Frédéric Branchaud-Charron
Frédéric Branchaud-Charron
Hello! Thank you for submitting your issue, I'm happy to see that this is useful. The CSG of 1.0 for Cifar10 is when using CNN autoencoder features and t-sne. Table...
@dgistJihunJung This can happen when no anchor matches the target box.
Hello, I'm not super familiar with NER, but Baal should be able to handle it easily. Here is a script that load a NER model and outputs prediction. We use...
Hello, 1. Yes those would have the highest uncertainty. I had to do some modifications to Baal to make ActiveLearningLoop work on NER. I opened #263 to fix that. I...
Oh I made a mistake in the Gist, now updated. We first load the initial weights before training. ```python for _ in range(2): trainer.load_state_dict(init_weights) print(f"Active learning: labelled={al_dataset.n_labelled} unlabelled={al_dataset.n_unlabelled}") trainer.train() trainer.lr_scheduler...
That make sense! Something like `wrapper.train_and_test_on_datasets(eval_set='val')`? For backward compatibility, we would still keep `test` as the default. What do you think?
Hello, sorry for the delay. Do you mean that you take the loss as a measure of uncertainty?
Awesome! Thank you very much :) We would be happy to help you as well if needed
I'll read the full paper over the weekend. I would know more about how we could work on this.
After reading the paper and looking at the codebase, I think the main issue is that we can't easily supply all `targets` to the heuristic. We could directly provide the...