This Matlab package implements machine learning algorithms described in the great textbook: Pattern Recognition and Machine Learning by C. Bishop (PRML).
It is written purely in Matlab language. It is self-contained. There is no external dependency.
Note: this package requires Matlab R2016b or latter, since it utilizes a new Matlab syntax called Implicit expansion (a.k.a. broadcasting). It also requires Statistics Toolbox (for some simple random number generator) and Image Processing Toolbox (for reading image data).
- Succinct: The code is extremely compact. Minimizing code length is a major goal. As a result, the core of the algorithms can be easily spotted.
- Efficient: Many tricks for speeding up Matlab code are applied (e.g. vectorization, matrix factorization, etc.). Usually, functions in this package are orders faster than Matlab builtin ones (e.g. kmeans).
- Robust: Many tricks for numerical stability are applied, such as computing probability in logrithm domain, square root matrix update to enforce matrix symmetry\PD, etc.
- Readable: The code is heavily commented. Corresponding formulas in PRML are annoted. Symbols are in sync with the book.
- Practical: The package is not only readable, but also meant to be easily used and modified to facilitate ML research. Many functions in this package are already widely used (see Matlab file exchange).
- Download the package to a local folder (e.g. ~/PRMLT/) by running:
git clone https://github.com/PRML/PRMLT.git-
Run Matlab and navigate to the folder (~/PRMLT/), then run the init.m script.
-
Run some demos in ~/PRMLT/demo folder. Enjoy!
If you find any bug or have any suggestion, please do file issues. I am graceful for any feedback and will do my best to improve this package.
Released under MIT license
sth4nth at gmail dot com