wormhole/linear solves the following linear method

min_w sum_i loss(y_i, <w, x_i>) + penalty(w)

where (x_i, y_i) is a data pair and vector w is the model we will learn.

Use build.sh to build, and demo_local.sh to train on the sample dataset in local machine, or demo_yarn.sh to run the job on Yarn.

• A Step-by-step tutorial of training sparse logistic regression using the Criteo Terabytes CTR dataset on Amazon EC2. It processes 10 millions examples per second using only 5 machines.
• More about build
• Job configuration
• How to contribute
• Use a block coordinate descent solver
• Use a L-BFGS solver