pattern matching | Page 5 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Account Matching or Prop Trading 3 replies

Pattern Matching Concept EA 61 replies

Two matching highs trend breakout? 3 replies

Matching metatrader times to alerts 0 replies

Oanda's NOT a matching engine...? 4 replies

Attachments: pattern matching
Exit Attachments

pattern matching

  • #81
  • Edited 1:36pm Jun 13, 2010 1:23pm | Edited 1:36pm
Quoting finimej
So my question is: can that pattern finder used to find the pattern on indicators as well as it find in the price pattern?
Not in its current form, there would be modifications necessary to read the indicator but in theory it should be possible. Unfortunately I cannot do this at the moment because I have put this whole pattern matching project on pause because of lack of time and some other things that have much higher priority at the moment.

BTW: I'm not totally sure whether the screenshot with the indicator you posted is actually a good candidate for this kind of approach, I would try to identify these simple patterns in a more conventional manner, especially in renko charts where the number of possible combinations of bars during a certain time is very limited.

The project itself is actually not half finished, it is in a state where it can be used (if supplied with a good pattern function) and the minimum needed features that are included are complete and working and extensively debugged. It can be used as a starting point or inspiration for further research.
  • #82
  • Edited 1:50pm Jun 13, 2010 1:36pm | Edited 1:50pm
thanks 7bit for your patiently answers.
1) I find the fucntion pf code. I also read through your other thread on make dll for mt4 using lazarus. It is a real pleasure to read your clean and structured code.

2) ok, put the indicator pattern maching to side. Just use the pure price pattern matching on the renko chart shall do. For exemple, take the same picture shown on the previous thread, I shall have lots of ABC,length 0toA=BC, (some call this ABCD) marked with red line. Take me a while to get it start this pattern code. It is better at this moment, I am your tester only. If you show me how to make the AB=BC this simple pattern in the code, I will then understand where to edit your function code and start to make other patterns for matching. Thanks so much for inspiring.

http://www.formulatrader.com/download/schaff2.gif
  • #83
  • Edited 5:44am Jun 18, 2010 5:09am | Edited 5:44am
7Bit,
Say if I understand your code correctly.

In the current code, you have one pattern defined in the pf funciton, then in the error function, you compare each bar scaled X(bar number =time),Y (price) and check how many bars that does not match the pf xy function. This is hardcoded inside the error funtion.
Inserted Code
 for i := 0 to width do
  begin
    err += (abs(scale[I_OFFSET] + fact_y * pf(fact_x * i, par) - history[end_i - i].Close));
  end;

so if the EA have multipule patterns such as bearish1, bearish2, bullish1, bullish2

then the error funtion need to search the against this patterns
Inserted Code
 for i := 0 to width do
  begin
    err += (abs(scale[I_OFFSET] + fact_y * bearish1(fact_x * i, par) - history[end_i - i].Close));
  end;

I wonder if you can update your code, so that error funtion will takin the pf(fact_x * i, par) as input parameter, and I can call the error function to compare any patterns, not just pf hardcoded.

Correct me, if I am wrong. I just newly gettings into your way of thinking for pattern matching.

-------------------------------------
After thinking it, the error function shall NOT be touched.
if I have more patterns to match, I just put it inside the pf function, with case switch. right?

why I think this way, because current ea only matching one single pattern, and made for the purpose to find various patterns that useful. However, for an EA to be able to become even profitable, it at least need to find the several patterns for the turning points.
  • #84
  • Jun 18, 2010 9:02am Jun 18, 2010 9:02am
This is fun. played with py. and will check out later how it is matching in ea. but maybe we can scale each tuning point instead? as long as shape of 12345 wave is ok, then length wave 3> wave1 , dose not matter how long the wave shall be?
Attached Image (click to enlarge)
Click to Enlarge Name: fun.jpg Size: 57 KB
  • #85
  • Jul 4, 2010 3:29pm Jul 4, 2010 3:29pm
Nice thread

Any updates ?
  • #86
  • Jul 5, 2010 3:25am Jul 5, 2010 3:25am
Quoting finimej
7Bit,
After thinking it, the error function shall NOT be touched.
if I have more patterns to match, I just put it inside the pf function, with case switch. right?
This would be one possibility, the other one (faster) would be to let a function pointer point directly to the actual pattern function (or the same thing implemented in a cleaner way: make it a class and pf() a virtual method and let subclasses override it). Currently pf() is declared inline, this means a complete copy of the function is directly inserted inside the loop in the error function (and in other places where it is called) to save a function call and a few clock cycles (push/pop the stack etc.) in every loop iteration but I have not done any benchmarking. It would probably not make a big difference to make it an ordinary function call.

