Closed dietmarwo closed 4 years ago
My testing revealed clipping issues (boundary violations) for some of the algorithms - for HHO I already provided a PR related to this. The underlying GTOP C-code will log warnings about "negative flight time" which cannot be suppressed if the bounds are violated. Some links with GTOP-results for multi-threaded algorithms: http://www.midaco-solver.com/index.php/about/benchmarks/gtopx (there is also an alternative python wrapper) and https://github.com/dietmarwo/fast-cma-es/blob/master/Results.adoc . Midaco and fcmaes are the only optimizers I know which can solve all of these problems. The Midaco team writes: "Therefore researchers should not be discouraged if struggling to solve these benchmarks. Instead, the GTOPX library aims at senior researchers wishing to put advanced algorithms to the test."
Am Di., 23. Juni 2020 um 11:26 Uhr schrieb Raneem Qaddoura < notifications@github.com>:
Merged #48 https://github.com/7ossam81/EvoloPy/pull/48 into master.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/7ossam81/EvoloPy/pull/48#event-3472012897, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOIZMKRBQR4B2UQHVPNH4VTRYBYLRANCNFSM4ODLCC6A .
This PR adds real world benchmark examples from https://www.esa.int/gsp/ACT/projects/gtop/ . Testing with these can further test/improve the algorithms. One issue with HHO - handling of lists as bounds - is already fixed in this PR. Beside that I noticed exceptions with GA, which I locally fixed by catching these, but it would be better if the root cause would be analyzed and fixed. A new dependency - fcmaes - is required because these benchmarks are implemented there.