The "secret recipe" for Median is not secret - it is available in Numerical Recipes book.
Best regards, Tomasz Janeczko amibroker.com ----- Original Message ----- From: "brian_z111" <[email protected]> To: <[email protected]> Sent: Friday, June 05, 2009 3:51 PM Subject: [amibroker] Re: Newbie Array / Looping Question Z I'll leave it for a while too. I think you were right though, when you indicated that looping is required and it probably doesn't get any better than what you have already done .... pity we don't know Tomasz's secret algorithmic recipe for Median ... I guess it is doing looping in the background ... if he told us we would have to die. Thanks for the code ... pretty neat. Anyway, I'm on a two year plan with that stuff. --- In [email protected], i cs <ics4...@...> wrote: > > Hi, > > One way of doing it might be to run the analysis to a "synthetic > stock" - eg ~~~CBA for CBA and bring it in with a foreign() > when required or perhaps an ODBC call. > > Anyway, I might leave it on the backburner for now and keep exploring > the rest of AB. Still lots to have a look at! > > Z > > > > > ________________________________ > From: brian_z111 <brian_z...@...> > To: [email protected] > Sent: Friday, 5 June, 2009 7:13:39 PM > Subject: [amibroker] Re: Newbie Array / Looping Question > > > > > > Yes, AB is written in C++. > > I am only up to AFL101 but I have been around the AB environment for 4 years. > > The ABDevelopmentKit might be the place to find out more about what is under > the bonnet. > > > I'm keener on your suggestion regarding using the arithmetic > > now that I now that MA(array, array) works! And yes, > > I'd prefer the faster option every time! > > If you are used to programming but not used to AB's array processing it might > take a bit of a mental adjustment .. a lot of > programmers seem to find that.... it is all done at the machine level in AB > speed ... there is a post from Tomasz at the KB about > the technical details of part of the process (it went over my head but I get > the concept). > > It's a piece of cake except that StDev is limited to a constant number also > so you will have to calc your own StDev (also note > that StDev in AB is StDevP by default).... . I started on the AFL for Skew > until I hit the StDev wall and I had to leave it for a > while ... I'll be back on it later or tomorrow.... the first problem I found > is that you will need to run a lot of datapoints > before you get a meaningful distribution (pseudo dist in this case) i.e. I > used GF and they are all different at the default > number of decimal points .. I used Prec() to round down to 2 integers to to > test the code I started to write ... in the real world > you will need N to be very big to start to find the meaningful mode ... I > could use Prec(tradeseries, 3) and that would be the > equivalent of 1000 bins so you need quite a few N to get a decent bin count > ... if you round down to 2 decimal places perhaps the > end result won't be that meaningful. > > <snip>But I also like to think that I'll find other, and perhaps more > meaningful "profile attributes" which have nothing to do > with the > old distributions! That's why I've been looking for a > generic solution<snip> > > That's the way to go ... I never fail to learn something when I dig in. > Our analytical approach seems to be synched to some degree ... now that you > have raised the subject I am trying to relate Skew etc > to my BinomialModels ... I expect a few new metrics to come out of that but > they are a fair way away. > It is always hard to say, in advance, if the new 'profile attributes' will be > useful on a day to day basis but I find the > understanding gained from thrashing it all out is worth the effort .. > somewhere in the back of my mind I use it all at some stage. > > Once I got that far I started to wonder if the progressive Skew, Median etc > would be that meaningful ... a rolling period might be > just as effective ... still if you are like me you need to do it and look at > it to see if it meaningful or not. > > > The following construct works - but it is very, very > > slow..... > > Yep! > > --- In amibro...