MQL5 MACHINE LEARNING 02: Deep Learning For Algo-Trading

Identify ranging markets with a perceptron-based neural network using the 30 and 50 moving averages to mark bullish, bearish, or range conditions; explain why deep networks require multiple layers.

Build a neural network in Excel to detect energy market range conditions using two inputs and moving averages, with a hidden layer, weight initialization, sigmoid activation, and output predictions.

Learn to perform backpropagation in the output layer by computing how weights and a bias affect the cost, using sigmoid derivatives to update weights and reduce loss.

Calculate how weight three, weight four, and the bias affect the cost in a hidden layer using backpropagation, tracing activation effects through weighted sums and activation derivatives to guide updates.

Calculate effects of weight one, weight two, and bias one on cost via backpropagation for node three, using activation derivative and weighted sum terms to update weights and bias.

Practice backpropagation through multiple epochs to update weights and biases using gradient descent, track loss reductions, and explore deeper neural networks for algo-trading in the MQL5 platform.

Design a neural network to classify market direction as bullish or bearish using MacD and RSI, with two inputs, three hidden layers, and output nodes for bullish and bearish predictions.

Build a deep neural network in mql5 by creating an expert advisor, configuring inputs and chart appearance, and implementing initialization and on-tick logic with new-bar detection.

Create a tensor in mql5 by building a c matrix class with a public matrix and a tensor class that stores an array of matrices, plus add and get functions.

Initialize the tensor size, resize the matrices array, and create matrix pointers in the constructor; validate pointers, get last error, and in the destructor delete valid matrices and free memory.

Define and implement a tensor class with add and get functions to store and retrieve matrices, validating indices and assigning matrix contents into the tensor's matrix member.

Declare and manage c tensor objects for weights, bias, inputs, and outputs, check validity, and delete them to prepare tensors for neural network data flow.

Get data from macd histogram and rsi indicators, set ema 12/26 and signal 9, configure rsi 18, and build a 5000-path, 3-column training matrix for bullish or bearish candlestick prediction.

Create a function to collect, train, and test a neural network by building a three-column dataset matrix with MacD histogram, RSI, and a bullish or bearish target.

Develop a live trading pipeline that runs on each new price quote, building an inputs vector from macd and rsi indicators using copy buffer.

Split the data matrix into training and testing sets by implementing a randomize function with seed 42, using Fisher's algorithm to shuffle rows and ensure reproducible training.

Create a custom copy function to copy from one vector to another in MQL5, using copy from, copy to, start from, and total, returning a boolean and resizing the destination.

Implement a data split function that randomizes the dataset and partitions it into x_train, y_train, x_test, y_test with a 0.7 training ratio to train a neural network and prevent overfitting.

Define a random function in MQL5 that generates numbers within a specified range by using Rand, normalizing to 0–1, scaling by (max−min), and offsetting by the minimum, with double typecasting.

generate and initialize random tensor parameters for a neural network by creating weight and bias tensors, seeded randomization, and layer-wise population of values between -1 and 1.

Configure hidden layers by declaring the number of layers and neurons per layer, storing sizes in a vector (e.g., 15, 10, 7). Warn against overfitting in forex time series.

Execute the neural network forward pass by applying weights, biases and activation functions across hidden layers, using sigmoid, tanh, leaky ReLU, and ReLU, with softmax on the output layer.

Implement a single forward pass to process a vector of inputs through a neural network and return the class with the highest probability, and a batch forward pass for matrices.

Begin backpropagation by selecting a loss function (binary, categorical cross-entropy, mean squared error, hinge), configure inputs (epochs, learning rate), and initialize neural network parameters for epoch-by-epoch training for algo-trading.

Implement backpropagation by looping through epochs, performing forward passes, computing loss gradients for the output with softmax, and updating weights and biases via the learning rate.

Split 70% training and 30% testing data, normalize inputs with min max normalization, train the neural network with back propagation, then evaluate with a confusion matrix.

Create an arrow drawing function to render chart arrows representing neural network signals, using parameters like name, time, price, arrow code, color, style, and width.

Display neural network signals by normalizing MACD and RSI inputs, performing a forward pass to generate buy or sell signals, and plotting timestamped arrows on the chart to prevent repaint.

Test and debug a neural network for algo-trading by compiling code, fixing input issues, correcting backpropagation parameters, and diagnosing weights, bias, and randomization with RSI and MACD signals.

Test the neural network to assess reliability, debug coding errors, and evaluate training, while noting bullish/bearish signals and MacD limitations for improved accuracy.

Learn to build and apply deep neural networks in forex trading using Mql5, explore convolutional and recurrent networks, and master data processing techniques like feature engineering and normalization.