PrincetonUniversity / PsyNeuLink / 28116648026

84%

master: 85%

push

github

Fix PEC degenerate-likelihood warning re-printing stale parameter sets (#3571)

# Summary
The Parameter Estimation Composition (PEC) prints a warning summarizing
parameter sets that produced degenerate objective-function values (e.g.
BadLikelihoodWarning when a simulation has zero variance and the KDE
can't be computed). A user reported that this warning repeatedly lists
parameter sets that were not just evaluated — pinning the first bad set
seen to the top of the log and re-printing it on every subsequent
iteration, cluttering the output.

# Root cause
The optimizer callback (progress_callback in both
_fit_differential_evolution and _fit_optuna) summarized warnings from a
warns_with_params list that was never cleared. Each callback re-printed
the entire accumulated history of degenerate parameter sets, so the
earliest bad set kept reappearing regardless of what was actually
evaluated in the current iteration/trial.

A related latent bug: the per-evaluation warns buffer was only cleared
inside the display_iter branch, so with display_iter=False a stale
warns[-1] could be mis-attributed to a later (possibly warning-free)
parameter set.

# Changes
Extracted the duplicated "print summary + clear" logic into a new
_report_degenerate_warnings helper, which clears warns_with_params after
printing so each callback only reports sets that went degenerate since
the last callback.
Always clear the per-evaluation warns buffer at the end of each
evaluation (not just in the display_iter branch), and hoisted the
warning check into a single got_pec_warning flag so the captured warning
and parameters always come from the same evaluation.
Fixed a missing space in the footer text (fitting todata → fitting to
data).

11020 of 14380 branches covered (76.63%)

Branch coverage included in aggregate %.

14 of 14 new or added lines in 1 file covered. (100.0%)

37665 of 43558 relevant lines covered (86.47%)

0.86 hits per line