A primal-dual framework for distributed L1-regularized optimization.
This code trains a standard least squares sparse regression with L1 or elastic net regularizer. The proxCoCoA+ framework runs on the primal optimization problem (called D in the paper). To solve the data-local subproblems on each machine, an arbitrary solver can be used. In this example we use randomized coordinate descent as the local solver, as the L1-regularized single coordinate problems have simple closed-form solutions.
The code can be easily adapted to include other internal solvers or to solve other data-fit objectives or regularizers.
How to run the code locally:
sbt/sbt assembly ./run-demo-local.sh
sbt script to run, make sure you have downloaded CoCoA into a directory whose path contains no spaces.)
The algorithmic framework is described in more detail in the following paper:
Smith, V., Forte, S., Jordan, M.I., Jaggi, M. L1-Regularized Distributed Optimization: A Communication-Efficient Primal-Dual Framework