@yahoogrou ps.com, i cs <ics4mer@ > wrote: > > > > > Hi Brian, > > > > Hmmmm... could be a typo in the manual > > I certainly had no idea that MA() could do this. > > > > myX = MA(C, 1); // AFL happy > > myX = Median(C, 1); // AFL happy > > myX = MA(C, Cum(1)); // AFL happy > > myY = Median(C, Cum(1)); // "Argument has incorrect type" > > > > > > You must have a huge amout of hands on with AFL! > > Some of the functions in AFL take either arrays, or > > numbers, and a few seem to be able to take both: > > Not sure what language AB is written in suspect its > > probably C++ which allows you to do this sort of > > thing. > > > > Looking at the entry for "int()" > > > > INT > > âË?' integer part > > SYNTAX int( NUMBER ) > > int( ARRAY) > > > > I suspect the manual entry for MA() should be the > > same! > > > > I'm currently looking at the "_TRACE()" command to try > > and garner some insight into the execution process. > > > > I'm keener on your suggestion regarding using the arithmetic > > now that I now that MA(array, array) works! And yes, > > I'd prefer the faster option every time! But I also like to > > think that I'll find other, and perhaps more meaningful > > "profile attributes" which have nothing to do with the > > old distributions! That's why I've been looking for a > > generic solution > > > > The R package looks intriging too. But I still have to > > learn how AB works in it's basic mode before I > > unduly complicate my life with other packages etc... > > > > The following construct works - but it is very, very > > slow..... > > > > function myMedian( array, i ) > > { > > x = array[i]; > > return x; > > } > > > > mp2 = 1; > > for( i = 1; i < BarCount; i++ ) > > { > > mp2[ i ] = myMedian(Median( C,i),i); > > } > > > > > > Thanks for your help once again! > > > > Robert Z > > > > > > > > > > ____________ _________ _________ __ > > From: brian_z111 <brian_z111@ ...> > > To: amibro...@yahoogrou ps.com > > Sent: Friday, 5 June, 2009 5:00:41 PM > > Subject: [amibroker] Re: Newbie Array / Looping Question > > > > > > > > > > > > <snip>Median( ) wants a number as its second argument. (The manual is > > clear!)<snip> > > > > To help me understand next time ... how did you decide that Median)array, > > period) wants a number from what is written in the > > manual? > > > > MA(array,periods) ; accepts an array and the only differential in the > > manual is plurality of the periods (do you think Tomasz > > intended the plurality to be significant? ) > > > > MA(C,Cum(1)) ;//this is OK? > > > > <snip> When are the "for" loops executed in the block of code? Knowing some > > more about the AFL execution order might be > > helpful.<snip> > > > > Sorry, don't know anything about execution. > > > > Recently Siddhartha said that learning the AFL execution was the best thing > > we can ever do ... I don't know where to find it > > though. > > > > You are not keen on my suggestion that calcs for the moments are arithmetic > > and so should fly in AFL without looping? > > I am a lay programmer but my instinct is to start there - looping will slow > > you down in RT indicators?? ? > > > > Plan C - use the RMath Plugin ... advanced stats available there. > > > > --- In amibro...@yahoogrou ps.com, i cs <ics4mer@ > wrote: > > > > > > Hi Brian, > > > > > > Thanks for getting back to me. > > > > > > Unfortunately, Cum() returns array, Median() wants a number as its > > > second argument. (The manual is clear!) > > > > > > I used median as an example because it's relatively easy to understand, > > > and I will be using it, but I will also be plotting the growth factor in > > > the same > > > way - as well as the other moments of distribution - just one step at a > > > time. > > > > > > Its possible to derive a lot of the moments of distribution from other the > > > parts, but it comes back to the same problem - I'd have to write functions > > > to derive kurtosis, mode etc. > > > > > > Some of the historic emails in the