forked from mnbvc188199/Warrior_EA
355 lines
30 KiB
MQL5
355 lines
30 KiB
MQL5
//+------------------------------------------------------------------+
| |
//| Study.mq5 |
| |
//| Copyright DNG® |
| |
//| https://www.mql5.com/en/users/dng |
| |
//+------------------------------------------------------------------+
| |
#property copyright "Copyright DNG®"
| |
#property link "https://www.mql5.com/en/users/dng"
| |
#property version "1.00"
| |
//+------------------------------------------------------------------+
| |
//| |
| |
//+------------------------------------------------------------------+
| |
#define Study
| |
#include "Trajectory.mqh"
| |
//+------------------------------------------------------------------+
| |
//| Input parameters |
| |
//+------------------------------------------------------------------+
| |
input int Iterations = 100000;
| |
input float Tau = 0.01f;
| |
//+------------------------------------------------------------------+
| |
//| |
| |
//+------------------------------------------------------------------+
| |
STrajectory Buffer[];
| |
CNet Actor;
| |
CNet Critic1;
| |
CNet Critic2;
| |
CNet TargetCritic1;
| |
CNet TargetCritic2;
| |
CNet Autoencoder;
| |
//---
| |
float dError;
| |
datetime dtStudied;
| |
//---
| |
CBufferFloat State;
| |
CBufferFloat Account;
| |
CBufferFloat Actions;
| |
CBufferFloat Gradient;
| |
CBufferFloat *Result;
| |
vector<float> AutoencoderResult;
| |
vector<float> check;
| |
vector<float> CriticResult;
| |
float MinCriticError;
| |
float MaxCriticError;
| |
float AvgCriticError;
| |
//---
| |
COpenCLMy *OpenCL;
| |
//+------------------------------------------------------------------+
| |
//| Expert initialization function |
| |
//+------------------------------------------------------------------+
| |
int OnInit()
| |
{
| |
//---
| |
ResetLastError();
| |
if(!LoadTotalBase())
| |
{
| |
PrintFormat("Error of load study data: %d", GetLastError());
| |
return INIT_FAILED;
| |
}
| |
//--- load models
| |
float temp;
| |
if(!Actor.Load(FileName + "Act.nnw", temp, temp, temp, dtStudied, true) ||
| |
!Critic1.Load(FileName + "Crt1.nnw", temp, temp, temp, dtStudied, true) ||
| |
!Critic2.Load(FileName + "Crt2.nnw", temp, temp, temp, dtStudied, true) ||
| |
!Autoencoder.Load(FileName + "AEnc.nnw", temp, temp, temp, dtStudied, true) ||
| |
!TargetCritic1.Load(FileName + "Crt1.nnw", temp, temp, temp, dtStudied, true) ||
| |
!TargetCritic2.Load(FileName + "Crt2.nnw", temp, temp, temp, dtStudied, true))
| |
{
| |
CArrayObj *actor = new CArrayObj();
| |
CArrayObj *critic = new CArrayObj();
| |
CArrayObj *autoencoder = new CArrayObj();
| |
if(!CreateDescriptions(actor, critic, autoencoder))
| |
{
| |
delete actor;
| |
delete critic;
| |
delete autoencoder;
| |
return INIT_FAILED;
| |
}
| |
if(!Actor.Create(actor) || !Critic1.Create(critic) || !Critic2.Create(critic) ||
| |
!Autoencoder.Create(autoencoder))
| |
{
| |
delete actor;
| |
delete critic;
| |
delete autoencoder;
| |
return INIT_FAILED;
| |
}
| |
if(!TargetCritic1.Create(critic) || !TargetCritic2.Create(critic))
| |
{
| |
delete actor;
| |
delete critic;
| |
delete autoencoder;
| |
return INIT_FAILED;
| |
}
| |
delete actor;
| |
delete critic;
| |
delete autoencoder;
| |
//---
| |
TargetCritic1.WeightsUpdate(GetPointer(Critic1), 1.0f);
| |
TargetCritic2.WeightsUpdate(GetPointer(Critic2), 1.0f);
| |
}
| |
//---
| |
OpenCL = Actor.GetOpenCL();
| |
Critic1.SetOpenCL(OpenCL);
| |
Critic2.SetOpenCL(OpenCL);
| |
