Calls build_posterior Args: domain (hyperopt.base.Domain): contains info about the obj function and the hp space passed to fmin prior_weight (float): smoothing factor for counts, to avoid having 0 prob # TODO: consider renaming or improving documentation
(domain, prior_weight, gamma)
| 792 | |
| 793 | |
| 794 | def build_posterior_wrapper(domain, prior_weight, gamma): |
| 795 | """ |
| 796 | Calls build_posterior |
| 797 | Args: |
| 798 | domain (hyperopt.base.Domain): contains info about the obj function and the hp |
| 799 | space passed to fmin |
| 800 | prior_weight (float): smoothing factor for counts, to avoid having 0 prob |
| 801 | # TODO: consider renaming or improving documentation for suggest |
| 802 | gamma (float): the threshold to split between l(x) and g(x), see eq. 2 in |
| 803 | https://papers.nips.cc/paper/4443-algorithms-for-hyper-parameter-optimization.pdf |
| 804 | |
| 805 | Returns: |
| 806 | |
| 807 | """ |
| 808 | |
| 809 | # -- these dummy values will be replaced in build_posterior() and never used |
| 810 | observed = {"idxs": pyll.Literal(), "vals": pyll.Literal()} |
| 811 | observed_loss = {"idxs": pyll.Literal(), "vals": pyll.Literal()} |
| 812 | |
| 813 | posterior = build_posterior( |
| 814 | # -- vectorized clone of bandit template |
| 815 | domain.vh.v_expr, |
| 816 | # -- this dict and next represent prior dists |
| 817 | domain.vh.idxs_by_label(), |
| 818 | domain.vh.vals_by_label(), |
| 819 | observed["idxs"], |
| 820 | observed["vals"], |
| 821 | observed_loss["idxs"], |
| 822 | observed_loss["vals"], |
| 823 | pyll.Literal(gamma), |
| 824 | pyll.Literal(float(prior_weight)), |
| 825 | ) |
| 826 | |
| 827 | return observed, observed_loss, posterior |
| 828 | |
| 829 | |
| 830 | def suggest( |
no test coverage detected