forked from dng/NN_in_Trading
672 lines
28 KiB
MQL5
672 lines
28 KiB
MQL5
//+------------------------------------------------------------------+
|
|
//| Study.mq5 |
|
|
//+------------------------------------------------------------------+
|
|
#property copyright "Copyright DNG®"
|
|
#property link "https://www.mql5.com/ru/users/dng"
|
|
#property version "1.00"
|
|
#define Study
|
|
#define StudyOnline
|
|
//+------------------------------------------------------------------+
|
|
//| Input parameters |
|
|
//+------------------------------------------------------------------+
|
|
input datetime Start = D'2024.01.01';
|
|
input datetime End = D'2026.01.01';
|
|
input int Epochs = 100;
|
|
input double MinBalance = 50;
|
|
input double MaxBalance = 150;
|
|
input bool LoadBaseMemorySnapshot = true;
|
|
input string BaseMemorySnapshotPath = "VLADriverBaseMemory.snapshot";
|
|
#include "Trajectory.mqh"
|
|
CNet cActor;
|
|
CNet cCritic;
|
|
CNet cStateEncoder;
|
|
|
|
datetime dtStudied;
|
|
CBufferFloat bState;
|
|
CBufferFloat bContext;
|
|
CBufferFloat bTime;
|
|
CBufferFloat bGradient;
|
|
CBufferFloat *Result;
|
|
CBufferFloat *Action;
|
|
|
|
// The library owns snapshot, records, consolidation and GPU retrieval.
|
|
CNeuronRAGMemory RAGMemory;
|
|
bool OfflineHasLastStoredAction = false;
|
|
float OfflineLastStoredAction[NActions];
|
|
const float OfflineActionChangeThreshold = 0.01f;
|
|
const float OfflineActionNonZeroThreshold = 0.001f;
|
|
ulong OfflineCollectionChecks = 0;
|
|
ulong OfflineTerminalFailures = 0;
|
|
ulong OfflineRawLotRejections = 0;
|
|
ulong OfflineScenarioAcquireFailures = 0;
|
|
ulong OfflineScenarioShapeFailures = 0;
|
|
ulong OfflineScenarioReadFailures = 0;
|
|
ulong OfflineScenarioReplacedValues = 0;
|
|
ulong OfflineNormalizationFailures = 0;
|
|
ulong OfflineNormalizationInputFailures = 0;
|
|
ulong OfflineNormalizationMarketFailures = 0;
|
|
ulong OfflineNormalizationVolatilityFailures = 0;
|
|
ulong OfflineNormalizationDescriptorFailures = 0;
|
|
ulong OfflineNormalizationZeroFailures = 0;
|
|
ulong OfflineDedupeRejections = 0;
|
|
ulong OfflineRecordFailures = 0;
|
|
ulong OfflineAcceptedRecords = 0;
|
|
ulong OfflinePublishAttempts = 0;
|
|
ulong OfflinePublishFailures = 0;
|
|
ulong OfflinePublishSuccesses = 0;
|
|
bool OfflineFirstRejectionCaptured = false;
|
|
bool OfflineDiagnosticsPrinted = false;
|
|
string OfflineFirstRejectionReason = "";
|
|
double OfflineFirstRejectedAction0 = 0;
|
|
double OfflineFirstRejectedAction3 = 0;
|
|
double OfflineFirstRejectedATR = 0;
|
|
//+------------------------------------------------------------------+
|
|
//| Reset per-run offline RAG collection diagnostics. |
|
|
//+------------------------------------------------------------------+
|
|
void ResetOfflineRAGDiagnostics(void)
|
|
{
|
|
OfflineCollectionChecks = 0;
|
|
OfflineTerminalFailures = 0;
|
|
OfflineRawLotRejections = 0;
|
|
OfflineScenarioAcquireFailures = 0;
|
|
OfflineScenarioShapeFailures = 0;
|
|
OfflineScenarioReadFailures = 0;
|
|
OfflineScenarioReplacedValues = 0;
|
|
OfflineNormalizationFailures = 0;
|
|
OfflineNormalizationInputFailures = 0;
|
|
OfflineNormalizationMarketFailures = 0;
|
