slow `predict_as_dataframe` function

[sevmag]

The function:

graphnet/src/graphnet/models/easy_model.py

Line 323 in 401f28b

def predict_as_dataframe(

Can be really slow when querying for additional attributes, which is handled here:

graphnet/src/graphnet/models/easy_model.py

Lines 381 to 397 in 401f28b

	for batch in dataloader:
	for attr in attributes:
	attribute = batch[attr]
	if isinstance(attribute, torch.Tensor):
	attribute = attribute.detach().cpu().numpy()
	# Check if node level predictions
	# If true, additional attributes are repeated
	# to make dimensions fit
	if pulse_level_predictions:
	if len(attribute) < np.sum(
	batch.n_pulses.detach().cpu().numpy()
	):
	attribute = np.repeat(
	attribute, batch.n_pulses.detach().cpu().numpy()
	)
	attributes[attr].extend(attribute)

In the prediction loop, we are already looping through the whole data. Can we think of a way to utilise that initial loop already to get additional attributes without looping through the dataloader again? I think the least we can do is put a progress bar in the second loop to let the user know that the program hasn't stalled (I was very confused at times, why does it take so long.?).