MicroML

Micro-ML is a plain C library for machine learning on edge devices (it can be used also on desktops). It currently implements the following ML algorithms:

• Linear SVM
• Decision Tree
• K-NN
• ANN. This algorithm requires the X-CUBE-AI library for STM32 microcontrollers. Please, check instructions in Desk-ML

We are extending the library to other algorithms, also unsupervised. Your voluntary contribution is welcome.

At present, the library is optimized for memory usage. The algorithms parameters are stored in source files created by Desk-LM, which performs the training phase.

We are currently working to provide dynamic configurability to all the algorithms.

Reference article for more infomation

F., Sakr, F., Bellotti, R., Berta, A., De Gloria, "Machine Learning on Mainstream Microcontrollers," Sensors 2020, 20, 2638. https://www.mdpi.com/1424-8220/20/9/2638

Usage

You can compile the code as an executable or as a static library, using gcc/g++ for a Microcontroller or a desktop (e.g., thorugh Eclipse CDT or Visual Studio Code).

The program expects to find files produced by Desk-LM for a given algorithm and dataset to be available in ./ds/source and in ./ds/include (i.e., the export directory of Desk-LM must be the root directory of the Micro-LM project)

The program must be configured in AI_main.h, where the user has to specify some #define:

• The algorithm: SVM or DT or KNN
• DS_TEST, if you want to test performance in a dataset, instead of doing one shot estimations
• if DS_TEST is defined, then, if you define SAMPLE_TESTINGSET, a minimal_testing_set (.c and .h files) is used, which is produced by Desk-LM if its nTests variable is different from 'full'. Otherwise (if DS_TEST is defined), testing_set (.c and .h files) is used, which is produced by Desk-LM if its nTests variable is equal to 'full'.
• REGRESSION, if you want to perform a regression. Default is classification (no regression)

The key functions are:

• algo_test_dataset(isRegression) for (whole or minimal) dataset testing. algo is knn, decisionTree, svm
• preprocess(X), where X is the sample vector
• algo_classification(X), where X is the sample vector. algo is knn, decisionTree, svm