Thread: مفهوم مطابقة الأنماط EA

Originally Posted by ;

وماذا يمكن أن نستنتج من تجارتنا؟ على سبيل المثال ، اكتشفت خوارزمية أن اتباع الكلمةالشموع الخمسة R-O-N-A-L هي فرصة محتملة للحصول على candle'D '- كيف سيؤثر ذلك على قرار التداول الخاص بك؟ قد تكون المشكلة الأخرى أنك تستخدم مثل هذه الأبجدية الضخمة. بالنظر إلى المثال أعلاه ، عندما نفترض أن جميع الشموع مشتتة بشكل متساوٍ تقريبًا - فقد تكون فرصة الشمعة D الخاصة بك حوالي 0.4٪ ... لأننا ذكرنا ، كشفت الخوارزمية D على الأرجح في هذه الحالة ، قد تزداد الفرصة ربما إلى 1 ٪... مع ذلك...

يسمح استخدام الأبجدية حقًا بفرص تجارية أكبر من مجرد استخدام المبالغ المستقيمة. لا يوجد سوى 255 نوعًا من الشموع ، وبتخيل نمطين للشموع ، فإن احتمالات العثور على نمط لا يمكن تمييزه هي 1/65025. عند استخدام مقادير مباشرة ، هناك اختلافات لا نهائية نظريًا. يمكنك محاولة إنشاء هامش خطأ من نوع ما ، وعمل الأشياء بهذه الطريقة. على الأقل مع الزخرفة ، فهي تحمي هامش الخطأ هذا بالنسبة لي.

Originally Posted by ;

ما زلت لا أفهم سبب اهتمامك ...

المشكلة الرئيسية هي أن النظام يأخذ في الحسبان الثغرات بشكل كامل. باستخدام الطريقة التي ذكرتها سابقًا ، توجد فجوة مماثلة للشمعة. بالطريقة التي طلبت بها الكود ، يجب أن أكون قادرًا على إضافة المزيد من الأبعادالمعلمات إلى الثلاثة التي لدي بالفعل ما لم أساء فهم شيء ما في حساب التفاضل والتكامل. في الوقت الحالي ، العلاقة الوحيدة بين الشموع التي أشعر بالقلق حيالها هي ما إذا كانت الشمعة ”R” تأتي بعد شمعة ”P” بشكل متكرر بدلاً من شمعة ”Q”. في التداول الحي ، نادرًا ما نحدث الفجوة على أي حال ، لذلك لا أزعجني.

Originally Posted by ;

[edit] You could change the supply of this 255'feature slots' so 1 feature is more weighted compared to another.

I was considering the exact same thing. You can count the number of different values of Top - Open, Close - Low and Close - Open occured since 2000. You may possibly get something like this: High - Open - 50 different values Close - Low - 50 different values Close - Open - 100 different values In this example you may assign can. Half of the'feature slots' to Close - Open.

Will pay attention.

Thanks to April Offer You Cannot Refuse. I deny (but it is really really hard)! Subscribe. RR

Ronald, allow me to know if I am off base... Patterns have to be generalizations, so that they could be skewed by a certain percentage but still fit in the pattern... so that if a pubs' range was 10 pips or 100 pips, it could still fit within a pattern. Can it be any use to agorize a bar by its range? An illustration... a bar range is always 100%. Therefore, the open will be at 50 percent of its range (center of the bar), and also the close could be at 75% (3/4 of this bar) of its range. Then you could classify the low of the current bar to the low of the previous bar as lt;100%, low of previous bar within the range of the current bar, etc.. I am unsure now the best way to handle when the previous bar is part of engulfing pattern... Just throwing an idea... Carl

Originally Posted by ;

Ronald, allow me to know if I am off base... Patterns have to be generalizations, so they could be skewed by a particular percentage but still fit in the pattern... so that if a bars' range was 10 pips or 100 pips, it could still fit within a pattern. Can it be any use to agorize a bar by its range? An example... a bar range is always 100%....

Hey Carl, Your strategy identifies bars based on their comparative form. I am not sure if you intended for this circumstance, but let us assume we have. When looking at a chart, do you classify them? I am classifying bars based in their absolute contour, it would be simpler to distinguish between patterns in which the price is ranging and patterns where the price is trending. Ideally, I would rather just write a neural network that could take into account all of this information. 2 layer is sufficient to think of every constant mathematical function, however that I suspect adding more nodes and layers will allow a conditional step function to be created by the NN. It is similar to our trading thinking. . -If the market price is moving sideways use this mindset -If the market is moving upward, use this mindset -If there is undefinable chaos, use a different mindset. That's the hope anyway

Originally Posted by ;

Ideally, I would rather just write a neural network that could take into account all this data. 2 layer is sufficient to come up with every possible continuous function, but I suspect adding more nodes and layers will make it possible for the NN to produce a step function that is conditional. It's similar to our own trading believing. .

Hi Ronald, Are you writing a neural network based on FANN? Can you share a NN? If NN is based on FANN I expect
http://articles.mql4.com/868 may be useful.

Originally Posted by ;

.. .This is a newer concept EA which is programmed to find its candle patterns. The pattern finding algorithm is simple: it assesses the previous x closed bars and generates a string something like LLDSSL that signifies that the bars from right (most recent) to left were long, short, brief, doji, long. The EA then scans the chart history looking for a pattern of pubs with this specific same arrangement. The EA checks x pubs after the sequence to see if this bar was long or short. The outcomes are subsequently added to the statistics generated by the EA...

Hey there Ronald... This reminds me of something I did a number of years ago for equities. My set-up was basically a generalization of the exact same thought you are working with. The way I generalized the strategy was that rather than attempting to egorize candles into traditional patterns (long, doji, brief, etc.) then match sequences, I simply generated sequences of percentage-based differentials for each of OHLC on every bar, when compared with the previous bar. By expressing each of these values from bar to bar in terms of similarity to the previous bar and then placing a particular number of pubs to examine, I ended-up using one mathematically derived value that expressed that the total similarity of all of OHLC to get a fixed number of bars to the current condition. Then I would check the following bars in the near historical match to see if it was indiive of a fantastic move and to get an immediate term sign to trade according to the previous matching pattern. Anyway, I just thought I would mention this as this sort of set-up it might be something to grow your pattern recognition to provide a generalization effect. (To elaborate, my machine did something like this, for the past n pubs being examined: open_2percent = (open_2-open_1)/open_1 open_3percent = (open_3-open_2)/open_2 open_npercent = (open_n-open_n-1)/open_n-1 This has been done also for n pubs to get large, low, and near. The next step was to find and sum the differences between every bar position from the historical data to the same offset bar within the current pubs (with exactly the exact same bar to bar percentages computed for the current bars.) The sum of the amounts of those differences for the past and current OHLC will be taken. Then it was a simple thing to rate the final result -- the closer the amount was to zero, with each of OHLC results being close to zero similarity between the current bars and the historical bars. Once similarity has been decided, big moves either up or down at the previous data were of interest as the current conditions might be expected to act in a similar way.) Anyway, I thought I would mention that, as you might have the ability to create your pattern matching more abstract and less determined by human-recognizable structures using a set-up like that. Theo Buitendyk.

The, can you tell us the result of your implementation?

I wonder, what was the consequence of this method's implementation? The idea sounds really interesting. .