TargetCritic1.SetOpenCL(OpenCL);
| |
TargetCritic2.SetOpenCL(OpenCL);
| |
Autoencoder.SetOpenCL(OpenCL);
| |
//---
| |
Actor.getResults(Result);
| |
if(Result.Total() != NActions)
| |
{
| |
PrintFormat("The scope of the actor does not match the actions count (%d <> %d)", NActions, Result.Total());
| |
return INIT_FAILED;
| |
}
| |
//---
| |
Actor.GetLayerOutput(0, Result);
| |
if(Result.Total() != (HistoryBars * BarDescr))
| |
{
| |
PrintFormat("Input size of Actor doesn't match state description (%d <> %d)", Result.Total(), (HistoryBars * BarDescr));
| |
return INIT_FAILED;
| |
}
| |
//---
| |
Actor.GetLayerOutput(LatentLayer, Result);
| |
int latent_state = Result.Total();
| |
Critic1.GetLayerOutput(0, Result);
| |
if(Result.Total() != latent_state)
| |
{
| |
PrintFormat("Input size of Critic doesn't match latent state Actor (%d <> %d)", Result.Total(), latent_state);
| |
return INIT_FAILED;
| |
}
| |
//---
| |
Critic1.GetLayerOutput(1, Result);
| |
latent_state = Result.Total();
| |
Autoencoder.GetLayerOutput(0, Result);
| |
if(Result.Total() != latent_state)
| |
{
| |
PrintFormat("Input size of Autoencoder doesn't match latent state Critic (%d <> %d)", Result.Total(), latent_state);
| |
return INIT_FAILED;
| |
}
| |
//---
| |
Gradient.BufferInit(AccountDescr, 0);
| |
//---
| |
if(!EventChartCustom(ChartID(), 1, 0, 0, "Init"))
| |
{
| |
PrintFormat("Error of create study event: %d", GetLastError());
| |
return INIT_FAILED;
| |
}
| |
//---
| |
return(INIT_SUCCEEDED);
| |
}
| |
//+------------------------------------------------------------------+
| |
//| Expert deinitialization function |
| |
//+------------------------------------------------------------------+
| |
void OnDeinit(const int reason)
| |
{
| |
//---
| |
TargetCritic1.WeightsUpdate(GetPointer(Critic1), Tau);
| |
TargetCritic2.WeightsUpdate(GetPointer(Critic2), Tau);
| |
Actor.Save(FileName + "Act.nnw", 0, 0, 0, TimeCurrent(), true);
| |
TargetCritic1.Save(FileName + "Crt1.nnw", Critic1.getRecentAverageError(), 0, 0, TimeCurrent(), true);
| |
TargetCritic1.Save(FileName + "Crt2.nnw", Critic2.getRecentAverageError(), 0, 0, TimeCurrent(), true);
| |
Autoencoder.Save(FileName + "AEnc.nnw", Autoencoder.getRecentAverageError(), 0, 0, TimeCurrent(), true);
| |
delete Result;
| |
}
| |
//+------------------------------------------------------------------+
| |
//| ChartEvent function |
| |
//+------------------------------------------------------------------+
| |
void OnChartEvent(const int id,
| |
const long &lparam,
| |
const double &dparam,
| |
const string &sparam)
| |
{
| |
//---
| |
if(id == 1001)
| |
Train();
| |
}
| |
//+------------------------------------------------------------------+
| |
//| Train function |
| |
//+------------------------------------------------------------------+
| |
void Train(void)
| |
{
| |
int total_tr = ArraySize(Buffer);
| |
uint ticks = GetTickCount();
| |
//---
| |
for(int iter = 0; (iter < Iterations && !IsStopped()); iter ++)
| |
{
| |
int tr = (int)((MathRand() / 32767.0) * (total_tr - 1));
| |
int i = (int)((MathRand() * MathRand() / MathPow(32767, 2)) * (Buffer[tr].Total - 2));
| |
//--- Target
| |
State.AssignArray(Buffer[tr].States[i + 1].state);
| |
float PrevBalance = Buffer[tr].States[i].account[0];
| |
float PrevEquity = Buffer[tr].States[i].account[1];
| |
Account.Clear();
| |
Account.Add((Buffer[tr].States[i + 1].account[0] - PrevBalance) / PrevBalance);
| |
Account.Add(Buffer[tr].States[i + 1].account[1] / PrevBalance);
