libsvm is the famous SVM library, which provides interfaces for almost all programming languages. However, it seems not friendly for embeded system. In my application, I trained data on my Linux machine for a model file, and use it on STM32 board, which do not have a file system. So the easiest way maybe is converting the generated model file to c files like svm_model.h and svm_model.c which can be compiled to any embeded system.
convert_model_to_c.py reads an model file and replace variables to STM32_files/svm_model.c.in and generate STM32_files/svm_model.c as output.
./convert_model_to_c.py [model_path] [prefix]
[prefix]svm_model.h and [prefix]svm_model.c can then be utilized in STM32 project.
prefix is used to distinguish models as I used two SVM models in my application. It can also be leave alone, thus no prefix will be added.
STM32_files/main.c tests the two files and also gives the demonstration of invoking functions in svm_model.c
To make sure the svm_model.c is correct, tests could be conducted. Following are the steps:
Firstly, run gen_test_model.py to get test.model file.
./gen_test_model.py
Secondly, execute convert_model_to_c.py to generate newest svm_model.c file
./convert_model_to_c.py
Finally, test same data both by standard libsvm and converted svm_model.c
test standard libsvm:
cat data.txt | ./test_model.py
test converted svm_model.c:
cd STM32_files
make
cat ../data.txt | ./svm_test
the output will tell whether the converted program is right or not.