|
OfflineNormalizationVolatilityFailures = 0;
|
|
OfflineNormalizationDescriptorFailures = 0;
|
|
OfflineNormalizationZeroFailures = 0;
|
|
OfflineDedupeRejections = 0;
|
|
OfflineRecordFailures = 0;
|
|
OfflineAcceptedRecords = 0;
|
|
OfflinePublishAttempts = 0;
|
|
OfflinePublishFailures = 0;
|
|
OfflinePublishSuccesses = 0;
|
|
OfflineFirstRejectionCaptured = false;
|
|
OfflineDiagnosticsPrinted = false;
|
|
OfflineFirstRejectionReason = "";
|
|
OfflineFirstRejectedAction0 = 0;
|
|
OfflineFirstRejectedAction3 = 0;
|
|
OfflineFirstRejectedATR = 0;
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Retain one compact sample without changing collection decisions. |
|
|
//+------------------------------------------------------------------+
|
|
void CaptureOfflineRAGRejection(const string reason, const uint position)
|
|
{
|
|
if(OfflineFirstRejectionCaptured)
|
|
return;
|
|
OfflineFirstRejectionCaptured = true;
|
|
OfflineFirstRejectionReason = reason;
|
|
if(Action != NULL && Action.Total() > 3)
|
|
{
|
|
OfflineFirstRejectedAction0 = Action[0];
|
|
OfflineFirstRejectedAction3 = Action[3];
|
|
}
|
|
if(position < Rates.Size())
|
|
OfflineFirstRejectedATR = ATR.Main(position);
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Print one final concise collection summary. |
|
|
//+------------------------------------------------------------------+
|
|
void PrintOfflineRAGDiagnostics(void)
|
|
{
|
|
if(OfflineDiagnosticsPrinted || (OfflineCollectionChecks == 0 && OfflinePublishAttempts == 0))
|
|
return;
|
|
OfflineDiagnosticsPrinted = true;
|
|
PrintFormat("Offline RAG collection: checked=%I64u accepted=%I64u terminal=%I64u lot=%I64u acquire=%I64u shape=%I64u read=%I64u replaced=%I64u normalize=%I64u dedupe=%I64u add=%I64u",
|
|
OfflineCollectionChecks, OfflineAcceptedRecords, OfflineTerminalFailures, OfflineRawLotRejections,
|
|
OfflineScenarioAcquireFailures, OfflineScenarioShapeFailures, OfflineScenarioReadFailures,
|
|
OfflineScenarioReplacedValues, OfflineNormalizationFailures, OfflineDedupeRejections,
|
|
OfflineRecordFailures);
|
|
PrintFormat("Offline RAG normalization: input=%I64u market=%I64u volatility=%I64u descriptor=%I64u zero=%I64u",
|
|
OfflineNormalizationInputFailures, OfflineNormalizationMarketFailures,
|
|
OfflineNormalizationVolatilityFailures, OfflineNormalizationDescriptorFailures,
|
|
OfflineNormalizationZeroFailures);
|
|
PrintFormat("Offline RAG publication: attempts=%I64u success=%I64u failures=%I64u scenarios=%u pending=%u",
|
|
OfflinePublishAttempts, OfflinePublishSuccesses, OfflinePublishFailures,
|
|
RAGMemory.ScenarioCount(), RAGMemory.PendingRecordCount());
|
|
if(OfflineFirstRejectionCaptured)
|
|
PrintFormat("Offline RAG first rejection: reason=%s action0=%.8f action3=%.8f atr=%.8f",
|
|
OfflineFirstRejectionReason, OfflineFirstRejectedAction0,
|
|
OfflineFirstRejectedAction3, OfflineFirstRejectedATR);
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Persist the immutable library-owned Base Memory snapshot. |
|
|
//+------------------------------------------------------------------+
|
|
bool SaveOfflineBaseMemory(void)
|
|
{
|
|
// Do not replace a previously published image with an empty run.
