Confusion about the diversity metric and other loss terms

[TobiasLee]

Hi, thanks for the nice paper and neat code, well done.
After read the paper and check the code, I was confused about several lines in model_task.py, from 306 to 312.

            result['pi_kl'] = pi_kl
            result['diversity'] = th.mean(p)
            result['nll'] = self.nll(dec_outputs, labels)
            result['b_pr'] = b_pr
            result['mi'] = mi

According to the config in sl_cat.py, it seems that the loss of diversity, pi_kl and mi is not used. From my understanding, the diversity is checking whether the learned log_qy is a orthogonal matrix, rather than that diversity described in the paper:

Diversity is mea- sured by the number of unique responses the model used in all scenarios from the test data

Am I right? Could you please tell more about these metrics? Thx a lot.