learn)
self.assertEqual((None, 2, 0, 3 / 4), base2.received_learn)
def test_importance_predict(self):
learner = CorralLearner(
[FixedLearner([1 / 2, 1 / 2]),
FixedLearner([1 / 4, 3 / 4])],
eta=0.5,
mode="importance")
mean_predict = list(
map(mean,
zip(*[learner.predict(None, [1, 2])[0] for _ in range(1000)])))
self.assertAlmostEqual(1 / 2 * 1 / 2 + 1 / 2 * 1 / 4, mean_predict[0],
2)
self.assertAlmostEqual(1 / 2 * 1 / 2 + 1 / 2 * 3 / 4, mean_predict[1],
2)
def test_off_policy_learn(self):
actions = [1, 2]
base1 = ReceivedLearnFixedLearner([1 / 2, 1 / 2], 'a')
base2 = ReceivedLearnFixedLearner([1 / 4, 3 / 4], 'b')
learner = CorralLearner([base1, base2], eta=0.5, mode="off-policy")
predict, info = learner.predict(None, actions)
action = actions[0]
probability = predict[0]
reward = 1
learner.learn(None, action, reward, probability, info)
info = InteractionContext.learner_info
self.assertDictEqual({'a': 1, 'b': 1, **info}, info)
self.assertEqual((None, action, reward, predict[0]),
base1.received_learn)
self.assertEqual((None, action, reward, predict[0]),
base2.received_learn)
def test_rejection_learn(self):
actions = [0, 1]
base1 = ReceivedLearnFixedLearner([1 / 2, 1 / 2], 'a')
base2 = ReceivedLearnFixedLearner([1 / 4, 3 / 4], 'b')
learner = CorralLearner([base1, base2], eta=0.5, mode="rejection")
predict, info = learner.predict(None, actions)
action = actions[0]
probability = predict[0]
reward = 1
base1_learn_cnt = [0, 0]
base2_learn_cnt = [0, 0]
random = CobaRandom(1)
for _ in range(1000):
action = random.choice(actions, predict)
probability = predict[actions.index(action)]
learner.learn(None, action, reward, probability, info)
base1_learn_cnt[action] += int(base1.received_learn is not None)
base2_learn_cnt[action] += int(base2.received_learn is not None)
base1.received_learn = None
base2.received_learn = None
self.assertLessEqual(
abs(base1_learn_cnt[0] / sum(base1_learn_cnt) - 1 / 2), .02)
self.assertLessEqual(
abs(base1_learn_cnt[1] / sum(base1_learn_cnt) - 1 / 2), .02)
self.assertLessEqual(
abs(base2_learn_cnt[0] / sum(base2_learn_cnt) - 1 / 4), .02)
self.assertLessEqual(
abs(base2_learn_cnt[1] / sum(base2_learn_cnt) - 3 / 4), .02)
def test_params(self):
LinUCBLearner
base1_name = 'A'
base2_name = 'B'
expected = {
'family': 'corral',
'eta': 0.075,
'mode': 'importance',
'T': float('inf'),
'B': [base1_name, base2_name],
'seed': 1
}
actual = CorralLearner([FamilyLearner("A"), FamilyLearner("B")]).params
self.assertEqual(expected, actual)
"""
This is an example script that creates a Benchmark that matches the bandit bakeoff paper.
This script requires that the matplotlib and vowpalwabbit packages be installed.
"""
from coba.learners import RandomLearner, EpsilonLearner, VowpalLearner, UcbTunedLearner, CorralLearner
from coba.benchmarks import Benchmark
if __name__ == '__main__':
benchmark = Benchmark.from_file("./examples/benchmark_short.json")
learners = [
RandomLearner(),
EpsilonLearner(epsilon=0.025),
UcbTunedLearner(),
VowpalLearner(bag=5, seed=10),
CorralLearner([VowpalLearner(bag=5, seed=10),
UcbTunedLearner()],
eta=.075,
T=40000,
seed=10),
]
benchmark.evaluate(learners, './examples/bakeoff.log',
seed=10).standard_plot()