Machine Learning for Embbedded Devices

sklearn2c is a tool that converts scikit-learn library classification algorithms to C code. It can be used to generate C code from trained models, which can then be used in microcontrollers or other embedded systems. The generated code can be used for real-time classification tasks, where the computational resources are limited.

Supported Models

Classification

• Bayes Classifier*

• Decision Trees

• KNN Classifier

• C-SVC**

  *: sklearn2c does not use scikit-learn GaussianNB(), instead it uses the following cases to compute decision function.

  **: linear, poly and rbf kernels are supported.

Regression

• Linear Regression
• Polynomial Regression
• KNN
• Decision Trees

Clustering

• kmeans
• DBSCAN

Installation

You can install the library via pip either using:

pip install sklearn2c

or

pip install git+git@github.com:EmbeddedML/sklearn2c.git

Alternatively, you can install conda package:

conda install sklearn2c or mamba install sklearn2c

Usage

Please check examples directory under this repository. For example, decision tree classifier is created as follows:

• train method trains the model and optionally saves the model file to save_path. This method is totally compatible with scikit-learn library.
• predict method runs the model on the given data.
• static method load loads the model from saved path.
• export method generates model parameters as C functions.

dtc = DTClassifier()
dtc.train(train_samples, train_labels, save_path="<path/to/model>")
dtc.predict(test_samples)
dtc2 = DTClassifier.load(dtc_model_dir)
dtc2.export("<path/to/config_dir>")

For inference on C(specifically for STM32 boards), you can take a look at STM32_inference directory for the corresponding model.

License

MIT