Logits are the same as softmax scores

[mihow]

The logits returned with classifications are a copy of the softmax scores! This was fixed in the ood branch but never brought into main. Noticed by @rhine3

The bug can be seen here:
https://github.com/RolnickLab/ami-data-companion/blob/main/trapdata/api/models/classification.py#L78

One fix can be found here:

ami-data-companion/trapdata/api/models/classification.py

Lines 73 to 128 in 6a1b16f

	def post_process_batch(
	self, logits: torch.Tensor, features: torch.Tensor | None = None
	) -> list[ClassifierResult]:
	"""
	Return the labels, softmax/calibrated scores, and the original logits for
	each image in the batch.
	Almost like the base class method, but we need to return the logits as well.
	each image in the batch, along with optional feature vectors.
	"""
	predictions = torch.nn.functional.softmax(logits, dim=1)
	predictions = predictions.cpu().numpy()
	if self.class_prior is None:
	ood_scores = np.max(predictions, axis=-1)
	else:
	ood_scores = np.max(predictions - self.class_prior, axis=-1)
	# Ensure higher scores indicate more likelihood that it is OOD.
	ood_scores = 1 - ood_scores
	features = features.cpu() if features is not None else None
	logits = logits.cpu()
	batch_results = []
	for i, pred in enumerate(predictions):
	class_indices = np.arange(len(pred))
	labels_single = [self.category_map[i] for i in class_indices]
	ood_score = ood_scores[i]
	logits_single = logits[i].float().tolist()
	feature_vector = (
	features[i].float().tolist() if features is not None else None
	)
	result = ClassifierResult(
	features=feature_vector,
	labels=labels_single,
	logits=logits_single,
	scores=pred,
	ood_score=ood_score,
	)
	batch_results.append(result)
	logger.debug(f"Post-processing result batch with {len(batch_results)} entries.")
	return batch_results
	def predict_batch(self, batch, return_features: bool = False):
	batch_input = batch.to(self.device, non_blocking=True)
	if return_features:
	features = self.get_features(batch_input)
	logits = self.model(batch_input)
	return logits, features
	logits = self.model(batch_input)
	return logits, None