class FilterConfig(NamedTuple):
attribution_method: str = IntegratedGradients.get_name()
attribution_arguments: Dict[str, Any] = {
arg: config.value
for arg, config in ATTRIBUTION_METHOD_CONFIG[
IntegratedGradients.get_name()].items()
}
prediction: str = "all"
classes: List[str] = []
count: int = 4
class FilterConfig(NamedTuple):
attribution_method: str = IntegratedGradients.get_name()
# issue with mypy github.com/python/mypy/issues/8376
attribution_arguments: Dict[str, Any] = {
arg: config.value # type: ignore
for arg, config in ATTRIBUTION_METHOD_CONFIG[
IntegratedGradients.get_name()].params.items()
}
prediction: str = "all"
classes: List[str] = []
num_examples: int = 4
for cls in SUPPORTED_ATTRIBUTION_METHODS
}
def _str_to_tuple(s):
if isinstance(s, tuple):
return s
return tuple([int(i) for i in s.split()])
def _str_to_bool(s):
return False if s == "False" else True
ATTRIBUTION_METHOD_CONFIG: Dict[str, ConfigParameters] = {
IntegratedGradients.get_name():
ConfigParameters(
params={
"n_steps":
NumberConfig(value=25, limit=(2, None)),
"method":
StrEnumConfig(limit=SUPPORTED_METHODS, value="gausslegendre"),
},
post_process={"n_steps": int},
),
FeatureAblation.get_name():
ConfigParameters(params={
"perturbations_per_eval":
NumberConfig(value=1, limit=(1, 100))
}, ),
Deconvolution.get_name():
