import copy
import inspect
import itertools
from geomstats.vectorization import repeat_point
KNOWN_POINT_NAMES = (
"point",
"base_point",
"point_a",
"point_b",
"fiber_point",
"mat",
)
KNOWN_VECTOR_NAMES = (
"vec",
"vector",
"tangent_vec",
"tangent_vec_a",
"tangent_vec_b",
"tangent_vec_c",
"tangent_vec_d",
"cotangent_vec_a",
"cotangent_vec_b",
"direction",
)
KNOWN_ARGS = KNOWN_POINT_NAMES + KNOWN_VECTOR_NAMES
def _filter_combs(n_args, combs, vectorization_type):
if vectorization_type == "sym" or n_args == 1:
return combs
repeat_indices = [int(val) for val in vectorization_type.split("-")[1:]]
if len(repeat_indices) == n_args:
return combs
remove_indices = [val for val in range(n_args) if val not in repeat_indices]
for comb in combs.copy():
vals_remove = [comb[remove_index] for remove_index in remove_indices]
if max(vals_remove) == 0 or min(vals_remove) == 1 and min(comb) == 1:
continue
combs.remove(comb)
return combs
def _check_vectorization_type(vectorization_type, n_args):
if vectorization_type in ("sym", "basic") or n_args == 1:
return
try:
val = None
for val in vectorization_type.split("-")[1:]:
int(val)
if val is None or int(val) > n_args:
raise ValueError(
"Unable to repeat unexisting args for vectorization type "
f"`{vectorization_type}` and `n_args = {n_args}"
)
except ValueError:
raise ValueError(
f"Unable to understand vectorization type `{vectorization_type}`. "
"Can handle `sym` and `repeat-(int)` format."
)
def _generate_datum_vectorization_data(
datum, comb_indices, arg_names, expected_name, n_reps=2
):
if expected_name is not None:
has_expected = True
if type(expected_name) is str:
expected_name = [expected_name]
expected_combs = []
for expected_name_ in expected_name:
expected = datum.get(expected_name_)
expected_combs.append([expected, repeat_point(expected, n_reps=n_reps)])
else:
has_expected = False
args_combs = []
for arg_name in arg_names:
arg = datum.get(arg_name)
args_combs.append([arg, repeat_point(arg, n_reps=n_reps, expand=True)])
new_data = []
for indices in comb_indices:
new_datum = copy.copy(datum)
if has_expected:
rep = int(1 in indices)
for expected_i, expected_name_ in enumerate(expected_name):
new_datum[expected_name_] = expected_combs[expected_i][rep]
for arg_i, (index, arg_name) in enumerate(zip(indices, arg_names)):
new_datum[arg_name] = args_combs[arg_i][index]
new_data.append(new_datum)
return new_data
[docs]
def generate_vectorization_data(
data,
arg_names,
expected_name=None,
n_reps=2,
vectorization_type="sym",
):
"""Create new data with vectorized version of inputs.
Parameters
----------
data : list of dict
Data. Each to vectorize.
arg_names: list
Name of inputs to vectorize.
expected_name: str or list of str
Output name in case it needs to be repeated.
n_reps: int
Number of times the input points should be repeated.
vectorization_type: str
Possible values are 'sym', 'basic', or the format 'repeat-(int)'
(e.g. "repeat-0-2").
'sym': tests all repetition combinations.
'basic': tests only no repetition and repetition of all.
'repeat-(int)': tests repetition of provided indices.
"""
n_args = len(arg_names)
_check_vectorization_type(vectorization_type, n_args)
if vectorization_type == "basic":
comb_indices = [tuple(i for _ in range(n_args)) for i in range(2)]
else:
comb_indices = list(itertools.product(*[range(2)] * n_args))
comb_indices = _filter_combs(n_args, comb_indices, vectorization_type)
comb_indices.pop(0)
new_data = []
for datum in data:
new_data.extend(
_generate_datum_vectorization_data(
datum,
comb_indices,
arg_names,
expected_name=expected_name,
n_reps=n_reps,
)
)
return new_data
def _generate_random_data(data_generator, arg_names):
data = {}
base_point = None
for arg_name in arg_names:
if arg_name in KNOWN_POINT_NAMES:
base_point = data[arg_name] = data_generator.random_point()
for arg_name in arg_names:
if arg_name in KNOWN_VECTOR_NAMES:
data[arg_name] = data_generator.random_tangent_vec(base_point)
return data
def _get_vectorization_type(test_case, arg_names):
if test_case.tangent_to_multiple:
return "sym"
tangent_vec_type = ""
for k, arg_name in enumerate(arg_names):
if arg_name in KNOWN_VECTOR_NAMES:
tangent_vec_type += f"-{k}"
return "repeat" + tangent_vec_type if tangent_vec_type else "sym"
[docs]
def test_vectorization(self, test_func, n_reps, atol):
# TODO: move this to decorator?
# TODO: accept kwargs?
arg_names = list(inspect.signature(test_func).parameters.keys())
arg_names = list(filter(lambda x: x in KNOWN_ARGS, arg_names))
data = _generate_random_data(self.data_generator, arg_names)
geometry = (
self.space.metric
if hasattr(self, "is_metric") and self.is_metric
else self.space
)
data["expected"] = getattr(geometry, test_func.__name__[5:])(**data)
data["atol"] = atol
vec_data = generate_vectorization_data(
data=[data],
arg_names=arg_names,
expected_name="expected",
n_reps=n_reps,
vectorization_type=_get_vectorization_type(self, arg_names),
)
self._test_vectorization(vec_data, test_fnc_name=test_func.__name__)