NNMQL5 — Lightweight Neural Network DLL for MetaTrader 5 (x64, MSVC)

A lightweight, dependency-free C++ DLL that brings dense multilayer perceptrons (MLP) and batch training wrappers directly into MQL5.
Ideal for time-series analysis and custom indicator prototyping — no Python runtime or external frameworks required.

Build: Visual Studio 2022 • x64 • MSVC
License: MIT-like spirit — use freely, please credit the author.
Author: Tomáš Bělák — Remind
Downloads & Documentation: remind.cz/neural-networks-in-mql5

---

Features

• C ABI for painless import in MQL5
• Dense MLP with SGD, He/Xavier initialization, gradient clipping
• Activations: SIGMOID, TANH, RELU, LINEAR, SYM_SIG
• Batch training & inference: NN_ForwardBatch, NN_TrainBatch
• Weights access: NN_GetWeights, NN_SetWeights
• Internal tensor utilities (sliding windows, tensor→matrix)
• Conv1DLayer (forward only, internal prototype)

---

Quick Start (MQL5)

1. Install the DLL

Copy NNMQL5.dll to:

%APPDATA%\MetaQuotes\Terminal\<your-terminal-id>\MQL5\Libraries\

In MT5 enable: Tools → Options → Expert Advisors → Allow DLL imports.

---

2. Import functions in your .mq5

#import "NNMQL5.dll"
int  NN_Create();
void NN_Free(int h);
bool NN_AddDense(int h, int inSz, int outSz, int act);
int  NN_InputSize(int h);
int  NN_OutputSize(int h);
bool NN_Forward(int h, const double& in[], int in_len, double& out[], int out_len);
bool NN_TrainOne(int h, const double& in[], int in_len,
                 const double& tgt[], int tgt_len, double lr, double& mse);
bool NN_ForwardBatch(int h, const double& in[], int batch, int in_len,
                     double& out[], int out_len);
bool NN_TrainBatch(int h, const double& in[], int batch, int in_len,
                   const double& tgt[], int tgt_len, double lr, double& mean_mse);
bool NN_GetWeights(int h, int i, double& W[], int Wlen, double& b[], int blen);
bool NN_SetWeights(int h, int i, const double& W[], int Wlen, const double& b[], int blen);
#import

---

3. Create a network

int h = NN_Create();
NN_AddDense(h, 240, 64, 1);   // ReLU
NN_AddDense(h, 64, 15, 3);    // Linear

double x[240], y[15]; 
ArrayInitialize(x, 0.0);

NN_Forward(h, x, 240, y, 15);
NN_Free(h);

---

4. Train one sample

double mse;
NN_TrainOne(h, x, 240, t, 15, 0.001, mse);

---

5. Train batch

const int BATCH=32, IN=240, OUT=15;
double X[BATCH*IN], T[BATCH*OUT], mean_mse;

NN_TrainBatch(h, X, BATCH, IN, T, OUT, 0.001, mean_mse);

---

API Reference

Lifecycle

• NN_Create, NN_Free

Topology

• NN_AddDense
  • Activation codes: 0=SIGMOID, 1=RELU, 2=TANH, 3=LINEAR, 4=SYM_SIG

Introspection

• NN_InputSize, NN_OutputSize

Inference

• NN_Forward, NN_ForwardBatch

Training

• NN_TrainOne, NN_TrainBatch

Weights

• NN_GetWeights, NN_SetWeights

---

Example: Sliding Window Forecast

int h = NN_Create();
NN_AddDense(h, 240, 128, 1);  // ReLU
NN_AddDense(h, 128, 15, 3);   // Linear

double x[240], t[15], y[15], mse;

// Train 10 epochs
for(int e=0; e<10; e++)
   NN_TrainOne(h, x, 240, t, 15, 0.001, mse);

NN_Forward(h, x, 240, y, 15);
NN_Free(h);

---

Build From Source

• Toolchain: Visual Studio 2022, MSVC, C++17, x64
• Configuration: Release | x64
• Entry: dllmain.cpp (precompiled header pch.cpp/h)
• Outputs: NNMQL5.dll (+ optional .lib, not required for MT5)
• Note: Compiler warning D9007 /C can be safely ignored.

---

Data Conventions

• Type: double
• Shape: flat vectors, batch = [batch × in_len]
• Loss: MSE
• Optimizer: SGD with gradient clipping ±5
• Init: He (ReLU) / Xavier (others)
• Normalization: scale inputs/targets (e.g., [-1,1])
• Determinism: initial weights random (std::rand()); set via NN_SetWeights for reproducibility

---

Roadmap

• Conv1D backpropagation
• Momentum / Adam optimizers
• FP32/FP64 switch
• Save / Load weights

---

License

MIT-like spirit — free for commercial and non-commercial use.
Please attribute:

Author: Tomáš Bělák — Remind