Exploitation set in WD classifiers

[HeitorBoschirolli]

If I understood the paper correctly, the exploitation set is used only to train the feature extractor, if that's the case why is it necessary to specify the users for the exploitation set to train the WD model?

I tried to pass an empty set as the exploitation set but that caused an error.

[luizgh]

The development set is used for feature learning, and the exploitation set is used during WD training.

The experimental protocol for GPDS also uses signatures from the development set as negative examples for WD training, but it is possible to train WD classifiers with both positive and negative examples from the exploitation set: you just need to set --forg-from_dev=0 and --forg-from_exp=$n with $n being the number of random forgeries from the exploitation set (per user) that should be used.

[HeitorBoschirolli]

I mistook the sets. What I want to do is train WD classifiers using all the signatures in the dataset, so I want to set the development set to an empty set.

For example, to train WD classifiers for the MCYT using the feature extractor trained in the GPDS, the exploitation set should contain any signatures?

[luizgh]

In our experimental protocol, for the MCYT there is only the exploitation set (no development set). So for training with signatures from all users in the dataset, you just need the exp-users to have all users (e.g.--exp-users 0 75) and inform that you will not use random forgeries from the development set (--forg-from_dev=0), but rather from the exploitation set: --forg-from_exp=$n. I hope this helps

[HeitorBoschirolli]

Helped a lot, thanks