MlModel

Namespace: MLLIB  ·  Library: Machine Learning Library

#include <ml/ml_model.h>

class MLLIB::MlModel

Backend-agnostic inference interface: load, predict, save plus model/task descriptors.

src/examples/ex_ml/main.cpp

std::unique_ptr<MlModel> model = std::make_unique<MlOnnxModel>();
const bool loaded = model->load(MNE_EX_ML_DATA_DIR "/affine_softmax.onnx");
const MlTensor probabilities = model->predict(batch); // shape {2, 2}, one row per sample
const QString classes = static_cast<MlOnnxModel&>(*model).metadata("classes");

Inheritance

---

Public Methods

~MlModel()

Virtual destructor.

---

predict(input)

Run inference on the given input tensor.

Parameters:

• input : const MlTensor & The input data.

Returns:

• MlTensor — The prediction result.

---

save(path)

Serialise the model to disk.

Parameters:

• path : const QString & File path.

Returns:

• bool — True if successful.

---

load(path)

Load a model from disk.

Parameters:

• path : const QString & File path.

Returns:

• bool — True if successful.

---

modelType()

Returns:

• QString — Human-readable model type name.

---

taskType()

Returns:

• MlTaskType — The task type this model is configured for.

---

Example

• Full source: src/examples/ex_ml/main.cpp
• Build target: ex_ml (runs under CTest)

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 85

#ifdef MNE_USE_ONNXRUNTIME
    std::unique_ptr<MlModel> model = std::make_unique<MlOnnxModel>();
    const bool loaded = model->load(MNE_EX_ML_DATA_DIR "/affine_softmax.onnx");
    const MlTensor probabilities = model->predict(batch); // shape {2, 2}, one row per sample
    const QString classes = static_cast<MlOnnxModel&>(*model).metadata("classes");
    // NumPy: softmax(x @ W + b) = [[0.6681878, 0.33181223], [0.00235232, 0.9976477]]
    Matrix2f expected;
    expected << 0.6681878f, 0.33181223f, 0.00235232f, 0.9976477f;
    ok &= expect(loaded && model->modelType() == "onnx" && probabilities.shape() == std::vector<int64_t>({2, 2}) && (probabilities.toMatrixXf() - expected).cwiseAbs().maxCoeff() < 1e-6f && classes == "rest,task",
                 QString("MlOnnxModel: p(task) = %1, %2 match NumPy; classes '%3'").arg(probabilities.matrix()(0, 1)).arg(probabilities.matrix()(1, 1)).arg(classes));
#else
    qInfo().noquote() << "  skip  MlOnnxModel (built without ONNX Runtime)";

Authors of this file

• Christoph Dinh <christoph.dinh@mne-cpp.org>