|
|
if(RAGMemory.ScenarioCount() == 0)
|
|
return true;
|
|
return RAGMemory.SavePublishedSnapshot(BaseMemorySnapshotPath);
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Rebind the Actor immediately after COW snapshot publication. |
|
|
//+------------------------------------------------------------------+
|
|
bool RebindOfflineRAGMemory(void)
|
|
{
|
|
if(cActor.BindRAGMemory(2, GetPointer(RAGMemory)))
|
|
return true;
|
|
PrintFormat("%s -> %d offline RAG Actor rebind failed", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
ReturnFalse;
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Detach before the layer destroys an old COW inference buffer. |
|
|
//+------------------------------------------------------------------+
|
|
bool DetachOfflineRAGMemory(void)
|
|
{
|
|
if(cActor.UnbindRAGMemory(2))
|
|
return true;
|
|
PrintFormat("%s -> %d offline RAG Actor detach failed", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
ReturnFalse;
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Publish only records that were consumed into an immutable |
|
|
//| snapshot |
|
|
//+------------------------------------------------------------------+
|
|
bool PublishOfflineMemory(const bool episode_closed = false)
|
|
{
|
|
const uint pending = RAGMemory.PendingRecordCount();
|
|
if(pending == 0)
|
|
return true;
|
|
if(!episode_closed && pending < OnlineMemorySize)
|
|
return true;
|
|
OfflinePublishAttempts++;
|
|
if(!DetachOfflineRAGMemory())
|
|
{
|
|
OfflinePublishFailures++;
|
|
ReturnFalse;
|
|
}
|
|
if(!RAGMemory.Publish(episode_closed))
|
|
{
|
|
if(!RebindOfflineRAGMemory())
|
|
{
|
|
OfflinePublishFailures++;
|
|
ReturnFalse;
|
|
}
|
|
OfflinePublishFailures++;
|
|
ReturnFalse;
|
|
}
|
|
if(!RebindOfflineRAGMemory())
|
|
{
|
|
OfflinePublishFailures++;
|
|
ReturnFalse;
|
|
}
|
|
if(!SaveOfflineBaseMemory())
|
|
{
|
|
OfflinePublishFailures++;
|
|
ReturnFalse;
|
|
}
|
|
OfflinePublishSuccesses++;
|
|
return true;
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Expert initialization function |
|
|
//+------------------------------------------------------------------+
|
|
int OnInit()
|
|
{
|
|
ResetLastError();
|
|
// Load Models.
|
|
float temp;
|
|
if(!cStateEncoder.Load(FileName + "StEnc.nnw", temp, temp, temp, dtStudied, true))
|
|
{
|
|
PrintFormat("Error of load StateEncoder: %d", GetLastError());
|
|
return INIT_FAILED;
|
|
}
|
|
CArrayObj *actor = new CArrayObj();
|
|
CArrayObj *critic = new CArrayObj();
|
|
if(!CreateDescriptions(actor, critic))
|
|
{
|
|
DeleteObj(actor)
|
|
DeleteObj(critic)
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
return INIT_FAILED;
|
|
}
|
|
if(!cActor.Load(FileName + "Act.nnw", temp, temp, temp, dtStudied, true))
|
|
{
|
|
Print("Create new Actor");
|
|
if(!cActor.Create(actor))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
DeleteObj(actor)
|
|
DeleteObj(critic)
|
|
return INIT_FAILED;
|
|
}
|
|
}
|
|
if(!cCritic.Load(FileName + CriticCheckpointFile, temp, temp, temp, dtStudied, true))
|
|
{
|
|
Print("Create new Critic");
|
|
if(!cCritic.Create(critic))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
DeleteObj(actor)
|
|
DeleteObj(critic)
|
|
return INIT_FAILED;
|
|
}
|
|
}
|
|
DeleteObj(actor)
|
|
DeleteObj(critic)
|
|
cActor.TrainMode(true);