| |
Account.Add((Buffer[tr].States[i + 1].account[1] - PrevEquity) / PrevEquity);
| |
Account.Add(Buffer[tr].States[i + 1].account[2]);
| |
Account.Add(Buffer[tr].States[i + 1].account[3]);
| |
Account.Add(Buffer[tr].States[i + 1].account[4] / PrevBalance);
| |
Account.Add(Buffer[tr].States[i + 1].account[5] / PrevBalance);
| |
Account.Add(Buffer[tr].States[i + 1].account[6] / PrevBalance);
| |
double x = (double)Buffer[tr].States[i + 1].account[7] / (double)(D'2024.01.01' - D'2023.01.01');
| |
Account.Add((float)MathSin(x != 0 ? 2.0 * M_PI * x : 0));
| |
x = (double)Buffer[tr].States[i + 1].account[7] / (double)PeriodSeconds(PERIOD_MN1);
| |
Account.Add((float)MathCos(x != 0 ? 2.0 * M_PI * x : 0));
| |
x = (double)Buffer[tr].States[i + 1].account[7] / (double)PeriodSeconds(PERIOD_W1);
| |
Account.Add((float)MathSin(x != 0 ? 2.0 * M_PI * x : 0));
| |
x = (double)Buffer[tr].States[i + 1].account[7] / (double)PeriodSeconds(PERIOD_D1);
| |
Account.Add((float)MathSin(x != 0 ? 2.0 * M_PI * x : 0));
| |
//---
| |
if(Account.GetIndex() >= 0)
| |
Account.BufferWrite();
| |
if(!Actor.feedForward(GetPointer(State), 1, false, GetPointer(Account)))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
ExpertRemove();
| |
break;
| |
}
| |
//---
| |
if(!TargetCritic1.feedForward(GetPointer(Actor), LatentLayer, GetPointer(Actor)) ||
| |
!TargetCritic2.feedForward(GetPointer(Actor), LatentLayer, GetPointer(Actor)))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
break;
| |
}
| |
TargetCritic1.getResults(Result);
| |
float reward = Result[0];
| |
TargetCritic2.getResults(Result);
| |
reward = Buffer[tr].Revards[i] + DiscFactor * (MathMin(reward, Result[0]) - Buffer[tr].Revards[i + 1]);
| |
//--- Q-function study
| |
State.AssignArray(Buffer[tr].States[i].state);
| |
PrevBalance = Buffer[tr].States[MathMax(i - 1, 0)].account[0];
| |
PrevEquity = Buffer[tr].States[MathMax(i - 1, 0)].account[1];
| |
Account.Update(0, (Buffer[tr].States[i].account[0] - PrevBalance) / PrevBalance);
| |
Account.Update(1, Buffer[tr].States[i].account[1] / PrevBalance);
| |
Account.Update(2, (Buffer[tr].States[i].account[1] - PrevEquity) / PrevEquity);
| |
Account.Update(3, Buffer[tr].States[i].account[2]);
| |
Account.Update(4, Buffer[tr].States[i].account[3]);
| |
Account.Update(5, Buffer[tr].States[i].account[4] / PrevBalance);
| |
Account.Update(6, Buffer[tr].States[i].account[5] / PrevBalance);
| |
Account.Update(7, Buffer[tr].States[i].account[6] / PrevBalance);
| |
x = (double)Buffer[tr].States[i].account[7] / (double)(D'2024.01.01' - D'2023.01.01');
| |
Account.Add((float)MathSin(x != 0 ? 2.0 * M_PI * x : 0));
| |
x = (double)Buffer[tr].States[i].account[7] / (double)PeriodSeconds(PERIOD_MN1);
| |
Account.Add((float)MathCos(x != 0 ? 2.0 * M_PI * x : 0));
| |
x = (double)Buffer[tr].States[i].account[7] / (double)PeriodSeconds(PERIOD_W1);
| |
Account.Add((float)MathSin(x != 0 ? 2.0 * M_PI * x : 0));
| |
x = (double)Buffer[tr].States[i].account[7] / (double)PeriodSeconds(PERIOD_D1);
| |
Account.Add((float)MathSin(x != 0 ? 2.0 * M_PI * x : 0));
| |
Account.BufferWrite();
| |
//---
| |
if(!Actor.feedForward(GetPointer(State), 1, false, GetPointer(Account)))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
break;
| |
}
| |
//---
| |
Actions.AssignArray(Buffer[tr].States[i].action);
| |
if(Actions.GetIndex() >= 0)
| |
Actions.BufferWrite();
| |
//---
| |
if(!Critic1.feedForward(GetPointer(Actor), LatentLayer, GetPointer(Actions)) ||
