Pattern Matching Concept EA

There hasn't been much activity on the "Old Dogs with New Tricks" thread save the posts which have begun to detract from the original topic. I have noticed in the tick collection logs that there were 110 unique users this week, so I guess these people are making money.

This is a newer concept EA which is programmed to find its own candle patterns. The pattern finding algorithm is simple: it checks the previous x closed bars and generates a string something like "LLDSSL" that means the bars from right (most recent) to left were long, short, short, doji, long, long.

The EA then scans the chart history looking for a pattern of bars with that exact same sequence. The EA checks x bars after the sequence to see if that bar was long or short. The results are then added to the statistics generated by the EA.

If the likelihood of an event happening is greater than the threshold, the EA will take the trade accordingly.


Settings:

 

1. extern string Remark1 = "== Main Settings ==";
2. extern int MagicNumber = 0;
   1. This number should be unique when multiple EAs are attached to a terminal.

3. extern bool SignalsOnly = False;
   1. If set to true, the EA will only alert you to a trade using any combination of the alerts below.

4. extern bool Alerts = False;
   1. If set to true, the EA will send a pop-up alert when a trading signal arrives.

5. extern bool SignalMail = False;
   1. If set to true, the EA will send you an e-mail (assuming you set up your mail in your MT4 Terminal.) when a trading signal arrives.

6. extern bool PlaySounds = False;
   1. If set to true, the EA will play the file, "alerts.wav" when a trading signal arrives.

7. extern bool ECNBroker = False;
   1. Some brokers require EAs to first place a trade, and then modify the trade with the stoploss and takeprofit. If your broker requires this, set this value to True.

8. extern bool EachTickMode = True;
   1. If set to true, the EA will take a trade immediately if it sees a signal. If set to false, the EA will take a trade only if there is a signal at the end of the bar.

9. extern double Lots = 0;
   1. Fixed lot size

10. extern bool MoneyManagement = False;
    1. Enables/Disables the lot sizing function.

11. extern int Risk = 0;
    1. The percent of your available margin that you are willing to put into a trade.

12. extern int Slippage = 5;
    1. Maximum slippage allowed when placing an order.

13. extern int StopLoss = 100;
    1. Hard Stoploss in pips.

14. extern int TakeProfit = 60;
    1. Hard take profit in pips.

15. extern string Remark2 = "== Pattern Settings ==";
16. extern int NumberOfBars = 4;
    1. The number of bars to use when generating a pattern.

17. extern int PredictionBar = 1;
    1. The number of bars after a pattern string to check a prediction.

18. extern int HistoryBars = 500;
    1. The number of bars to check when searching for matching pattern strings. Set to 0 to use all bars.

19. extern int TimeLimit = 10;
    1. The maximum number of seconds after the current bar opens to place a trade.

20. extern int BarsClose = 1;
    1. The number of bars after a trade is opened to close a trade.

21. extern double Threshold = 50.0;
    1. The minimum percentage required to place a trade.

22. extern bool ReverseTrades = True;
    1. Changes long signals to short signals and short signals to long signals.

Note: This EA is highly experimental and has not yet been proven as profitable. Use at your own risk.

nice concept Ron, it can be extended to some complex pattern.

after tried it, i see you should be carefull with this line:
double LongProbability = LongFound / TotalPattern * 100;
when TotalPattern = 0, zero divide error.
i suggest to change it to
double LongProbability = LongFound / MathMax(1,TotalPattern) * 100;

regards,

Problem with doing that is it might skew the value of the long threshold when applied in the tradetrigger lines further down.

it won't affect the value of LongProbs.

LongValue will always less or equal to TotalPattern.
when TotalPattern=0, LongValue will be 0. LongProbs = LongValue/1 = 0/1. Zero divide avoided.

when TotalPattern =1, LongProbs = LongValue/(MathMax(1,TotalPattern) = LongValue/MathMax(1,1)=LongValue/1.
when TotalPattern =2, LongProbs = LongValue/MathMax(1,TotalPattern)=LongValue/MathMax(1,2)=LongValue/2;
so, for TotalPattern greater than 0(1,2,3..), the value of MathMax(1,TotalPattern) will return TotalPattern. No skew occured.

I get it. Thanks for the clarification Sangmane!

Non-zero divide version attached.

What time frame is best for this EA?

