beathespread

Here I would like another shot of out of sample results of the Spinal Implant EA. However in this EA, I am using experimental activation function for the percetrone. And the whole project is not ready for release yet, as it is in the experimental stage.

The system is looking to enter into a trade only if the market conditions are sufficiently predictable. That is achieved thanks to the fractal geometry filtering. Basically we are looking to implement several technologies.

-Fractal dimension filtering. We are looking for a chaos treshhold and only to enter into a trade if the fractal dimesnions is low enogh (or Hurst exponent large enough). 

- Evolutionary and "unsupervised" neural network that is finding the general direction when the market conditions are favorable for directionnal prediction. As the chaos filter is applied we train the percetrone only with the predictable market conditions. The comparative tests show that the percetrone mod of the fractal scalper outperforms its more traditional counter part based on rough indicator readings.

-Percetron function with special activation gauss function optimized by genetic optmimizer. This is experimental architecture with non conclusive results. The idea is to map the gauss function into the 3d space using two optimized parameters per neurone.