Package ml.shifu.guagua.example.lnr

Class LinearRegressionMaster

  • All Implemented Interfaces:
    ml.shifu.guagua.master.MasterComputable<LinearRegressionParams,​LinearRegressionParams>
    public class LinearRegressionMaster
extends ml.shifu.guagua.master.AbstractMasterComputable<LinearRegressionParams,​LinearRegressionParams>
    LinearRegressionMaster defines logic to update global linear regression model.

    At first iteration, master builds a random model then send to all workers to start computing. This is to make all workers use the same model at the starting time.

    At other iterations, master works:

    • 1. Accumulate all gradients from workers.
    • 2. Update global models by using accumulated gradients.
    • 3. Send new global model to workers by returning model parameters.