| |
!Critic2.feedForward(GetPointer(Actor), LatentLayer, GetPointer(Actions)))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
break;
| |
}
| |
Critic1.getResults(Result);
| |
float error = reward - Result[0];
| |
if(iter == 0)
| |
{
| |
MaxCriticError = error;
| |
MinCriticError = error;
| |
AvgCriticError = error;
| |
}
| |
else
| |
{
| |
MaxCriticError = MathMax(error, MaxCriticError);
| |
MinCriticError = MathMin(error, MinCriticError);
| |
AvgCriticError = 0.99f * AvgCriticError + 0.01f * error;
| |
}
| |
Critic2.getResults(Result);
| |
error = reward - Result[0];
| |
MaxCriticError = MathMax(error, MaxCriticError);
| |
MinCriticError = MathMin(error, MinCriticError);
| |
AvgCriticError = 0.99f * AvgCriticError + 0.01f * error;
| |
//---
| |
Result.Update(0, reward);
| |
if(!Critic1.backProp(Result, GetPointer(Actions), GetPointer(Gradient)) ||
| |
!Critic2.backProp(Result, GetPointer(Actions), GetPointer(Gradient)))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
break;
| |
}
| |
//--- Policy study
| |
CNet *critic = NULL;
| |
if(Critic1.getRecentAverageError() <= Critic2.getRecentAverageError())
| |
critic = GetPointer(Critic1);
| |
else
| |
critic = GetPointer(Critic2);
| |
//---
| |
if(!critic.feedForward(GetPointer(Actor), LatentLayer, GetPointer(Actor)) ||
| |
!Autoencoder.feedForward(critic, 1, NULL, -1))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
break;
| |
}
| |
Autoencoder.getResults(AutoencoderResult);
| |
critic.GetLayerOutput(1, Result);
| |
Result.GetData(CriticResult);
| |
critic.getResults(Result);
| |
//---
| |
float alpha = (MaxCriticError == MinCriticError ? 0 : 10.0f * (AvgCriticError - MinCriticError) / (MaxCriticError - MinCriticError));
| |
alpha = 1.0f / (1.0f + MathExp(-alpha));
| |
alpha = 1 - alpha;
| |
reward = Result[0];
| |
reward = (reward > 0 ? reward + PoliticAdjust : PoliticAdjust);
| |
reward += AutoencoderResult.Loss(CriticResult, LOSS_MSE) * alpha;
| |
//---
| |
Result.Update(0, reward);
| |
critic.TrainMode(false);
| |
if(!critic.backProp(Result, GetPointer(Actor)) ||
| |
!Actor.backPropGradient(GetPointer(Account), GetPointer(Gradient), LatentLayer) ||
| |
!Actor.backPropGradient(GetPointer(Account), GetPointer(Gradient)))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
critic.TrainMode(true);
| |
break;
| |
}
| |
critic.TrainMode(true);
| |
//--- Autoencoder study
| |
Result.AssignArray(CriticResult);
| |
if(!Autoencoder.backProp(Result, critic, 1))
| |
{
| |
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
| |
break;
| |
}
| |
//--- Update Target Nets
| |
TargetCritic1.WeightsUpdate(GetPointer(Critic1), Tau);
| |
TargetCritic2.WeightsUpdate(GetPointer(Critic2), Tau);
| |
//---
| |
if(GetTickCount() - ticks > 500)
| |
{
| |
string str = StringFormat("%-15s %5.2f%% -> Error %15.8f\n", "Critic1", iter * 100.0 / (double)(Iterations), Critic1.getRecentAverageError());
| |
str += StringFormat("%-15s %5.2f%% -> Error %15.8f\n", "Critic2", iter * 100.0 / (double)(Iterations), Critic2.getRecentAverageError());
| |
Comment(str);
| |
ticks = GetTickCount();
| |
}
| |
}
| |
Comment("");
| |
//---
| |
PrintFormat("%s -> %d -> %-15s %10.7f", __FUNCTION__, __LINE__, "Critic1", Critic1.getRecentAverageError());
| |
PrintFormat("%s -> %d -> %-15s %10.7f", __FUNCTION__, __LINE__, "Critic2", Critic2.getRecentAverageError());
| |
ExpertRemove();
| |
//---
| |
}
| |
//+------------------------------------------------------------------+
|