as q neural net pure mq l5 neural network library

Related: MLP Neural Network Class for MetaTrader - another powerful library for MetaTrader 5 traders.

asq neuralnet is a comprehensive neural network library developed entirely in native mql5, eliminating the need for external dependencies like dlls or python bridges.

Also recommended: rbf Neural Network Class - similar library with strong performance on MetaTrader 5.

this library empowers mql5 developers with a full-featured deep learning framework, comprising:

- a dense matrix algebra engine with over 40 operations, including multiplication, transpose, hadamard, he/xavier initialization, and nan detection.
- 13 activation functions with analytical derivatives: relu, leakyrelu, elu, selu, sigmoid, tanh, softmax, swish, mish, gelu, softplus, hardsigmoid, and linear.
- dense layers supporting forward and backward propagation, dropout, and gradient clipping.
- 3 optimizers: sgd with momentum, adam, and adamw (decoupled weight decay).
- 7 learning rate schedulers: constant, step decay, exponential, cosine annealing, linear, reduceonplateau, warmup, and cyclic lr.
- 5 loss functions: mse, mae, huber, cross-entropy, and binary cross-entropy.
- a complete training pipeline with mini-batch sgd, fisher-yates shuffle, and epoch logging.

quick start:

building a neural network is straightforward, requiring just 6 lines of code. here's an example:

cneuralnetwork net;
net.init(32); // 32 input features
net.addlayer(64, act_relu); // hidden layer 1
net.addlayer(32, act_relu, 0.2); // hidden layer 2 + dropout
net.addlayer(3, act_softmax); // output: buy/sell/hold
net.build();

training is equally simple, with a single function call:

net.setoptimizer(opt_adam, 0.001);
net.setloss(loss_cross_entropy);
net.fit(trainx, trainy, 100, 32);

prediction is also a one-line operation:

int action = net.predictclass(features); // 0=buy, 1=sell, 2=hold

use cases:

- train classification models for trading signals (buy/sell/hold).
- perform price direction regression using mse or huber loss.
- recognize patterns in candlestick formations.
- detect market regimes (trending, ranging, volatile).
- analyze feature importance.
- approximate q-value functions for reinforcement learning agents.

performance:

- inference latency: < 0.1ms for typical architectures with under 1000 parameters.
- memory usage is proportional to the total number of parameters (e.g., 5kb for a 32→64→32→3 network).
- no dynamic allocation during inference.
- numerical stability features: nan detection, gradient clipping, and safe softmax with max-subtraction.

installation:

place the 5 library files in mql5/include/algosphere/neuralnet/ and include the main header:

#include <algosphere/neuralnet/nn_network.mqh>

the demo script showcases matrix operations, activation functions, xor classification, and synthetic market direction prediction.

library files:

- nn_matrix.mqh (908 lines) - dense matrix algebra engine.
- nn_activations.mqh (300 lines) - 13 activations + derivatives.
- nn_layer.mqh (374 lines) - dense layer with forward/backward/dropout.
- nn_optimizer.mqh (454 lines) - sgd/adam/adamw + 7 lr schedulers.
- nn_network.mqh (734 lines) - complete feedforward network with training.
- asq_neuralnet_demo.mq5 (283 lines) - 4 runnable demonstrations.

total: 3,053 lines of pure mql5.

technical notes:

- weight initialization: he init for relu-family, xavier for sigmoid/tanh.
- box-muller transform for normal distribution (mql5 native mathrand).
- softmax + cross-entropy gradient shortcut (ŷ - y) to avoid full jacobian.
- inverted dropout (scaled during training, identity during inference).
- fisher-yates shuffle for mini-batch training.
- gradient norm clipping per layer (default max norm = 1.0).

You may also like: PNN Neural Network Class - excellent alternative for library users on MetaTrader 5.

#property copyright "robotfx"
#property link "https://robotfx.org"
#property version   "1.00"
#property script_show_inputs
#include "asq_neuralnet.mqh"
input int    inpepochs     = 500;
input double inplr         = 0.01;
input bool   inpverbose    = true;
void onstart()
{

.......

• This tool is provided for educational and testing purposes only.
• Past performance does not guarantee future results.
• Trading involves substantial risk of loss. Use on a demo account first.
• Results may vary depending on market conditions, broker, and settings.
• We recommend thorough backtesting and forward testing before using with real funds.