steffen/Mel_Vision_Transformer_ComParE-21
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

bug in metric calculation

Browse Source
This commit is contained in:
Steffen Illium
2021-03-27 16:39:07 +01:00
parent 82835295a1
commit 37e36df0a8
12 changed files with 967 additions and 239 deletions

43
util/module_mixins.py

@ -56,18 +56,17 @@ class ValMixin:
for file_name in sorted_y:
sorted_y.update({file_name: torch.stack(sorted_y[file_name])})
y_max = torch.stack(
[torch.argmax(x.mean(dim=0)) if x.shape[0] > 1 else torch.argmax(x) for x in sorted_y.values()]
).squeeze()
y_one_hot = torch.nn.functional.one_hot(y_max, num_classes=self.params.n_classes).float()
target_y = torch.stack(tuple(sorted_batch_y.values())).long()
if self.params.n_classes <= 2:
if y_one_hot.ndim == 1:
y_one_hot = y_one_hot.unsqueeze(0)
if target_y.ndim == 1:
target_y = target_y.unsqueeze(-1)
self.metrics.update(y_one_hot, target_y)
mean_sorted_y = torch.stack([x.mean(dim=0) if x.shape[0] > 1 else x for x in sorted_y.values()])
self.metrics.update(mean_sorted_y, target_y)
else:
y_max = torch.stack(
[torch.argmax(x.mean(dim=0)) if x.shape[0] > 1 else torch.argmax(x) for x in sorted_y.values()]
).squeeze()
y_one_hot = torch.nn.functional.one_hot(y_max, num_classes=self.params.n_classes).float()
self.metrics.update(y_one_hot, target_y)
if self.params.n_classes <= 2:
val_loss = self.bce_loss(y.squeeze().float(), batch_y.float())
else:
@ -109,14 +108,15 @@ class ValMixin:
#mean_vote_loss = self.ce_loss(y_mean, sorted_batch_y)
#summary_dict.update(val_mean_vote_loss=mean_vote_loss)
y_max = torch.stack(
[torch.argmax(x.mean(dim=0)) if x.shape[0] > 1 else torch.argmax(x) for x in sorted_y.values()]
).squeeze()
y_one_hot = torch.nn.functional.one_hot(y_max, num_classes=self.params.n_classes).float()
if self.params.n_classes >= 2:
max_vote_loss = self.ce_loss(y_one_hot, sorted_batch_y)
if self.params.n_classes <= 2:
mean_sorted_y = torch.stack([x.mean(dim=0) if x.shape[0] > 1 else x for x in sorted_y.values()])
max_vote_loss = self.bce_loss(mean_sorted_y.float(), sorted_batch_y.float())
else:
max_vote_loss = self.bce_loss(y_one_hot, sorted_batch_y)
y_max = torch.stack(
[torch.argmax(x.mean(dim=0)) if x.shape[0] > 1 else torch.argmax(x) for x in sorted_y.values()]
).squeeze()
y_one_hot = torch.nn.functional.one_hot(y_max, num_classes=self.params.n_classes).float()
max_vote_loss = self.ce_loss(y_one_hot, sorted_batch_y)
summary_dict.update(val_max_vote_loss=max_vote_loss)
summary_dict.update({f'mean_{key}': torch.mean(torch.stack([output[key]
@ -124,7 +124,10 @@ class ValMixin:
for key in keys if 'loss' in key}
)
# Sklearn Scores
additional_scores = self.additional_scores(dict(y=y_one_hot, batch_y=sorted_batch_y))
if self.params.n_classes <= 2:
additional_scores = self.additional_scores(dict(y=y_max, batch_y=sorted_batch_y))
else:
additional_scores = self.additional_scores(dict(y=y_one_hot, batch_y=sorted_batch_y))
summary_dict.update(**additional_scores)
pl_metrics, pl_images = self.metrics.compute_and_prepare()
@ -132,7 +135,9 @@ class ValMixin:
summary_dict.update(**pl_metrics)
summary_dict.update(epoch=self.current_epoch)
self.log_dict(summary_dict, on_epoch=True)
self.log_dict(summary_dict)
# For Debugging:
# print(f'Summary Metrics are: {summary_dict}')
for name, image in pl_images.items():
self.logger.log_image(name, image, step=self.global_step)