Quote:
Originally Posted by doshur
Some crazy idea just pop up in my head creating a Probabilistic Perceptron Neural Network
this is how it works.
every new bar, it will open a buy or a sell
the perceptron will adjust the weights and learn
give me your thoughts
How to use
Use the trainer to train,
inputs 0 - 200, step 1 for all
then sort by profit factor
|
Doshur,
Thanks for the perceptron,I have just downloaded it,so my comments are preliminary.
By looking at this code:
double ATR = iATR(NULL, 0,
13, 0);
double StopLoss = ATR *
3;
double TakeProfit = ATR;
I suggest that you modify it to
double ATR = iATR(NULL, 0,
M,0);
double StopLoss = ATR *
N;
double TakeProfit = ATR*
P;
This way the perceptron will be able to find the most suitable ATR period and ATR multipliers for both stoploss and Take Profit.
I would use a range for M,N,P from 0.5 to 10 in steps of 0.5...as an alternative, you could make them external doubles and then optimize,but I believe that the fun of using the perceptron is for it to do the optimization on a continuous basis.If this is too wide a range and it slows the process ,you cold then make them external doubles,optimize to find where the best settings range for several different pairs and then reduce the "range of search" of the perceptron accordingly(if,for example,trough optimization of several pairs,you find optimal M(ATR period) ranges from 5 to 8..optimal N(Stoploss ATR multiplier) between 0.5 and 2.5..and optimalP(Atr multiplier for take profit) from 1.5 to 5.0..you can use those limits and program the perceptron accordingly).
Simba