def on_epoch_end(self, last_metrics, **kwargs):
prec = self._precision()
rec = self._recall()
metric = (1 + self.beta2) * prec * rec / (prec * self.beta2 + rec + self.eps)
metric[metric != metric] = 0 # removing potential "nan"s
if self.avg: metric = (self._weights(avg=self.avg) * metric).sum()
return add_metrics(last_metrics, metric)
def on_epoch_end(self, last_metrics, **kwargs):
label = np.concatenate(self._label)
score = np.concatenate(self._score)
if len(set(label)) != 2:
auc = 0.5
else:
score = np.concatenate(self._score)
auc = roc_auc_score(label,score)
return add_metrics(last_metrics, auc)
def on_epoch_end(self, last_metrics, **kwargs):
return add_metrics(last_metrics, self._precision())
def on_epoch_end(self, last_metrics, **kwargs):
return add_metrics(last_metrics, self._recall())
def on_epoch_end(self, last_metrics, **kwargs):
return add_metrics(
last_metrics, self.compute_f1(self.tp, self.fp, self.fn))
