This shows you the differences between two versions of the page.
Both sides previous revision Previous revision | |||
derivative_free_optimization [2012/05/24 09:58] prokop |
derivative_free_optimization [2012/05/24 09:59] |
||
---|---|---|---|
Line 1: | Line 1: | ||
- | ====== Derivative Free Optimization ====== | ||
- | |||
- | -http://en.wikipedia.org/wiki/Pattern_search_(optimization) | ||
- | -http://en.wikipedia.org/wiki/Random_search | ||
- | -http://en.wikipedia.org/wiki/Nelder%E2%80%93Mead_method (Ameba) | ||
- | |||
- | Pattern search je relativne rychly protoze sampluje pouze vyznacne smery. Minimalne sampluje 1 smer, maximalne 2N. Pamatuje si predchozi smer takze se pohybuje effektivne. | ||
- | |||
- | problem nastava v okamziku kdy se objevi uzke udoli sikme na hlavni osy. V tom pripade Pattern_search sice konverguje ale s velice kratkym krokem (aby se vlez do sirky udoli) coz vede k velkemu poctu potrebnych iteraci. | ||
- | |||
- | Resenim je pouzit znalosti okolnich bodu k odhadu gradientu, pripadne i stredu paraboly. | ||
- | |||
- | |||
- | === algoritmus === | ||
- | |||
- | == Pattern run == | ||
- | |