def _validate_metrics_map(self):
"""Ensure `metrics_map` input parameter is properly formatted and yields callable functions for all metrics"""
if not (isinstance(self.metrics_map, dict)
or isinstance(self.metrics_map, list)):
raise TypeError(
'metrics_map must be one of: [dict, list]. Received type: {}.'.
format(type(self.metrics_map)))
#################### If metrics_map is list, convert to dict with None values ####################
if isinstance(self.metrics_map, list):
self.metrics_map = {_: None for _ in self.metrics_map}
for _m_key, _m_val in self.metrics_map.items():
if not isinstance(_m_key, str):
raise TypeError(
'metrics_map ids must be strings. Received type {}: {}'.
format(type(_m_key), _m_key))
if not any(
[callable(_m_val),
isinstance(_m_val, str), _m_val is None]):
raise TypeError(
'metrics_map values must be one of: [callable, str, None]. Received {}'
.format(type(_m_val)))
#################### Check sklearn.metrics for: _m_val if str, or _m_key if _m_val is None ####################
if not callable(_m_val):
try:
self.metrics_map[_m_key] = sk_metrics.__getattribute__(
_m_key if _m_val is None else _m_val)
except AttributeError:
raise AttributeError(
'"sklearn.metrics" has no attribute "{}".'.format(
_m_key if _m_val is None else _m_val))
def _set_metric_function(self, f):
"""Ensure provided `f` is a valid callable
Parameters
----------
f: Callable, string, None
See `metric_function` documentation of :meth:`Metric.__init__`
Returns
-------
Callable
A function derived from `f` if `f` was not already callable. Else `f`"""
if not callable(f):
try:
return sk_metrics.__getattribute__(self.name if f is None else f)
except AttributeError:
raise AttributeError(f"`sklearn.metrics` has no attribute: {f or self.name}")
return f
def _validate_metrics_map(self):
"""Ensure `metrics_map` input parameter is properly formatted and yields callable functions
for all metrics"""
if not (isinstance(self.metrics_map, dict) or isinstance(self.metrics_map, list)):
raise TypeError(f"metrics_map must be a dict, or list, not {type(self.metrics_map)}")
# If metrics_map is list, convert to dict with None values
if isinstance(self.metrics_map, list):
self.metrics_map = {_: None for _ in self.metrics_map}
for _m_key, _m_val in self.metrics_map.items():
if not isinstance(_m_key, str):
raise TypeError(f"`metrics_map` ids must be strings. Received {_m_key}")
if not any([callable(_m_val), isinstance(_m_val, str), _m_val is None]):
raise TypeError(f"metrics_map values must be callable, str, or None, not {_m_val}")
# Check sklearn.metrics for: _m_val if str, or _m_key if _m_val is None
if not callable(_m_val):
try:
self.metrics_map[_m_key] = sk_metrics.__getattribute__(
_m_key if _m_val is None else _m_val
)
except AttributeError:
raise AttributeError(f"`sklearn.metrics` has no attribute: {_m_val or _m_key}")
def get_metric(metric_name):
metric = metrics.__getattribute__(metric_name)
return metric
def get_metric(config: DictConfig):
metric_config = config["metric"]
metric_name = metric_config["name"]
metric = skm.__getattribute__(metric_name)
return metric