|
|
cCritic.TrainMode(true);
|
|
cStateEncoder.TrainMode(false);
|
|
COpenCL *opencl = cActor.GetOpenCL();
|
|
cCritic.SetOpenCL(opencl);
|
|
cStateEncoder.SetOpenCL(opencl);
|
|
CLayerDescription memory_description;
|
|
if(!CreateRAGMemoryDescription(memory_description) ||
|
|
!RAGMemory.Init(0, 0, opencl, memory_description) ||
|
|
!RAGMemory.SetOnlineMemorySize(OnlineMemorySize))
|
|
return INIT_FAILED;
|
|
const bool snapshot_loaded = (LoadBaseMemorySnapshot &&
|
|
RAGMemory.LoadPublishedSnapshot(BaseMemorySnapshotPath));
|
|
if(!snapshot_loaded && !RAGMemory.BindNullInference())
|
|
return INIT_FAILED;
|
|
if(!RebindOfflineRAGMemory())
|
|
return INIT_FAILED;
|
|
if(!bGradient.BufferInit(EmbeddingSize, 0.0f) ||
|
|
!bGradient.BufferCreate(opencl))
|
|
return INIT_FAILED;
|
|
if(!Symb.Name(_Symbol) || !Symb.Refresh())
|
|
return INIT_FAILED;
|
|
if(!RSI.Create(Symb.Name(), TimeFrame, RSIPeriod, RSIPrice))
|
|
return INIT_FAILED;
|
|
if(!CCI.Create(Symb.Name(), TimeFrame, CCIPeriod, CCIPrice))
|
|
return INIT_FAILED;
|
|
if(!ATR.Create(Symb.Name(), TimeFrame, ATRPeriod))
|
|
return INIT_FAILED;
|
|
if(!MACD.Create(Symb.Name(), TimeFrame, FastPeriod, SlowPeriod, SignalPeriod, MACDPrice))
|
|
return INIT_FAILED;
|
|
cActor.GetLayerOutput(0, Result);
|
|
if(Result.Total() != AccountDescr)
|
|
{
|
|
PrintFormat("Input size of Actor doesn't match context description (%d <> %d)", Result.Total(), AccountDescr);
|
|
return INIT_FAILED;
|
|
}
|
|
cStateEncoder.GetLayerOutput(0, Result);
|
|
if(Result.Total() != (HistoryBars * BarDescr))
|
|
{
|
|
PrintFormat("Input size of StateEncoder doesn't match market state description (%d <> %d)", Result.Total(), (HistoryBars * BarDescr));
|
|
return INIT_FAILED;
|
|
}
|
|
cStateEncoder.GetLayerOutput(StateTokenLayer, Result);
|
|
if(Result.Total() != (BarDescr * EmbeddingSize))
|
|
{
|
|
PrintFormat("StateEncoder RankTCM layer doesn't match Critic context (%d <> %d)", Result.Total(), (BarDescr * EmbeddingSize));
|
|
return INIT_FAILED;
|
|
}
|
|
cStateEncoder.GetLayerOutput(StateScenarioLayer, Result);
|
|
if(Result.Total() != EmbeddingSize)
|
|
{
|
|
PrintFormat("StateEncoder token layer doesn't match pooled scenario embedding (%d <> %d)", Result.Total(), EmbeddingSize);
|
|
return INIT_FAILED;
|
|
}
|
|
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)
|
|
{
|
|
if(reason != REASON_INITFAILED && !PublishOfflineMemory(true))
|
|
Print("Error of publish Base Memory snapshot");
|
|
PrintOfflineRAGDiagnostics();
|
|
if(!(reason == REASON_INITFAILED || reason == REASON_RECOMPILE))
|
|
{
|
|
if(!cActor.Save(FileName + "Act.nnw", 0, 0, 0, TimeCurrent(), true))
|
|
PrintFormat("Error of save model: %s", "Actor");
|
|
if(!cCritic.Save(FileName + CriticCheckpointFile, 0, 0, 0, TimeCurrent(), true))
|
|
PrintFormat("Error of save model: %s", "Critic");
|
|
}
|
|
delete Result;
|
|
delete Action;
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| ChartEvent function |
|
|
//+------------------------------------------------------------------+
|
|
void OnChartEvent(const int id,
|
|
const long &lparam,
|
|
const double &dparam,
|
|
const string &sparam)
|
|
{
|
|
switch(id)
|
|
{
|
|
case 1001:
|
|
Train();
|
|
break;
|
|
case 1007:
|
|
Print("Event 1007");
|
|
