def test_constraint_dropping(self):
from lale.lib.sklearn import LogisticRegression
from lale.operators import make_operator
from lale.search.schema2search_space import op_to_search_space
orig_schemas = LogisticRegression._schemas
mod_schemas = {
**orig_schemas,
"properties": {
**orig_schemas["properties"],
"hyperparams": {
"allOf": [
s if i == 0 else {**s, "forOptimizer": False}
for i, s in enumerate(
orig_schemas["properties"]["hyperparams"]["allOf"]
)
]
},
},
}
orig_space = op_to_search_space(LogisticRegression)
mod_op = make_operator(LogisticRegression._impl_class(), mod_schemas)
mod_space = op_to_search_space(mod_op)
# dropping constraints makes the search space smaller
self.assertGreater(len(str(orig_space)), len(str(mod_space)))
def op_to_search_space_grids(
op: PlannedOperator,
pgo: Optional[PGO] = None,
data_schema: Dict[str, Any] = {}) -> List[SearchSpaceGrid]:
search_space = op_to_search_space(op, pgo=pgo, data_schema=data_schema)
grids = search_space_to_grids(search_space)
return grids
def hyperopt_search_space(
op: "PlannedOperator", schema=None, pgo: Optional[PGO] = None, data_schema={}
):
search_space = op_to_search_space(op, pgo=pgo, data_schema=data_schema)
if search_space:
name = op.name()
if should_print_search_space("true", "all", "backend", "hyperopt"):
print(
f"hyperopt search space for {name}: {search_space_to_hp_str(search_space, name)}"
)
return search_space_to_hp_expr(search_space, name)
else:
return None
def test_override_int_param2(self):
from lale.lib.sklearn import PCA
from lale.search.schema2search_space import op_to_search_space
from lale.search.search_space import SearchSpaceNumber, SearchSpaceObject
pca = PCA.customize_schema(
relevantToOptimizer=["iterated_power"],
iterated_power=schemas.Float(
minimum=0,
minimumForOptimizer=1,
maximum=6,
maximumForOptimizer=5,
exclusiveMaximumForOptimizer=False,
exclusiveMinimumForOptimizer=True,
),
)
search = op_to_search_space(pca)
assert isinstance(search, SearchSpaceObject)
num_space = list(search.choices)[0][0]
assert isinstance(num_space, SearchSpaceNumber)
self.assertEqual(num_space.minimum, 1)
self.assertEqual(num_space.maximum, 5)
self.assertFalse(num_space.exclusiveMaximum)
self.assertTrue(num_space.exclusiveMinimum)
def test_override_float_param1(self):
from lale.lib.sklearn import PCA
from lale.search.schema2search_space import op_to_search_space
from lale.search.search_space import SearchSpaceNumber, SearchSpaceObject
pca = PCA.customize_schema(
relevantToOptimizer=["tol"],
tol=schemas.Float(
minimum=0.25,
minimumForOptimizer=0,
maximum=0.5,
maximumForOptimizer=1.0,
exclusiveMaximum=True,
exclusiveMinimum=False,
),
)
search = op_to_search_space(pca)
assert isinstance(search, SearchSpaceObject)
num_space = list(search.choices)[0][0]
assert isinstance(num_space, SearchSpaceNumber)
self.assertEqual(num_space.minimum, 0.25)
self.assertEqual(num_space.maximum, 0.5)
self.assertTrue(num_space.exclusiveMaximum)
self.assertFalse(num_space.exclusiveMinimum)
def op_to_search_space_grids(op:PlannedOperator,
pgo:Optional[PGO]=None)->List[SearchSpaceGrid]:
search_space = op_to_search_space(op, pgo=pgo)
grids = search_space_to_grids(search_space)
return grids