group included the following solutions: > > > - code the functions myself - which I'm trying to avoid. > > > - create a dynamic set of variables using VarSet - this seems a bit > > > brutish > > > especially if you are playing with a number of "profile attributes". > > > > > > When are the "for" loops executed in the block of code? Knowing some more > > > about the AFL execution order might be helpful. Might go down that track > > > for > > > a while....and check out the other 1500 references to "loop" in the email > > > group. > > > > > > Thanks again... > > > Robert Z > > > > > > > > > > > > > > > > > > ____________ _________ _________ __ > > > From: brian_z111 <brian_z111@ ...> > > > To: amibro...@yahoogrou ps.com > > > Sent: Friday, 5 June, 2009 2:57:43 PM > > > Subject: [amibroker] Re: Newbie Array / Looping Question > > > > > > > > > > > > > > > > > > I am not sure about this one now ... I think we need some help from the > > > code/maths experts. > > > > > > Median() doesn't appear to accept an array as input (the manual doesn't > > > say one way or another?) ... it might be demanding a > > > constant (that wouldn't surprize me since mean/mode/median are moments of > > > a distribution so massive calcs might be required to > > > find them for skewed dists without having the distriution at hand) ... AB > > > doesn't do that distributions off the shelf ... not > > > AFAIK. > > > > > > I think you can calc Skewness from N, StDev, arithmetic mean etc ... once > > > you have S you can back calc the Median or the Mode > > > from the Mean and the StDev. > > > > > > Another brute method might be an algorithmic trial and error test for the > > > Mode (loops required)... I forget what it is called > > > in programming but if the value to the right of the mean has a higher > > > frequency than the mean value the mode is somewhere to > > > the right .. by halving the range between the freq test you will zero in > > > on the mode and then you can calc the median by using > > > the mean and mode values..... AFAIK array processing will always be more > > > timely than looping though. > > > > > > --- In amibro...@yahoogrou ps.com, "brian_z111" <brian_z111@ ...> wrote: > > > > > > > > Median(array, periods) > > > > > > > > I assume you are measuring, say ROC(C,1) or similar for % price change. > > > > I guess you just need to get a progressive count, of the number of ROC > > > > datapoints, to make the function work. > > > > > > > > Things to look out for (if the AB function is going to work for you): > > > > > > > > - every element in the array you are measuring needs a value (nulls or > > > > zeros might trip up the math e.g. STDev, mean) ... I > > > > guess if you are interested in daily bars you will first create a daily > > > > ROC(C,1) or ROC of some other point of interest from > > > > within the bar HL range > > > > - the number of periods, in your array count, is range dependent i.e. > > > > in AB it can vary with mode ... QuickAFL can be used > > > > to change the range (in AA) and QickAFL will autoset the range in > > > > charts (or might ... I am not the full bottle on QuickAFL > > > > but there is an article by Tomasz in the AB KnowledgeBase) > > > > - BarsSince type functions might count one less period than you expect > > > > ... depending if the function is a zerobased count > > > > and how you want to use the count.... sometimes I have to add 1 to make > > > > the adjustment manually > > > > - BarIndex() is the bar by bar count since the start of the range, > > > > where range == the range of the database ... AB gets a > > > > bit tricky with when and where it uses the complete database range as > > > > the default .... probably to do with fast array > > > > processing etc ...sometimes I prefer to use Cum(1) as my count ... I > > > > seem to get more visually stable indicators in the > > > > charts because it doesn't reference the chart range it references from > > > > the start of the database. > > > > > > > > - so, making a few assumptions, and if Median works for you as > > > > required, Median(ROC(C, 