Machine Learning with Spark - Second Edition

Iterations

Many machine learning methods are iterative in nature, converging to a solution (the optimal weight vector that minimizes the chosen loss function) over a number of iteration steps. SGD typically requires relatively few iterations to converge to a reasonable solution, but can be run for more iterations to improve the solution. We can see this by trying a few different settings for the numIterations parameter, and comparing the AUC results like this:

val iterResults = Seq(1, 5, 10, 50).map { param =>   val model = trainWithParams(scaledDataCats, 0.0, param, new  SimpleUpdater, 1.0)   createMetrics(s"$param iterations", scaledDataCats, model) } iterResults.foreach { case (param, auc) => println(f"$param, AUC =  ${auc * 100}%2.2f%%") ...