The idea and the simplest algorithm are provided in the article "Random decision forest in reinforcement learning"
The library has advanced functionality allowing you to create an unlimited number of "Agents".
In addition, variations of the "Arguments group accounting method" are used
Using the library:
#include <RL gmdh.mqh> CRLAgents *ag1=new CRLAgents("RlExp1iter",1,100,50,regularize,learn); //created 1 RL agent accepting 100 entries (predictor values) and containing 50 trees
An example of filling input values with normalized close prices:
void calcSignal() { sig1=0; double arr[]; CopyClose(NULL,0,1,10000,arr); ArraySetAsSeries(arr,true); normalizeArrays(arr); for(int i=0;i<ArraySize(ag1.agent);i++) { ArrayCopy(ag1.agent[i].inpVector,arr,0,0,ArraySize(ag1.agent[i].inpVector)); } sig1=ag1.getTradeSignal(); }
Training takes place in the tester in one pass with the parameter learn=true. After training, we need to change it to false.
Demonstrating the trained "RL gmdh trader" EA operation on training and test samples.
