Log in
  • Files
  • jaguar1637
  • My Idea regarding optimization taken from Combo Right from http://codebase.mql4.com/2792

My Idea regarding optimization taken from Combo Right from http://codebase.mql4.com/2792

My Idea regarding optimization taken from Combo Right from http://codebase.mql4.com/2792
  • Public
By jaguar1637 2917 days ago Comments (13)

For the Combo Rights, the situation is :

a two-layer neural network consisting of two perceptrons in the lower layer and one perceptron in the upper layer.  The output of the neural network can be in one of these three states:

  • Entering the market with a long position
  • Entering the market with a short position
  • Indeterminate state

there are 4 stages

  • Stage 1 of optimization. Optimization of the BTS
  • Stage 2. Teaching the perceptron responsible for short positions
  • Stage 3. Teaching the perceptron responsible for long positions
  • Stage 4 (final). Teaching the first layer, i.e., teaching the perceptron that is in the upper layer

x14, x24, x34, x44 - weight numbers of the perceptron of the first layer. It is optimized with the values within the range of 0 to 200, step 1.
p4 - the period of the values of price difference to be analyzed by the perceptron. It is optimized with the values within the range of 3 to 100, step 1.

____________________________________________________________________________

Now, my idea =>

  • calculating the correlation to the price

instead of stage 2 & 3, Teaching the perceptron for short/long positions, I preconize Teaching the perceptron for correlation to the price

 

 

Comments

  • vasi 2916 days ago

    I wish to see the indicator, if exist.

  • jaguar1637 2916 days ago

    Hi Vasi

    Some people here (john and me) wrote perceptrons. You can find them in section Files

    Here, the aim is to tune this kind of indicators. A mix of those tuned perceptrons will provide a very optimized entry/exit signals.

    Simple EA will be able to work easily from those signals

     

  • vasi 2916 days ago

    Hi Jag !

    Nice to see you ! I've asked about indicator only to see a backtest of this.

    Ok. I hop e to see your work soon.

    I've found something that could help you, to create more fast your EA's:

    http://www.stevehopwoodforex.com/phpBB3/viewtopic.php?p=803#p803 ( the best. shell EA code)

    http://www.stevehopwoodforex.com/phpBB3/viewtopic.php?p=5733#p5733

    http://www.stevehopwoodforex.com/phpBB3/viewtopic.php?p=18306#p18306

  • jaguar1637 2916 days ago

    OK VAsi

    Steve is very good and is a better MQ4 programmer than me, for havind built a very good and reliable squeletton.

    Except some points that are complex to study in deep  :

    • I am not sure partial closing is a good idea. I read the code and found very complex calculations regarding such things in the code. Still wandering if there are necessary or not
    • I did not see a best calculation for taking profit. For example, if there is a high trend, and if the price is going to reach the initial tp, it could be nice to redeclare the tp (may be w/ the help of the FIBO) to another level.
    • there is no jumping stop loss, nor instant trailing stop (may be they are useless)

     

  • jaguar1637 2916 days ago

    john,

    As I read on the web page, it seemed, the past periods should be reduced to one or two pasted weeks.

    Do you have any idea to fetch a long cycle period in this threshold ?

     

  • JohnLast 2916 days ago

    No I do not know, every strategy has its own requirements, basically it is trial and error for me.

  • JohnLast 2916 days ago

    All that is very interesting you can easily change the linear activation function with something else: many examples here: radial, sigmoid, rational etc.

     

  • JohnLast 2916 days ago

    All that is very interesting you can easily change the linear activation function with something else: many examples here: radial, sigmoid, rational etc.

    It is easy to have two perceptrons for example

    if perceyton 1>0 buy

    if perceptron 2<0 sell

    if perceptrone 1<0 and percetron 2 >0 no trade that is a possible logic

  • jaguar1637 2916 days ago

    John,

    You remember what I wrote before

    the purpose of this optimization is to get the perceptron lines as closed from the correlation to the price.

    now, there is also 2 levels provided by the perceptrons

    if perceptron signals is below (or crossing up) oversold  (20% or 25% of the perceptron) implies BUY

    if perceptron signals is above (or crossing down) overbought (75% to 80% of the perceptron) implies SELL

    Note  the kind of signals are breakout likes.

     

    I hope every body understand this concept

  • JohnLast 2916 days ago

    Yes I do, it is great.

     

  • jaguar1637 2915 days ago

    So, the main idea is

    first : - to get the best weights to fetch the best correlation to the price (I do not know yet which algo to apply, I thought first about cascade correlation ... but ... not sure )

    secondly : - to run the NN to find the break out signals (below, above). (the last tuning regards the oversold/overbought levels )

    notice that this stuff is dedicated for short and medium timeframes.

    I think the best for me is to write directly to Reshetov. OK, I will do it this morning

     

  • jaguar1637 2915 days ago

    If a clear correlation to the price is not realized, there will be a loss instead of won.

    as detail ed in  :

    http://www.forex-tsd.com/expert-advisors-metatrader-4/9034-artificial-intelligence-11.html

    many guys lost money

  • JohnLast 2915 days ago

    The Artificial Intelligence is a conceptual EA. The core is the idea. 

    Making money out of an idea is a completely different thing, a lot of real intelligence lol is needed.

    So that does not surprise me that many loose money. 

     

    If I understand correctly what needed to be optimized are mainly the inputs (not only the weights).

    And as you can see in the native mt4 genetic algorythm you can optimize only (Balance minimum, profit maximum, consecutive loss etc.) but not towards a particular function.

    Maybe that is possible with mt5 (there is custom max).

     

    Or you need to export the data and use a neural net simulator.