pyPESTO: Mismatch of history vs optimizer reported optimal values
When executing the rosenbrock.ipynb notebook on develop branch, I get massive complaints:
Function values from history and optimizer do not match: 1.3168387678656086, 2.0179911928533514
Parameters obtained from history and optimizer do not match: [0.99553374 0.98995083 0.97544456 0.94768476 0.89550744 0.8019584
0.64185456 0.40180156 0.14794723 0.00844711], [0.9940831 0.98497745 0.96635783 0.93780681 0.87380595 0.74153588
0.53395112 0.25523609 0.03957006 0.00100739]
Function values from history and optimizer do not match: 1.6349167203078765, 2.336760892779954
I guess something must still be wrong about history values vs. values returned by optimizers. Happens for various optimizers.
About this issue
- Original URL
- State: closed
- Created 4 years ago
- Comments: 21 (21 by maintainers)
Agreed that it’s good to have both available. My guess would be that users mostly care about good parameters and less so about whether they were (what some optimizer considers) the final parameters, or whether they occurred elsewhere. So
history_beats_optimizer_result=Truewould be a reasonable default for me. Can live with either, though. Regarding warning: If we have ahistory_beats_optimizer_resultoption, I would not issue a warning anymore. If at all, then only debug-level logging.Same here…