if(!EventChartCustom(ChartID(), 1, 0, 0, "ChartEvent"))
|
|
{
|
|
PrintFormat("Error of create study event: %d", GetLastError());
|
|
ExpertRemove();
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
//| Train function |
|
|
//+------------------------------------------------------------------+
|
|
void Train(void)
|
|
{
|
|
ResetOfflineRAGDiagnostics();
|
|
int start = iBarShift(Symb.Name(), TimeFrame, Start);
|
|
int end = iBarShift(Symb.Name(), TimeFrame, End);
|
|
int bars = CopyRates(Symb.Name(), TimeFrame, 0, start, Rates);
|
|
if(bars < 0)
|
|
{
|
|
PrintFormat("%s -> %d CopyRates failed (%d)", __FUNCTION__, __LINE__, GetLastError());
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
if(!RSI.BufferResize(bars) || !CCI.BufferResize(bars) ||
|
|
!ATR.BufferResize(bars) || !MACD.BufferResize(bars))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
int count = -1;
|
|
bool calculated = false;
|
|
do
|
|
{
|
|
calculated = (RSI.BarsCalculated() >= bars &&
|
|
CCI.BarsCalculated() >= bars &&
|
|
ATR.BarsCalculated() >= bars &&
|
|
MACD.BarsCalculated() >= bars
|
|
);
|
|
Sleep(100);
|
|
count++;
|
|
}
|
|
while(!calculated && count < 100);
|
|
if(!calculated)
|
|
{
|
|
PrintFormat("%s -> %d The training data has not been loaded", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
RSI.Refresh();
|
|
CCI.Refresh();
|
|
ATR.Refresh();
|
|
MACD.Refresh();
|
|
if(!ArraySetAsSeries(Rates, true))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
bars -= end + HistoryBars + NForecast;
|
|
if(bars < 0)
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
vector<float> result, target, neg_target;
|
|
bool Stop = false;
|
|
uint ticks = GetTickCount();
|
|
for(int epoch = 0; (epoch < Epochs && !IsStopped() && !Stop); epoch ++)
|
|
{
|
|
if(!cActor.Clear() ||
|
|
!cCritic.Clear() ||
|
|
!cStateEncoder.Clear())
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
for(int posit = start - HistoryBars - NForecast - 1; posit >= end; posit--)
|
|
{
|
|
if(!CreateBuffers(posit, GetPointer(bState), GetPointer(bTime), Result))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
const vector<float> account = SampleAccount(GetPointer(bState), datetime(bTime[0]), MaxBalance, MinBalance);
|
|
const vector<float> target_action = OraculAction(account, Result);
|
|
if(!bContext.AssignArray(account))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
ExpertRemove();
|
|
return;
|
|
}
|
|
// Feed Forward.
|
|
if(!cStateEncoder.feedForward((CBufferFloat*)GetPointer(bState), 1, false, (CBufferFloat*)NULL))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!RAGMemory.Retrieve(GetPointer(cStateEncoder), StateScenarioLayer))
|
|
{
|
|
PrintFormat("%s -> %d RAG retrieval failed", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!cActor.feedForward((CBufferFloat*)GetPointer(bContext), 1, false,
|
|
GetPointer(cStateEncoder), StateScenarioLayer))
|
|
{
|
|
PrintFormat("%s -> %d Actor forward failed", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
#ifdef _DEBUG
|
|
vector<float> res;
|
|
cActor.getResults(res);
|
|
#endif
|
|
if(!cCritic.feedForward(GetPointer(cActor), -1, GetPointer(cStateEncoder), StateTokenLayer))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
#ifdef _DEBUG
|
|
cCritic.getResults(res);
|
|
#endif
|
|
// Study.
