def test_is_numeric():
assert _is_numeric(0)
assert _is_numeric(0.0)
assert not _is_numeric(True)
assert not _is_numeric(False)
assert not _is_numeric("0")
assert not _is_numeric(None)
def _get_autolog_metrics(fitted_model):
result_metrics = {}
failed_evaluating_metrics = set()
for metric in _autolog_metric_allowlist:
try:
if hasattr(fitted_model, metric):
metric_value = getattr(fitted_model, metric)
if _is_numeric(metric_value):
result_metrics[metric] = metric_value
except Exception:
failed_evaluating_metrics.add(metric)
if len(failed_evaluating_metrics) > 0:
_logger.warning(
f"Failed to autolog metrics: {', '.join(sorted(failed_evaluating_metrics))}."
)
return result_metrics
def results_to_dict(results):
"""
Turns a ResultsWrapper object into a python dict
:param results: instance of a ResultsWrapper returned by a call to `fit`
:return: a python dictionary with those metrics that are (a) a real number, or (b) an array
of the same length of the number of coefficients
"""
has_features = False
features = results.model.exog_names
if features is not None:
has_features = True
nfeat = len(features)
results_dict = {}
for f in dir(results):
try:
field = getattr(results, f)
# Get all fields except covariances and private ones
if (
not callable(field)
and not f.startswith("__")
and not f.startswith("_")
and not f.startswith("cov_")
):
if (
has_features
and isinstance(field, np.ndarray)
and field.ndim == 1
and field.shape[0] == nfeat
):
d = dict(zip(features, field))
renamed_keys_dict = prepend_to_keys(d, f)
results_dict.update(renamed_keys_dict)
elif _is_numeric(field):
results_dict[f] = field
except AttributeError:
pass
return results_dict
