load_state_dict

[YiDanCheung]

Thank you for your outstanding work!

But I'm wondering how to use load_state_dict(), in episodic_training.ipynb you use few_shot_classifier.load_state_dict(best_state), but classical_training.ipynb you use model.load_state_dict(best_state) while the the former is PrototypicalNetworks and the latter is resnet12.

Looking for your reply!

[ebennequin]

It's because in episodic training we train the few-shot classifier, while in classical training we only train the feature extractor.

[YiDanCheung]

It's because in episodic training we train the few-shot classifier, while in classical training we only train the feature extractor.

So maybe in classical_training.ipynb best_state = copy.deepcopy(few_shot_classifier.state_dict()) should be best_state = copy.deepcopy(model.state_dict()), and it would not cause error like RuntimeError: Error(s) in loading state_dict for ResNet : Missing key(s) in state_dict: "conv1.weight", ...

[ebennequin]

Indeed! It's a mistake that we introduced in this commit. We should make the change.