|
|
cActor.getResults(Action);
|
|
double balance = account[0] * EtalonBalance;
|
|
const uint terminal_position = uint(posit - NForecast + 1);
|
|
SRAGTerminalOutcome terminal;
|
|
OfflineCollectionChecks++;
|
|
bool terminal_evaluated = EvaluateTerminalOutcome(RAGMemory, Action, balance, terminal_position, terminal);
|
|
double reward = terminal.reward * balance / EtalonBalance;
|
|
if(!terminal_evaluated)
|
|
{
|
|
OfflineTerminalFailures++;
|
|
CaptureOfflineRAGRejection("terminal", terminal_position);
|
|
}
|
|
else
|
|
if(MathAbs(Action[0] - Action[3]) < OfflineActionNonZeroThreshold)
|
|
{
|
|
OfflineRawLotRejections++;
|
|
CaptureOfflineRAGRejection("lot", terminal_position);
|
|
}
|
|
else
|
|
{
|
|
CBufferFloat *scenario_embedding = NULL;
|
|
if(!cStateEncoder.GetLayerOutputDevice(StateScenarioLayer, scenario_embedding))
|
|
{
|
|
OfflineScenarioAcquireFailures++;
|
|
CaptureOfflineRAGRejection("scenario_acquire", terminal_position);
|
|
}
|
|
else
|
|
if(scenario_embedding.Total() != EmbeddingSize)
|
|
{
|
|
OfflineScenarioShapeFailures++;
|
|
CaptureOfflineRAGRejection("scenario_shape", terminal_position);
|
|
}
|
|
else
|
|
{
|
|
float normalized_action[NActions];
|
|
ENUM_MEMORY_ACTION_NORMALIZATION_STAGE normalization_stage;
|
|
if(NormalizeMemoryAction(RAGMemory, Action, terminal_position, balance, normalized_action,
|
|
normalization_stage))
|
|
{
|
|
bool materially_changed = !OfflineHasLastStoredAction;
|
|
for(int i = 0; i < NActions && !materially_changed; i++)
|
|
materially_changed = (MathAbs(normalized_action[i] - OfflineLastStoredAction[i]) >= OfflineActionChangeThreshold);
|
|
if(materially_changed)
|
|
{
|
|
float scenario_values[];
|
|
uint replaced_values = 0;
|
|
const bool scenario_read = RAGMemory.ReadScenarioEmbedding(scenario_embedding,
|
|
scenario_values, replaced_values);
|
|
OfflineScenarioReplacedValues += replaced_values;
|
|
if(!scenario_read)
|
|
{
|
|
OfflineScenarioReadFailures++;
|
|
CaptureOfflineRAGRejection("scenario_read", terminal_position);
|
|
PrintFormat("%s -> %d scenario anchor read failed", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!RAGMemory.AddCompletedRecord(scenario_values, normalized_action, terminal))
|
|
{
|
|
OfflineRecordFailures++;
|
|
CaptureOfflineRAGRejection("add_record", terminal_position);
|
|
PrintFormat("%s -> %d offline RAG record rejected", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
OfflineAcceptedRecords++;
|
|
ArrayCopy(OfflineLastStoredAction, normalized_action);
|
|
OfflineHasLastStoredAction = true;
|
|
}
|
|
else
|
|
{
|
|
OfflineDedupeRejections++;
|
|
CaptureOfflineRAGRejection("dedupe", terminal_position);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
OfflineNormalizationFailures++;
|
|
switch(normalization_stage)
|
|
{
|
|
case MEMORY_ACTION_NORMALIZATION_INPUT:
|
|
OfflineNormalizationInputFailures++;
|
|
break;
|
|
case MEMORY_ACTION_NORMALIZATION_MARKET:
|
|
OfflineNormalizationMarketFailures++;
|
|
break;
|
|
case MEMORY_ACTION_NORMALIZATION_VOLATILITY:
|
|
OfflineNormalizationVolatilityFailures++;
|
|
break;
|
|
case MEMORY_ACTION_NORMALIZATION_DESCRIPTOR:
|
|
OfflineNormalizationDescriptorFailures++;
|
|
break;
|
|
case MEMORY_ACTION_NORMALIZATION_ZERO:
|
|
OfflineNormalizationZeroFailures++;
|
|
break;
|
|
}
|
|
CaptureOfflineRAGRejection("normalize_" + MemoryActionNormalizationStageName(normalization_stage),
|
|
terminal_position);
|
|
}
|
|
}
|
|
}
|
|
Result.Clear();
|
|
if(!Result.Add(float(reward)))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!cCritic.backProp(Result, GetPointer(cStateEncoder), StateTokenLayer))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
// Oracul.