In theory, all of them.

In reality, 5M+

hi roland, the settings of the EA are good basic ones? or do I change it? thanks.

I should note that I haven't found any consistently profitable settings for this EA. I suggest you set the threshold really high (70+) and maybe the numberofbars relatively high (5-15).

hi Ron,
nice project, the idea is great but i have a general question: what is a bar exactly ? Stupid question at first glance but i think that the MT4 bars might be normalized:

- MT4 normal bar are time dependant, ie M1...D1..., so if you consider a bar as "Long" for example, the stats created by the bot will have 5 pips bar for the smallest and maybe 50 pips or else for an other. So it's difficult to only say 1Long=1Long, that mean nothing to my point of view.

So, what solutions ?
when following prices, there is trend, then consolidation, retrace, then trend and so on. Are you considering small retraces as Short or integrate them into the bigest move ?

- so constant bar value (10 or 20 pips or else) is more representative than time based bar
- how to analize (open-close) vs (high-low), and (high-low)/(open-close) ?

Really need to define a "bar" before building stats.

http://articles.mql4.com/340

Those are all very valid concerns and ones which I don't have ready answers for.

A bar is simply a range of time. It has an open, a high a low, and a close. The major drawback with this EA is a 50pip long bar is the same as a 1 pip long bar. I know that drawback exists. I don't know how to compensate for it. Options include:

 

1. Using renko bars
2. Factoring things such as:
   1. Wick length
   2. Pip count
   3. Tick count
   4. Volume
   5. High - Open
   6. High - Low
   7. High - Close
   8. etc
   9. etc
   10. etc

3. Writing a self-organizing map

This EA was originally written to look for marching candles (3 large long/short candles). I was curious as to what other patterns might exist, and I modified the original EA with the very (wrong) assumption that every bar is the same size.

I have looked into using arrays of ticks to find patterns, but realtime usage is almost impossible. Every week, my tick database receives roughly 4 million ticks. That boils down to roughly:

 

1. 727,000 Ticks / day
2. 30,000 Ticks / hour
3. 500 Ticks / minute
4. 8 Ticks / second

That's far too much data for all of my servers combined to process in real time.

hi Ron,
finally you catch it : we all want to trade eachtick move and that's where we fail. We can't trade "looking at our feet" because we can't see the landscape , small time frames just disturb us of main trend.

So, to comment your above post, i think it will probably be a good idea to follow W1 or MN1, D1 is to small but H1 will be able to give something good if we exclude "dead hours".

2 trades 2 won...(+60 and +2 )

for old dogs = 1 trade 1 won (+100)

im sorry to disturb, but why 'incorrect table name' error?

That's the EA trying to reach the tick database.

I'm guessing you attached the EA to a pair which the database doesn't currently support.


[Edit:]
I checked the error logs, apparently this EA wasn't sending the symbol it was on. This has been fixed below.

ok understood thnks

Hmmm... some ideas to improve your pattern-machine.... I wonder that a 100pip Candle with DIFF(Open, Close) = 1 is as same bullish as a 100pip Candle which is completely bullish...

Therefore I'm trying to classify candles as follow:

- if OPEN is above the first third of the candle we assign value '0'
- if OPEN is below the first third and above the second third --> value '1'
- if OPEN is below second third --> value '2'

Now you have to add information about the close-value.... all in all we then have 9 different candles (value 0 - 8).... which seems to be more precise, isn't it?

That's closer to what I'm trying to accomplish.

Do you know what an MD5 checksum is? It's basically a way to uniquely identify a file regardless of size, content, etc.

I'm trying to create something similar for a series of bars. There are currently 127 unique keys in the ASCII character set. That's 127 different potential classifications for bars. Each character (A, a, #, $, %, @, ~, etc) would represent a specific kind of bar. That representation can then form a string and that string can represent a particular pattern, wherein multiple instances of that string can be found in the bar history, and with those multiple instances, EAs with near-realtime pattern finding studies can be generated.

But before any of that can happen, I need a way to classify every single bar in the market into 127 specific groups.

Any suggestions?

Well 127 (or to say it better 128) means we have 7 different boolean values to use....

Imagine we'll take the pattern I described above as a base, then we have 4 Bits out of 7 used... so one could use further 3 bits to classify e.g. the length of the candle depending of ATR or something like that...