Class LogisticTrain

src/com/winvector/logistic/LogisticTrain.java:43–342

import com.winvector.variables.VariableEncodings;

public class LogisticTrain {
	
	public static PrimaVariableInfo buildVariableDefs(final Formula f, final Iterable<BurstMap> source) {
		// pass to get numeric columns and categorical columns
		final Log log = LogFactory.getLog(LogisticTrain.class);
		final PrimaVariableInfo def = new PrimaVariableInfo();
		def.readyForDefTracking(f);
		// not going to parallelize this as it it is cheaper than scans that evaluate model probabilities.
		log.info("start variable def scan 1/2");
		final Ticker ticker = new Ticker("build variable defs");
		for (BurstMap row : source) {
			ticker.tick();
			def.trackVariableDefsFromRow(row);
		}
		log.info("start variable def scan 2/2");
		ticker.start();
		// pass to get levels of categorical variables
		for(BurstMap row: source) {
			ticker.tick();
			def.trackVariableLevelsFromRow(row);
		}
		log.info("done variable def scans");
		def.trimStuckLevels();
		return def;
	}
	
	public static VariableEncodings buildAdpater(final Formula f, final String weightKey,
			final Iterable<BurstMap> source) {
		final PrimaVariableInfo def = buildVariableDefs(f,source);
		return new VariableEncodings(def,f.useIntercept,weightKey);
	}

	private static final String TRAINURIKEY = "trainURI";
	private static final String TRAINSEP = "sep";
	private static final String TRAINHDLKEY = "trainHDL";
	private static final String TRAINTBLKEY = "trainTBL";
	private static final String MEMKEY = "inmemory";
	private static final String FORMULAKEY = "formula";
	private static final String WEIGHTKEY = "weights";
	private static final String RESULTSERKEY = "resultSer";
	private static final String RESULTTSVKEY = "resultTSV";
	private static final String TRAINCLASSKEY = "trainClass";

	private static CommandLine parseCommandLine(final String[] args) throws org.apache.commons.cli.ParseException {
		final CommandLineParser clparser = new GnuParser();
		final Options cloptions = new Options();
		cloptions.addOption(TRAINURIKEY,true,"URI to get training TSV data from");
		cloptions.addOption(TRAINSEP,true,"(optional) training data input separator");
		cloptions.addOption(TRAINHDLKEY,true,"XML file to get JDBC connection to training data table");
		cloptions.addOption(TRAINTBLKEY,true,"table to use from database for training data");
		cloptions.addOption(MEMKEY, false, "(optional) if set data is held in memory during training");
		cloptions.addOption(FORMULAKEY,true,"formula to fit");
		cloptions.addOption(WEIGHTKEY,true,"(optional) symbol to user for weights");
		cloptions.addOption(RESULTSERKEY,true,"(optional) file to write seriazlized results to");
		cloptions.addOption(RESULTTSVKEY,true,"(optional) file to write TSV results to");
		cloptions.addOption(TRAINCLASSKEY,true,"(optional) alternate class to use for training");
		cloptions.getOption(FORMULAKEY).setRequired(true);
		final CommandLine cl = clparser.parse(cloptions, args);