repository·main·Indexed 42 days ago
https://github.com/megajoctan/male5A Python-inspired machine learning and data science library for the MQL5 programming language, enabling advanced data processing within MetaTrader 5. It includes implementations for dimensionality reduction (CLDA, CPCA, CNMF, CTruncatedSVD), ensemble learning (AdaBoost, CRandomForestClassifier, CRandomForestRegressor), and linear models (CLinearRegression, CLogisticRegression, CPolynomialRegression).
CNaiveBayes class implements a Naive Bayes classifier in MQL5. It is a probabilistic classifier that assumes feature independence to simplify prediction calculations. You can use it to train a model on existing data and then predict labels for new feature sets.Depending on your data and goal, you should choose one of the following models:
CLinearRegression): Use when you want to predict a continuous value and assume a proportional relationship between independent and dependent variables ($y = \beta_0 + \beta_1x_1 + ...$).CLogisticRegression): Use for binary classification (predicting 0 or 1). It uses a sigmoid function to map inputs to a probability between 0 and 1.CPolynomialRegression): Use when the relationship between variables is non-linear. It fits a polynomial function ($y = \beta_0 + \beta_1x + \beta_2x^2 + ...$) to capture curves in the data.When using CLDA or CPCA, you can specify how the number of components is determined using the m_criterion member variable via the CRITERION_ enum:
CRITERION_VARIANCE: Retains components that explain a specific percentage of variance (defined by a threshold).CRITERION_KAISER: Retains components with eigenvalues greater than 1.CRITERION_SCREE_PLOT: Uses visual analysis of the scree plot to determine significant eigenvalues.The AdaBoost class in adaboost.mqh implements the Adaptive Boosting algorithm, which combines multiple weak learners into a single strong learner. It works by iteratively training learners on a weighted dataset, where misclassified points receive higher weights in subsequent iterations.
Key features:
DecisionTree and LogisticRegression namespaces.random_state, and bootstrapping behavior.// Example conceptual usage of AdaBoost
// Note: Specific implementation details depend on the chosen namespace (DecisionTree or LogisticRegression)
AdaBoost model(50, 42, true);
model.fit(x_data, y_labels);
int prediction = model.predict(new_x_vector);To use ML5 in your MQL5 projects, you must manually install the library files into your MetaTrader 5 terminal directory.
ML5 folder within the extracted contents.ML5 folder into the Include directory of your MetaTrader 5 terminal data path.Your directory structure should look like: [MetaTrader5 Data Path]/MQL5/Include/ML5/.
The CDecisionTreeRegressor class is used for regression problems (predicting continuous numerical values). It specializes in maximizing variance reduction during the splitting process.
Hyperparameters:
min_samples_split: The minimum number of samples required to split an internal node.max_depth: The maximum depth of the tree.Core Methods:
fit(matrix &x, vector &y): Trains the model using a matrix of independent variables x and a vector of continuous target values y.predict(const vector &x): Returns the predicted continuous value for a single data point.// Example initialization and training
CDecisionTreeRegressor regressor(2, 5);
regressor.fit(features_matrix, target_values_vector);
// Predict a continuous value
double value = regressor.predict(new_sample_vector);The CPolynomialRegression class extends linear regression by fitting a higher-degree polynomial function, allowing it to capture complex non-linear relationships.
Set the polynomial degree during construction:
CPolynomialRegression(int degree=2)
fit(matrix &x, vector &y): Trains the model using the specified degree.predict(vector &x) or predict(matrix &x): Predicts the dependent variable for new data points.You can use the Bayesian Information Criterion (BIC) to find the optimal polynomial degree to avoid overfitting:
void BIC(ulong k, vector &bic, int &best_degree)
k: A vector of degree values to evaluate.bic: Output vector where BIC values for each degree in k are stored.best_degree: Output variable that will store the recommended degree (the one with the lowest BIC).// Initialize with degree 3
CPolynomialRegression model(3);
model.fit(x_data, y_data);
// Selecting the best degree using BIC
vector degrees = {1, 2, 3, 4, 5};
vector bic_values;
int best_deg;
CPolynomialRegression bic_helper;
bic_helper.BIC(degrees, bic_values, best_deg);To train the classifier, use the fit(matrix &x, vector &y) method.
x: A matrix representing the input features.y: A vector representing the target labels.During training, the class calculates the prior probability of each class and estimates the conditional probability of each feature value given each class (utilizing Laplace smoothing to handle unseen features).
void fit(matrix &x, vector &y);The CRandomForestClassifier class implements a random forest ensemble specifically for classification tasks. It builds multiple decision trees using random subsets of features and bootstrapped samples to reduce variance and prevent overfitting.
CRandomForestClassifier(uint n_trees=100, uint minsplit=NULL, uint max_depth=NULL, int random_state=-1)
n_trees: Number of trees in the forest.minsplit: Minimum samples required to split a node.max_depth: Maximum depth of the trees.random_state: Seed for reproducibility.void fit(matrix &x, vector &y, bool replace=true, errors_classifier err=ERR_ACCURACY):x: Independent variables.y: Class labels.replace: If true, uses bootstrapping (sampling with replacement).err: Error metric for training evaluation (default: ERR_ACCURACY).double predict(vector &x): Predicts the class label for a single data point.vector predict(matrix &x): Predicts class labels for multiple data points.The CLinearRegression class implements linear regression to model the relationship between a dependent variable and one or more independent variables. It supports two training methods: Least Squares for direct optimization and Gradient Descent for iterative optimization.
fit_LeastSquare(matrix &x, vector &y): Fits the model by minimizing the sum of squared errors using the least squares method.fit_GradDescent(matrix &x, vector &y, double alpha, uint epochs = 1000): Fits the model iteratively using gradient descent. You must provide a learning rate alpha and can optionally specify the number of epochs.Once trained, you can predict values using:
predict(vector &x): For a single new data point.predict(matrix &x): For multiple new data points (each row in the matrix is a data point).CLinearRegression model;
model.fit_GradDescent(x_data, y_data, 0.01, 500);
// Predict a single point
vector single_point = ...;
double prediction = model.predict(single_point);
// Predict multiple points
matrix multiple_points = ...;
vector predictions = model.predict(multiple_points);The CTruncatedSVD class implements Truncated Singular Value Decomposition for dimensionality reduction and feature extraction.
CTruncatedSVD(uint k=0)
k: Number of top singular values/components to retain. If 0, it is determined automatically.matrix fit_transform(matrix& X): Trains the model on X and returns the transformed data.matrix transform(matrix &X): Transforms new data X using the trained model.vector transform(vector &X): Transforms a single data point X using the trained model.// Example usage of CTruncatedSVD
CTruncatedSVD svd(10);
matrix reduced_X = svd.fit_transform(X);