Interesting. If meta-data existed to indicate which parameters are continuous and which are discreet, could a future optimization algorithm use that information to improve the optimization process?
--- In [email protected], "Tomasz Janeczko" <[EMAIL PROTECTED]> wrote: > > Differential Evolution is also for CONTINUOUS functions, see > the AUTHORS' page: > http://www.icsi.berkeley.edu/~storn/code.html > > All those methods use gradient of fitness function change > to decide in which direction they should move. For binary (0 or 1) parameters > gradients make no sense. > > Best regards, > Tomasz Janeczko > amibroker.com > ----- Original Message ----- > From: "Steve Davis" <[EMAIL PROTECTED]> > To: <[email protected]> > Sent: Wednesday, October 01, 2008 4:30 PM > Subject: [amibroker] Re: CMAE behavior when optimizing control parameters? > > > > Thanks Paul and Tomasz, > > > > I have also used IO for many years and consulted with Fred on this > > issue. Fred suggested using the Differential Evolution algorithm > > rather than Particle Swarm when a system has many non-continuous > > parameters. > > > > In any case, Tomasz gave me the answer I needed regarding CMAE. > > > > Thanks again, > > Steve > > > > --- In [email protected], "Paul Ho" <paul.tsho@> wrote: > >> > >> Tomasz > >> > >> What you said and what I said can co-exist quite happily if you want > > to read > >> it again, and want to read it that way! > >> It is not a debate that I want to enter into with you. I am just > > sharing my > >> experience - it is "possible" to do it. > >> All of these IO used simulated "Continuous" parameters, which by its own > >> nature are discrete, and it is the job of the user to get the best > > use out > >> of it. > >> > >> Finally, I have done tens of thousands of optimizations, lost of > > them with > >> success, so its about making your own luck in this game. > >> > >> for example consider this statement > >> xyz = m1 * (MA(C, pds) > C) + (!m1) * (ma(c,pds) <= C); > >> where m1 is a control parameters that decides whether xyz = ma(c, > > pds) > C > >> or the other way around, and pds is the period of ma, as it stands > > it wont > >> be get much "luck" as you say. because, pds that is optimimum in the > > case of > >> > is probably very different than in the case of <=. > >> so by making xyz = m1 * (ma(c, pds1) > C) + (!m1) * (ma(c, pds2) <= > > C); and > >> optimize pds1, m1 and pds2 separately, you will get pds1 and pds2 > > gathering > >> around a cluster of value closer to its optiminum, and m1 has own > > value of 0 > >> or 1 which sort out what way is better. > >> > >> I hope this will be useful those who wants to use it. > >> > >> > >> _____ > >> > >> From: [email protected] [mailto:[EMAIL PROTECTED] > > On Behalf > >> Of Tomasz Janeczko > >> Sent: Wednesday, 1 October 2008 7:18 PM > >> To: [email protected] > >> Subject: Re: [amibroker] Re: CMAE behavior when optimizing control > >> parameters? > >> > >> > >> > >> Paul, > >> > >> I don't want to enter into yet another useless debate, but if you learn > >> about > >> *MATHEMATICAL* background of > >> Particle Swarm Optimizers you will > >> know that they are all designed to be used for CONTINUOUS parameter > > spaces. > >> > >> The fact that non-exhaustive methods like CMAE, PSO, etc *may* work > > in some > >> cases for discrete spaces > >> is more a question of luck and relative simplicity (or more or less > >> "smoothness") of the problem > >> being optimized than anything else. > >> > >> Best regards, > >> Tomasz Janeczko > >> amibroker.com > >> ----- Original Message ----- > >> From: "Paul Ho" <[EMAIL PROTECTED] <mailto:paul.tsho%40gmail.com> com> > >> To: <[EMAIL PROTECTED] <mailto:amibroker%40yahoogroups.com> ps.com> > >> Sent: Wednesday, October 01, 2008 11:03 AM > >> Subject: [amibroker] Re: CMAE behavior when optimizing control > > parameters? > >> > >> > Talking from personal experience - and I've been using intelligent > >> > Optimizers for quite a number of years optimizing combinations of > >> > continuous and "discrete" control parameters. Fred's IO has worked > >> > extremely well - in that I'm able to find optiminiums successfully, > >> > it may be a little more tricky, but not impossible. There are things > >> > that would help to IO work better. Nevertheless, I do have more > >> > problems with cmae with a lot of discrete parameters. But I suspect > >> > that's more to do with configuration of cmae rather than the ability > >> > of cmae itself. > >> > > >> > --- In [EMAIL PROTECTED] <mailto:amibroker%40yahoogroups.com> > > ps.com, > >> "Tomasz Janeczko" <groups@> > >> > wrote: > >> >> > >> >> No, CMAE, PSO and most other non-exhaustive methods > >> >> are best for continuous parameter spaces. Discrete spaces > >> >> where adjacent param values result in wild changes in fitness > >> >> tend to be very difficult to optimize in "intelligent" manner. > >> >> > >> >> Best regards, > >> >> Tomasz Janeczko > >> >> amibroker.com > >> >> ----- Original Message ----- > >> >> From: "Steve Davis" <_sdavis@> > >> >> To: <[EMAIL PROTECTED] <mailto:amibroker%40yahoogroups.com> ps.com> > >> >> Sent: Wednesday, October 01, 2008 1:19 AM > >> >> Subject: [amibroker] CMAE behavior when optimizing control > >> > parameters? > >> >> > >> >> > >> >> > Does anyone know if the CMAE algorithm can be used effectively to > >> >> > optimize a system containing control parameters? By this I mean > >> >> > optimizable parameters that do not measure a quantity, but are > >> > instead > >> >> > used to control the flow of execution of the program. In this > >> > sort of > >> >> > system, adjacent parameter values could result in wildly > >> > different > >> >> > system fitness. > >> >> > > >> >> > Thanks, > >> >> > Steve > >> >> > > >> >> > > >> >> > > >> >> > ------------------------------------ > >> >> > > >> >> > **** IMPORTANT **** > >> >> > This group is for the discussion between users only. > >> >> > This is *NOT* technical support channel. > >> >> > > >> >> > ********************* > >> >> > TO GET TECHNICAL SUPPORT from AmiBroker please send an e-mail > >> > directly to > >> >> > SUPPORT {at} amibroker.com > >> >> > ********************* > >> >> > > >> >> > For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG: > >> >> > http://www.amibroke <http://www.amibroker.com/devlog/> > > r.com/devlog/ > >> >> > > >> >> > For other support material please check also: > >> >> > http://www.amibroke <http://www.amibroker.com/support.html> > >> r.com/support.html > >> >> > > >> >> > ********************************* > >> >> > Yahoo! Groups Links > >> >> > > >> >> > > >> >> > > >> >> > >> > > >> > > >> > > >> > ------------------------------------ > >> > > >> > **** IMPORTANT **** > >> > This group is for the discussion between users only. > >> > This is *NOT* technical support channel. > >> > > >> > ********************* > >> > TO GET TECHNICAL SUPPORT from AmiBroker please send an e-mail > > directly to > >> > SUPPORT {at} amibroker.com > >> > ********************* > >> > > >> > For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG: > >> > http://www.amibroke <http://www.amibroker.com/devlog/> r.com/devlog/ > >> > > >> > For other support material please check also: > >> > http://www.amibroke <http://www.amibroker.com/support.html> > >> r.com/support.html > >> > > >> > ********************************* > >> > Yahoo! Groups Links > >> > > >> > > >> > > >> > > > > > > > > ------------------------------------ > > > > **** IMPORTANT **** > > This group is for the discussion between users only. > > This is *NOT* technical support channel. > > > > ********************* > > TO GET TECHNICAL SUPPORT from AmiBroker please send an e-mail directly to > > SUPPORT {at} amibroker.com > > ********************* > > > > For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG: > > http://www.amibroker.com/devlog/ > > > > For other support material please check also: > > http://www.amibroker.com/support.html > > > > ********************************* > > Yahoo! Groups Links > > > > > > >
