lina

Linear Algebra library in C++ and OpenCL for machine learning algorithms. GPU-accelerated routines for multidimensional optimization, linear regression, logistic regression. Later: neural networks. Even later: SVM and recommender systems.

This is pretty much dead since Google published the TensorFlow machine learning library which does anything this project can ever hope to do.

Dependencies:

Boost uBLAS

sudo apt-get install libboost-dev

ViennaCl

sudo apt-get install libviennacl-dev

OpenCl headers and drivers

YMMV, packages to install depend on present GPU:

sudo apt-get install ocl-icd-libopencl1 ocl-icd-opencl-dev opencl-headers

Google Test

sudo apt-get install libgtest-dev
cd /usr/src/gtest
sudo cmake CMakeLists.txt
sudo make
sudo cp *.a /usr/lib

To do

• improve RegressionSolver
  • make the kind of regression a class template parameter
  • add predict() function
  • add function to determine training accuracy
• use smart pointers again (in gradient descent)
• logistic regression
  • minimization - is there a better way than gradient descent? does GD always give so bad results in nontrivial systems?
  • training accuracy
  • regularization
    • coursera example
  • multi-class classification
• ensure that only one matrix is stored in GPU memory at each time when using LinearRegressionSolver
• gradient descent seems to run on one CPU core only, at least with logistic regression
• factor out matrix and vector types so they can be used as template parameters
  • easier conversion between ublas and viennacl data types and algorithms?
• neural networks
• compile conditionally on presence of gtest so that it can be distributed without it