def testThompsonSamplerValidation(self):
generator = ThompsonSampler(min_weight=0.01)
# all Xs are not the same
with self.assertRaises(ValueError):
generator.fit(
Xs=[[[1, 1], [2, 2], [3, 3], [4, 4]], [[1, 1], [2, 2], [4, 4]]],
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
# multiple observations per parameterization
with self.assertRaises(ValueError):
generator.fit(
Xs=[[[1, 1], [2, 2], [2, 2]]],
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
# these are not the same observations, so should not error
generator.fit(
Xs=[[[1, 1], [2.0, 2], [2, 2]]],
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
# requires objective weights
with self.assertRaises(ValueError):
generator.gen(5, self.parameter_values, objective_weights=None)
def testThompsonSamplerOutcomeConstraints(self):
generator = ThompsonSampler(min_weight=0.0)
generator.fit(
Xs=self.multiple_metrics_Xs,
Ys=self.multiple_metrics_Ys,
Yvars=self.multiple_metrics_Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
arms, weights, _ = generator.gen(
n=4,
parameter_values=self.parameter_values,
objective_weights=np.array([1, 0]),
outcome_constraints=(
# pass in multiples of the same constraint
# to ensure that shapes are correct for multiple constraints
np.array([[0, 1], [0, 1], [0, 1]]),
np.array([[1], [1], [1]]),
),
)
self.assertEqual(arms, [[3, 3], [4, 4], [2, 2], [1, 1]])
for weight, expected_weight in zip(
weights, [4 * i for i in [0.4, 0.4, 0.15, 0.05]]
):
self.assertAlmostEqual(weight, expected_weight, delta=0.1)
def testThompsonSamplerUniformWeights(self):
generator = ThompsonSampler(min_weight=0.0, uniform_weights=True)
generator.fit(
Xs=self.Xs,
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
arms, weights, _ = generator.gen(
n=3, parameter_values=self.parameter_values, objective_weights=np.ones(1)
)
self.assertEqual(arms, [[4, 4], [3, 3], [2, 2]])
for weight, expected_weight in zip(weights, [1.0, 1.0, 1.0]):
self.assertAlmostEqual(weight, expected_weight, 1)
def testThompsonSamplerMinWeight(self):
generator = ThompsonSampler(min_weight=0.01)
generator.fit(
Xs=self.Xs,
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
arms, weights = generator.gen(
n=5, parameter_values=self.parameter_values, objective_weights=np.ones(1)
)
self.assertEqual(arms, [[4, 4], [3, 3], [2, 2]])
for weight, expected_weight in zip(weights, [0.725, 0.225, 0.05]):
self.assertAlmostEqual(weight, expected_weight, 1)
def testThompsonSampler(self):
generator = ThompsonSampler(min_weight=0.0)
generator.fit(
Xs=self.Xs,
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
arms, weights, gen_metadata = generator.gen(
n=3, parameter_values=self.parameter_values, objective_weights=np.ones(1)
)
self.assertEqual(arms, [[4, 4], [3, 3], [2, 2]])
for weight, expected_weight in zip(
weights, [3 * i for i in [0.725, 0.225, 0.05]]
):
self.assertAlmostEqual(weight, expected_weight, 1)
self.assertEqual(len(gen_metadata["arms_to_weights"]), 4)
def testThompsonSamplerPredict(self):
generator = ThompsonSampler(min_weight=0.0)
generator.fit(
Xs=self.Xs,
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
f, cov = generator.predict([[1, 1], [3, 3]])
self.assertTrue(np.array_equal(f, np.array([[1], [3]])))
self.assertTrue(np.array_equal(cov, np.ones((2, 1, 1))))
with self.assertRaises(ValueError):
generator.predict([[1, 2]])
def testThompsonSamplerInfeasible(self):
generator = ThompsonSampler(min_weight=0.9)
generator.fit(
Xs=self.Xs,
Ys=self.Ys,
Yvars=self.Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
with self.assertRaises(ModelError):
generator.gen(
n=3,
parameter_values=self.parameter_values,
objective_weights=np.ones(1),
)
def testThompsonSamplerOutcomeConstraintsInfeasible(self):
generator = ThompsonSampler(min_weight=0.0)
generator.fit(
Xs=self.multiple_metrics_Xs,
Ys=self.multiple_metrics_Ys,
Yvars=self.multiple_metrics_Yvars,
parameter_values=self.parameter_values,
outcome_names=self.outcome_names,
)
with self.assertRaises(ValueError):
generator.gen(
n=3,
parameter_values=self.parameter_values,
objective_weights=np.ones(2),
outcome_constraints=(np.array([[0, 1]]), np.array([[-10]])),
)
def get_thompson(
experiment: Experiment,
data: Data,
search_space: Optional[SearchSpace] = None,
num_samples: int = 10000,
min_weight: Optional[float] = None,
uniform_weights: bool = False,
) -> DiscreteModelBridge:
"""Instantiates a Thompson sampling model."""
if data.df.empty: # pragma: no cover
raise ValueError("Thompson sampler requires non-empty data.")
model = ThompsonSampler(num_samples=num_samples,
min_weight=min_weight,
uniform_weights=uniform_weights)
return DiscreteModelBridge(
experiment=experiment,
search_space=search_space
if search_space is not None else experiment.search_space,
data=data,
model=model,
transforms=TS_trans,
)
