def score(self,
actual: np.array,
predicted: np.array,
sample_weight: typing.Optional[np.array] = None,
labels: typing.Optional[np.array] = None,
X: typing.Optional[dt.Frame] = None,
**kwargs) -> float:
logger = None
# if self._make_logger:
# Example use of logger, with required import of:
# from h2oaicore.systemutils import make_experiment_logger, loggerinfo
# Can use loggerwarning, loggererror, etc. for different levels
# if self.context and self.context.experiment_id:
# logger = make_experiment_logger(experiment_id=self.context.experiment_id, tmp_dir=self.context.tmp_dir,
# experiment_tmp_dir=self.context.experiment_tmp_dir)
N = actual.shape[0]
if sample_weight is None:
sample_weight = np.ones(N)
# label actual values as 1 or 0
lb = LabelEncoder()
labels = lb.fit_transform(labels)
# create datatable with all data
DT = dt.Frame(actual=lb.transform(actual),
predicted=np.minimum(
1 - self.__class__._epsilon,
np.maximum(self.__class__._epsilon, predicted)),
cost_fn=self.make_cost_values(self.__class__._fn_cost, X,
N, 1.),
cost_tp=self.make_cost_values(self.__class__._tp_cost, X,
N, 0.),
cost_tn=self.make_cost_values(self.__class__._tn_cost, X,
N, 0.),
cost_fp=self.make_cost_values(self.__class__._fp_cost, X,
N, 1.),
sample_weight=sample_weight)
lloss = DT[:, f.sample_weight *
(f.actual * (f.cost_fn * dt.log(f.predicted) +
f.cost_tp * dt.log(1 - f.predicted)) +
(1 - f.actual) * (f.cost_fp * dt.log(1 - f.predicted) +
f.cost_tn * dt.log(f.predicted)))]
lloss = lloss.sum()[0, 0] * -1.0 / np.sum(sample_weight)
return lloss
def transform(self, X: dt.Frame):
return X[:, [(dt.f[i] / dt.abs(dt.f[i])) * dt.exp(0.5 * dt.log(dt.abs(dt.f[i]))) for i in range(X.ncols)]]
def transform(self, X: dt.Frame):
return X[:, [dt.log(dt.f[i]) for i in range(X.ncols)]]
def transform(self, X: dt.Frame):
if dtype_global() == np.float32:
return X[:, [dt.float32(dt.log(dt.f[i])) for i in range(X.ncols)]]
else:
return X[:, [dt.float64(dt.log(dt.f[i])) for i in range(X.ncols)]]
def transform(self, X: dt.Frame):
assert X.ncols == 1
return X[:, dt.stype.str32(dt.stype.int32(dt.log(dt.f[0])))]