The approach with the switch in pf() (this would be a case statement in Pascal (Note that the case in Pascal behaves much more natural and intuitive than the switch in C, there is no break needed)) is probably the easiest method. You could use one of the unused parameters that are there for parameterizing the pattern function without making any changes to the rest of the dll source code.
  • #87
  • Oct 15, 2010 4:30am Oct 15, 2010 4:30am
I have been reading through the thread and is very interesting. I have many patterns that I use on a daily basis and wondered if any more work has been done in advancing this pattern matching concept and possibly the logging of pattern statistics.

Cheers =)
  • #88
  • Oct 30, 2010 9:48am Oct 30, 2010 9:48am
hi guys, do you find some profitable pattern?
  • #89
  • Nov 2, 2010 2:14pm Nov 2, 2010 2:14pm
This formula appears to be our exact same patent-pending formula (minus additional filters) which was released two years ago: www.modulusfe.com/apr/

A similar patent (but different from ours) is pending by TT.
  • #90
  • Nov 3, 2010 2:21pm Nov 3, 2010 2:21pm
The algorithm was taken from our patent-pending APR product.

We are requesting FF to contact us but have not heard back.
  • #91
  • Nov 4, 2010 9:55am Nov 4, 2010 9:55am
Quoting modulusfe
The algorithm was taken from our patent-pending APR product.

We are requesting FF to contact us but have not heard back.
oh really ??
100 Fold Challenge->Interested? ->https://www.forexfactory.com/thread/32152
  • #92
  • Nov 4, 2010 11:00am Nov 4, 2010 11:00am
Quoting modulusfe
The algorithm was taken from our patent-pending APR product.

We are requesting FF to contact us but have not heard back.
They are busy making millions with it, please try later.
  • #93
  • Edited 11:57am Nov 4, 2010 11:36am | Edited 11:57am
Quoting modulusfe
The algorithm was taken from our patent-pending APR product.

We are requesting FF to contact us but have not heard back.
LOL!

You are joking. Algorithms cannot be patented.

And this one is surely not "taken" from somewhere else (instead it seems you have stolen your "patent" directly from good old Archimedes himself). This algorithm was taken right out of my memory and from the basic math knowledge that every kid learns at school at the age of ~15. Are you going to sue them all?

This patent bullshit has seriously come out of control. The only way to deal with these wannabe patent trolls who suddenly appear out of nowhere with some ridiculous claims more than half a year after "prior art" is already published (and this one is not even patent-worthy because it is so simple and obvious) is to simply ignore them.

And now try to sue me for my patented math knowledge if you can afford it. Also sue my old school where I learned how to calculate an integral to get the area below a curve and also sue all my teachers (at least all those who are still alive today). LOL.
  • #94
  • Nov 5, 2010 6:26pm Nov 5, 2010 6:26pm
What a world, What a world. They must not be selling to many of them.
  • #95
  • Nov 17, 2010 5:10am Nov 17, 2010 5:10am
What's more amusing is that 7bit managed to apply simple arithmatic to make this tool freely available on a freely available platform, with an infanitly smaller amount of resources, including time. Something tells me their business model needs some work.

Seems like they should be hiring you, 7bit. But please see his profile before you make an offer.

Oh yes, I make a claim on the recruitment fees (20% of total package) if this happens, as I suggested it. (See the irony).
  • #96
  • Nov 17, 2010 5:17am Nov 17, 2010 5:17am
Neat 7bit
  • #97
  • Jun 2, 2011 10:21am Jun 2, 2011 10:21am
Hello, I have created a short 3min video that looks at time patterns. In the video is an explanation looking for a time series to look for patterns. Does anyone have an idea of how to do what I have shown?
Thanks!
  • #98
  • Last Post: Sep 4, 2020 11:45am Sep 4, 2020 11:45am
Quoting 7bit
To not further delay the search for patterns I will post my current versions of the dll and the EA. Both have been rewritten to get rid of all globals variables in the dll, so it can be used on more than one pair at the same time and the code has been cleaned up, made understandable and commented as good as i can. The patttern function is now renamed to pf() and is found in a separate unit. It is still the same old dummy function from last time, don't expect it to be profitable. the zip file contains the dll source code, open the .lpi file with...
Can anyone compile the files please?They are giving errors.
0 traders viewing now