Tomasz, Thanks for this great step forward in basic capability of AmiBroker. This is exciting. You have indeed made it very simple to explore more optimization alternatives.
Along the lines of progster's suggestions, I would like to see the following: * First, I need some way of persisting the best results to be used by the script in an indicator or elsewhere. Some alternatives: ** Manually copying and pasting each value is OK up to about 3 variables once a day. Call me lazy but I have done that a few hundred times, and I am tired of it. It is error prone and makes me feel like a slave to my computer. This should be automatable by some means. ** IO modifies the script to write the best results as the default values of the Optimize calls. This has its drawbacks, however, since you can't use anything but simple Optimize calls - no expressions. ** Something like progster's suggestion, which allows backtesting with a specific row of the optimization results, but enhanced to make the values persistent so they would be used in indicators and elsewhere. ** Some specialized use of static vars to store a set of values, perhaps in a comma-separated-values list. Scripts would have to read the same static vars and set the corresponding variables, except during optimization where these would be overridden, of course. ** I need a separate set of variable values on a per-symbol or per-watchlist basis, by the way. Each symbol or collection of symbols has its own market-driven behavior, and at least for my system, it is best to tune the variables to the symbol. * Second need: I need some way to seed the optimizations from day to day, or week to week (or whatever your frequency of reoptimizing is). Market behavior changes, so systems should change as well. Since I can't even do one optimization run sufficiently well given limited time, and the fact that I need to re-run it for each symbol, there is no risk of over-optimizing. It should save a lot of time if the best results from the previous 30 days of runs, for example, could be used as starting points for the next run rather than starting from a random selection each time. * Third: I'd like a scriptable interface to the optimizer plugin. I expect that most of the time of running optimizations is the individual tests, not the overhead of figuring out what to test. If so, then it would be reasonable to support a scripting interface to the plugin rather than just the C/C++ interface. I would write in JScript. Writing an optimization controller in AFL would also be handy for more people, but the language is rather cumbersome for more traditional styles of programming (arrays of non-price data, objects composed of several properties, etc). ** Given a scriptable interface, and given access to the current optimization engines (assuming I could call them from my script in turn), I could probably build my own way of persisting the best result values. I couldn't provide the initial seed values without some cooperation from the engines, however. Daniel LaLiberte --- In [email protected], "progster01" <[EMAIL PROTECTED]> wrote: > > --- In [email protected], "Tomasz Janeczko" <groups@> wrote: > > > > Press Cancel. All backtests upto date will be available in the > result list. > > You can sort/ export the results list from AA window. > > > > Tomasz, > > Congratulations on all the new optimization capabilities you've > recently added to AB, and "Thank You!". I think this is great work, > really enhancing the product in several important ways. > > Re: "The EASIEST way to use" ... > > If at all possible, I think it would be _very_ useful if one of the > right-click options on optimization results would load and re-run the > selected optimization iteration in the background, creating the trade > lists for that run and doing any other necessary preparation for trade > arrows to be shown. > > Being able to view the trades taken by multiple optimizations visually > without having to go through the effort of manually typing in the > parameters for each one would be a major timesaver and workflow > enhancer here in the field. > > A further helpful extension would be if this capability were combined > with a way to persist (reload) optimization results for re-examination > and re-visualization at any future point. > > Thanks again for all the great new work! > > - Progster >
