def on_epoch_end(self, last_metrics, **kwargs):
# Convert from fastai format
my_targets = []
for batch_boxes, batch_labels in self.targets:
for boxes, labels in zip(batch_boxes, batch_labels):
non_pad_inds = labels != 0
boxes = to_box_pixel(boxes, self.h, self.w)
my_targets.append(
(boxes[non_pad_inds, :], labels[non_pad_inds]))
my_outputs = [(boxlist.boxes, boxlist.get_field('labels'),
boxlist.get_field('scores'))
for boxlist in self.outputs]
metric = compute_coco_eval(my_outputs, my_targets, self.num_labels)[0]
return add_metrics(last_metrics, metric)
def on_epoch_end(self, last_metrics, **kwargs):
return add_metrics(last_metrics, 100 * self.correct / self.total)
def on_epoch_end(self, last_metrics, **kwargs):
self.predictions = np.array(self.predictions)
metric = self.evaluator.get_final_metric(self.predictions)
self.predictions = []
return add_metrics(last_metrics, metric)
def on_epoch_end(self, last_metrics, **kwargs):
self.epsilon = self.learn.model.exploration_strategy.epsilon
return add_metrics(last_metrics, [self.epsilon])
def on_epoch_end(self, last_metrics, **kwargs: Any):
return add_metrics(last_metrics,
[sum(self.train_reward),
sum(self.valid_reward)])
def on_epoch_end(self, last_metrics, **kwargs):
self.epsilon = self.learn.exploration_method.epsilon
if last_metrics and last_metrics[-1] is None: del last_metrics[-1]
return add_metrics(last_metrics, [float(self.epsilon)])