Skip to main content

MLTrainer

Namespace: MLLIB  ·  Library: Machine Learning Library

#include <ml/ml_trainer.h>

class MLLIB::MLTrainer

Launches Python training scripts via PythonRunner with automatic venv handling and prerequisite checks.

Training logic (PyTorch, scikit-learn, MNE-Python, …) lives entirely in the .py scripts under scripts/ml/training/ and runs in a child process; no Python embedding or libpython linkage is required. If the configured PythonRunnerConfig carries a venvDir, the runner creates and updates the venv before each script invocation so dependencies are pinned per training job.

src/examples/ex_ml/main.cpp
QTemporaryDir dir;
QFile script(dir.filePath("train.py"));
const bool written = script.open(QIODevice::WriteOnly) && script.write("import sys\nprint('trained on', sys.argv[1], 'epochs')\n") > 0;
script.close();
MLTrainer trainer; // runs python3; a venv and pip install are configured via UTILSLIB::PythonRunnerConfig
const UTILSLIB::PythonRunnerResult result = trainer.run(script.fileName(), {"5"});

Public Methods​

MLTrainer()​

Construct an MLTrainer with default PythonRunner configuration.

MLTrainer(config)​

Construct an MLTrainer with explicit PythonRunner configuration.

Parameters:

runner()​

Access the underlying PythonRunner for callback/config changes.

Returns:

  • PythonRunner & — Reference to the PythonRunner instance.

run(scriptPath, args)​

Run a training script.

If the PythonRunner is configured with a venvDir and packageDir, the venv is created/updated automatically before the script runs.

Parameters:

  • scriptPath : const QString & Path to the .py training script.

  • args : const QStringList & Arguments forwarded to the script.

Returns:

checkPrerequisites(packages)​

Check that the required Python packages are importable.

Parameters:

  • packages : const QStringList & List of package names (e.g. {"torch", "mne"}).

Returns:

  • QStringList — List of packages that could NOT be imported (empty = all OK).

Example​

Used in examples​

ex_ml​

ML library: tensors, ONNX inference through the MlModel interface, and the Python trainer.

src/examples/ex_ml/main.cpp, line 103
QFile script(dir.filePath("train.py"));
const bool written = script.open(QIODevice::WriteOnly) && script.write("import sys\nprint('trained on', sys.argv[1], 'epochs')\n") > 0;
script.close();
MLTrainer trainer; // runs python3; a venv and pip install are configured via UTILSLIB::PythonRunnerConfig
const UTILSLIB::PythonRunnerResult result = trainer.run(script.fileName(), {"5"});
if (trainer.runner().isPythonAvailable()) {
    ok &= expect(written && result.success && result.stdOut.trimmed() == "trained on 5 epochs" && trainer.checkPrerequisites({"json", "no_such_package_xyz"}) == QStringList({"no_such_package_xyz"}),
                 "MLTrainer runs the script with its arguments and reports the missing package");
} else {
    qInfo().noquote() << "  skip  MLTrainer (python3 not found)";
}

Authors of this file​