1),Cum(1) ); // because Cum(1) is > > > > progressive and counts bars from the beginning > > > > > > > > - mathematically I find ROC is not as stable as GrowthFactor where ROC > > > > == 3% is expressed as 1.03 > > > > > > > > Haven't done it though. > > > > > > > > If that doesn't work maybe someone else in the forum knows more. > > > > > > > > In case of emergency RVince book two has the maths for median, mode, > > > > mean interelationships (possibly for advanced, or > > > > custom use, you could calculate the median from other values).... mean > > > > == median == mode for normal distributions (you > > > > should be so lucky!) > > > > > > > > > > > > > > > > > > > > --- In amibro...@yahoogrou ps.com, "ics4mer" <ics4mer@> wrote: > > > > > > > > > > Hi all, > > > > > > > > > > I'm after a general form for doing a particular type of algorithm > > > > > in AFL - and I'm happy to admit that I might be going about it > > > > > the totally wrong way... > > > > > > > > > > I am attempting to extract an evolving "profile" of a stock over > > > > > time. > > > > > > > > > > Using "median" as an example, lets say I want calculate the median > > > > > percentage price change for the life of the stock. To give an > > > > > example: > > > > > > > > > > The 1st trading day - do nothing. > > > > > The 2nd trading day - calc median for 2 days > > > > > The 3rd trading day - calc median for 3 days > > > > > Last traded day - calc median for n days. > > > > > > > > > > For simple functions, I have used a construct similar to; > > > > > for ( i=0; i < BarCount; i++ ){ > > > > > myArray[ i ] = simpleFunc( i ); > > > > > ..... > > > > > } > > > > > > > > > > However, in this case, I would have to write a new median function, > > > > > and this is irritating, since there is a perfectly good one > > > > > taunting me from the users manual! That's why I suspect that I > > > > > might be doing something in the wrong way. > > > > > > > > > > Any help or pointers would be much appreciated. A pointer to > > > > > a message in the user group would be fine too! ( Yes I am still > > > > > trawling through the email group as I write this....) > > > > > > > > > > TIA > > > > > > > > > > Robert Z > > > > > > > > > > > > > > > > > > > > > > > > > > > Need a Holiday? Win a $10,000 Holiday of your choice. Enter > > > now.http://us. lrd.yahoo. com/_ylc= X3oDMTJxN2x2ZmNp > > > BF9zAzIwMjM2MTY2 MTMEdG1fZG1lY2gD VGV4dCBMaW5rBHRt X2xuawNVMTEwMzk3 > > > NwR0bV9uZXQDWWFo b28hBHRtX3BvcwN0 YWdsaW5lBHRtX3Bw > > > dHkDYXVueg- -/SIG=14600t3ni/ **http%3A/ /au.rd.yahoo. com/mail/ > > > tagline/creative holidays/ *http%3A/ /au.docs. yahoo.com/ > > > homepageset/ %3Fp1=other% 26p2=au%26p3= mailtagline > > > > > > > > > > > > > > > Need a Holiday? Win a $10,000 Holiday of your choice. Enter > > now.http://us. lrd.yahoo. com/_ylc= X3oDMTJxN2x2ZmNp > > BF9zAzIwMjM2MTY2 MTMEdG1fZG1lY2gD VGV4dCBMaW5rBHRt X2xuawNVMTEwMzk3 > > NwR0bV9uZXQDWWFo b28hBHRtX3BvcwN0 YWdsaW5lBHRtX3Bw > > dHkDYXVueg- -/SIG=14600t3ni/ **http%3A/ /au.rd.yahoo. com/mail/ > > tagline/creative holidays/ *http%3A/ /au.docs. yahoo.com/ > > homepageset/ %3Fp1=other% 26p2=au%26p3= mailtagline > > > > > > > > Need a Holiday? Win a $10,000 Holiday of your choice. Enter > now.http://us.lrd.yahoo.com/_ylc=X3oDMTJxN2x2ZmNpBF9zAzIwMjM2MTY2MTMEdG1fZG1lY2gDVGV4dCBMaW5rBHRtX2xuawNVMTEwMzk3NwR0bV9uZXQDWWFob28hBHRtX3BvcwN0YWdsaW5lBHRtX3BwdHkDYXVueg--/SIG=14600t3ni/**http%3A//au.rd.yahoo.com/mail/tagline/creativeholidays/*http%3A//au.docs.yahoo.com/homepageset/%3Fp1=other%26p2=au%26p3=mailtagline > ------------------------------------ **** IMPORTANT PLEASE READ **** This group is for the discussion between users only. This is *NOT* technical support channel. TO GET TECHNICAL SUPPORT send an e-mail directly to SUPPORT {at} amibroker.com TO SUBMIT SUGGESTIONS please use FEEDBACK CENTER at http://www.amibroker.com/feedback/ (submissions sent via other channels won't be considered) For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG: http://www.amibroker.com/devlog/ Yahoo! Groups Links
