• 概述
  • 评论
  • 评论 (1)

Flash Neural Network EA

Flash Neural Network EA is Expert Advisor based on Neural Network with back propagation learning algorithm.

This software can be used on ANY market and on ANY time frame. Every parameter of network is fully configurable.
Most complicated and time-consuming part of building successful neural decision model is proper input data preparation.

Now everything is fully automated. You need to only pick input symbols, set neural network parameters as number of hidden layer neurons, maximal training epochs or target MSE and start testing this EA. All data will be formatted for Neural Network input correctly.
  • Input Instruments - Input instruments to use as input of Neural Network (detailed explanation below);
  • Activation Function - Activation Function for neurons (0 - sigmoid, 1 - tangent, 2 - hyperbolic tangent);
  • Learning Rate - Controls the step size when weights are iteratively adjusted (0-1, higher = faster learning but less accurate);
  • Max Epochs - Maximum number of iterations to stop learning;
  • Max MSE - Target Mean Square Error, if training MSE reach this limit, learning process stops;
  • Training Data Samples - Number of training data samples used in learning; (limited to 100 samples);
  • Hidden Layer Neurons - number of neurons in hidden layer;
  • Trigger - float value from 0 to 0.5, determine how strong signal will allow to open position;
  • Risk - Risk (in percents);
  • Take Profit - Take Profit (in pips);
  • Stop Loss - Stop Loss (in pips);
  • Maximum Lots - Maximum Lots allowed;
  • Control Balance Multiplier - Determine how aggressive will grow lot after loss;
  • Always Close on New Bar - EA will close opened position when new bar starts drawing.
Input Instruments:

Input instruments are strings separated by spaces in format:
  • "[INSTRUMENT_1_NAME][NUMBER(0-9)][SPACE][INSTRUMENT_2_NAME][NUMBER(0-9)][SPACE]...[INSTRUMENT_N_NAME][NUMBER(0-9)]"
for example "EURUSD1 EURUSD4 USDJPY1 SILVER1". Neural Network is limited to 100 inputs.

Number at the end of instrument name means price difference from current bar to number.
For example entry "EURUSD1" gives to Neural Network difference in price of EURUSD from last bar(t-1) to current price (t0).

Default parameters are suitable for EURUSD 1D interval.

Presented history results are in two options: without MM (Fixed Max Lot = 1.00) and with growing lot (Lot is calculated from free margin).
Initial deposit was in both cases $10 000. Input parameters set to default. Testing period from 2012-01-01 to 2012-04-01.

Results:
  • Option (Max Lot = 1) -> $21 516 net profit !
  • Option (Max Lot = 0) -> $53 559 net profit !! (above 500%)
无评论