MultiClassClassificator is a simple application of Machine Learning. Project in his assumptions should recognize digits which are presented on PGM files. In fact after splitting training set on training and testing set (70% for training and 30% for testing), I have got about 76% accuracy. It is not even a good result. I think there are two reasons why result is so poor - firstly, method Logic Regression with gradient descent, secondly lambda parameter. I am sure that if you choose a good lambda parameter results will be much much better.
Documentation is generated by doxygen in "doc" folder.
cd ~/Documents
git clone https://github.com/badeleux/MultiClassClassificationASM.git
cd MultiClassClassificationASM/
To compile application you need:
- Linux
- Armadillo library 32BIT
- If you are running 64bit system, you probably also need GCC-multilib (or equivalent package for your distro)
make
If all went good - now in folder bin you should have dynamic library (*.so) and executable Predict file.
In terminal:
./Predict ../digits/1337.pgm
Program will prompt you in which way would you like to train your Theta Matrix, in C or in Assembler. Next program will try to guess what number is in picture.
No license - which means - "To use the code, you must contact the author directly and ask permission."