|
|
if(!Action.AssignArray(target_action))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
reward = CheckAction(RAGMemory, Action, balance, posit - NForecast + 1) * balance / EtalonBalance;
|
|
CBufferFloat *actor_market_gradient = NULL;
|
|
if(!cStateEncoder.GetLayerOutputDevice(StateScenarioLayer, actor_market_gradient))
|
|
{
|
|
PrintFormat("%s -> %d actor market gradient input failed", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(reward > 0)
|
|
if(!cActor.backProp(Action, actor_market_gradient, GetPointer(bGradient)))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!cCritic.feedForward(Action, 1, false, GetPointer(cStateEncoder), StateTokenLayer))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!Result.Update(0, float(reward)))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(!cCritic.backProp(Result, GetPointer(cStateEncoder), StateTokenLayer))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
cCritic.TrainMode(false);
|
|
if(!cActor.feedForward((CBufferFloat*)GetPointer(bContext), 1, false,
|
|
GetPointer(cStateEncoder), StateScenarioLayer) ||
|
|
!cCritic.feedForward(GetPointer(cActor), -1, GetPointer(cStateEncoder), StateTokenLayer) ||
|
|
!cCritic.backProp(Result, GetPointer(cStateEncoder), StateTokenLayer) ||
|
|
!cActor.backPropGradient(actor_market_gradient, GetPointer(bGradient), -1, true))
|
|
{
|
|
PrintFormat("%s -> %d", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
cCritic.TrainMode(true);
|
|
if(RAGMemory.PendingRecordCount() >= OnlineMemorySize && !PublishOfflineMemory(false))
|
|
{
|
|
PrintFormat("%s -> %d offline memory publication failed", __FUNCTION__, __LINE__);
|
|
Stop = true;
|
|
break;
|
|
}
|
|
if(GetTickCount() - ticks > 500)
|
|
{
|
|
double percent = (epoch + 1.0 - double(posit - end) / (start - end - HistoryBars - NForecast)) / Epochs * 100.0;
|
|
string str = "";
|
|
str += StringFormat("%-12s %6.2f%% -> Error %15.8f\n", "Actor", percent, cActor.getRecentAverageError());
|
|
str += StringFormat("%-12s %6.2f%% -> Error %15.8f\n", "Critic", percent, cCritic.getRecentAverageError());
|
|
str += StringFormat("RAG scenarios %u, pending %u\n", RAGMemory.ScenarioCount(), RAGMemory.PendingRecordCount());
|
|
str += StringFormat("RAG collect accepted %I64u, lot %I64u, normalize %I64u, dedupe %I64u\n",
|
|
OfflineAcceptedRecords, OfflineRawLotRejections,
|
|
OfflineNormalizationFailures, OfflineDedupeRejections);
|
|
Comment(str);
|
|
ticks = GetTickCount();
|
|
}
|
|
}
|
|
if(!Stop && !PublishOfflineMemory(true))
|
|
PrintFormat("%s -> %d offline memory publication failed", __FUNCTION__, __LINE__);
|
|
}
|
|
Comment("");
|
|
PrintFormat("%s -> %d -> %-15s %10.7f", __FUNCTION__, __LINE__, "Actor", cActor.getRecentAverageError());
|
|
PrintFormat("%s -> %d -> %-15s %10.7f", __FUNCTION__, __LINE__, "Critic", cCritic.getRecentAverageError());
|
|
ExpertRemove();
|
|
}
|
|
//+------------------------------------------------------------------+
|
|
|