Credible set generation seems to have some flaws

[Daniel Suveges]

Hi Guys, I'm afraid, the function that generates credible sets based of the vectors of the inclusion probabilities is not right:

There's this condition:

    if np.min(np.abs(LDloc)) > purity:
      cred.append(sorted(list(credset)))

If ALL the abs(r) values reach the purity threshold, the logic keeps all elements, however, if at least one of r is "less pure", the credible set remains empty, as this condition doesn't have an else branch. I highly suspect this is not the intended behaviour.

[cuiran]

Thanks for the comment. This is indeed the intended behavior. We refer to the original SuSiE algorithm for justification. See the original SuSiE code for filtering credible set based on purity threshold. Also, see the SuSiE manuscript section 3.2.2 for their description and justification for this credible set filtering behavior.

[Daniel Suveges]

Oh, OK. Thank you so much for the clarification!