This is an implementation of Coordinate Descent Algorithms for optimization in C++ done as part of my course project. Following algorithms are implemented.

• Coordinate Descent
• Parallel Coordinate Descent
• Accelerated, Parallel and Proximal Coordinate Descent

Compilation:

$make clean
$make all

Running:

Use following format

./CD -o1 option1 -o2 option2 -o3 option3 .... 

Options:

• -A optimization algorithms. (required) * pcd: parallel coordinate descent * apcd: accelerated parallel coordinate descent
• -L Loss function. (required)
  • sq: squared loss
  • log: logistic loss
• -R Regularizer.
  • l1: L1 regularizer
  • l2: L2 regularizer
• -D Dataset name. (required) Files are mentioned in config/files.cfg. To add new dataset, please add it in config/files.cfg with details.
  • blog : Blog Feedback Data Set
  • msd : Year Prediction MSD
  • arcene : Arcene Data Set
  • internetAd : Internet Advertisements Data Set
  • theoremProving : First-order theorem proving Data Set
• -T number of iterations. (required)
• -alpha learning rate (required)
• -period Iterations to reduce learning rate ( use -1 for constant learning rate)
• -P Number of threads. (required)
• -lambda Regularization Constant
• -GS 1 for using Gauss-Southwell rule else 0

External Libraries Used:

• Eigen : Used for matrix operations
• ConfigParser : Used for reading config files