/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.mahout.classifier.sgd;
import com.google.common.base.Charsets;
import com.google.common.collect.Lists;
import com.google.common.io.Closeables;
import com.google.common.io.Resources;
import org.apache.commons.cli2.CommandLine;
import org.apache.commons.cli2.Group;
import org.apache.commons.cli2.Option;
import org.apache.commons.cli2.builder.ArgumentBuilder;
import org.apache.commons.cli2.builder.DefaultOptionBuilder;
import org.apache.commons.cli2.builder.GroupBuilder;
import org.apache.commons.cli2.commandline.Parser;
import org.apache.commons.cli2.util.HelpFormatter;
import org.apache.mahout.math.RandomAccessSparseVector;
import org.apache.mahout.math.Vector;
import java.io.BufferedReader;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.io.OutputStream;
import java.io.PrintWriter;
import java.util.List;
import java.util.Locale;
/**
* Train a logistic regression for the examples from Chapter 13 of Mahout in Action
*/
public final class TrainLogistic {
private static String inputFile;
private static String outputFile;
private static LogisticModelParameters lmp;
private static int passes;
private static boolean scores;
private static OnlineLogisticRegression model;
private TrainLogistic() {
}
public static void main(String[] args) throws Exception {
mainToOutput(args, new PrintWriter(System.out, true));
}
static void mainToOutput(String[] args, PrintWriter output) throws Exception {
if (parseArgs(args)) {
double logPEstimate = 0;
int samples = 0;
CsvRecordFactory csv = lmp.getCsvRecordFactory();
OnlineLogisticRegression lr = lmp.createRegression();
for (int pass = 0; pass < passes; pass++) {
BufferedReader in = open(inputFile);
try {
// read variable names
csv.firstLine(in.readLine());
String line = in.readLine();
int lineCount = 0;
while (line != null) {
// for each new line, get target and predictors
Vector input = new RandomAccessSparseVector(lmp.getNumFeatures());
int targetValue = csv.processLine(line, input);
// check performance while this is still news
double logP = lr.logLikelihood(targetValue, input);
if (!Double.isInfinite(logP)) {
if (samples < 20) {
logPEstimate = (samples * logPEstimate + logP) / (samples + 1);
} else {
logPEstimate = 0.95 * logPEstimate + 0.05 * logP;
}
samples++;
}
double p = lr.classifyScalar(input);
if (scores) {
output.printf(Locale.ENGLISH, "%10d %2d %10.2f %2.4f %10.4f %10.4f\n",
samples, targetValue, lr.currentLearningRate(), p, logP, logPEstimate);
}
// now update model
lr.train(targetValue, input);
line = in.readLine();
lineCount++;
}
} finally {
Closeables.closeQuietly(in);
}
}
OutputStream modelOutput = new FileOutputStream(outputFile);
try {
lmp.saveTo(modelOutput);
} finally {
Closeables.closeQuietly(modelOutput);
}
output.printf(Locale.ENGLISH, "%d\n", lmp.getNumFeatures());
output.printf(Locale.ENGLISH, "%s ~ ", lmp.getTargetVariable());
String sep = "";
for (String v : csv.getTraceDictionary().keySet()) {
double weight = predictorWeight(lr, 0, csv, v);
if (weight != 0) {
output.printf(Locale.ENGLISH, "%s%.3f*%s", sep, weight, v);
sep = " + ";
}
}
output.printf("\n");
model = lr;
for (int row = 0; row < lr.getBeta().numRows(); row++) {
for (String key : csv.getTraceDictionary().keySet()) {
double weight = predictorWeight(lr, row, csv, key);
if (weight != 0) {
output.printf(Locale.ENGLISH, "%20s %.5f\n", key, weight);
}
}
for (int column = 0; column < lr.getBeta().numCols(); column++) {
output.printf(Locale.ENGLISH, "%15.9f ", lr.getBeta().get(row, column));
}
output.println();
}
}
}
private static double predictorWeight(OnlineLogisticRegression lr, int row, RecordFactory csv, String predictor) {
double weight = 0;
for (Integer column : csv.getTraceDictionary().get(predictor)) {
weight += lr.getBeta().get(row, column);
}
return weight;
}
private static boolean parseArgs(String[] args) {
DefaultOptionBuilder builder = new DefaultOptionBuilder();
Option help = builder.withLongName("help").withDescription("print this list").create();
Option quiet = builder.withLongName("quiet").withDescription("be extra quiet").create();
Option scores = builder.withLongName("scores").withDescription("output score diagnostics during training").create();
ArgumentBuilder argumentBuilder = new ArgumentBuilder();
Option inputFile = builder.withLongName("input")
.withRequired(true)
.withArgument(argumentBuilder.withName("input").withMaximum(1).create())
.withDescription("where to get training data")
.create();
Option outputFile = builder.withLongName("output")
.withRequired(true)
.withArgument(argumentBuilder.withName("output").withMaximum(1).create())
.withDescription("where to get training data")
.create();
Option predictors = builder.withLongName("predictors")
.withRequired(true)
.withArgument(argumentBuilder.withName("p").create())
.withDescription("a list of predictor variables")
.create();
Option types = builder.withLongName("types")
.withRequired(true)
.withArgument(argumentBuilder.withName("t").create())
.withDescription("a list of predictor variable types (numeric, word, or text)")
.create();
Option target = builder.withLongName("target")
.withRequired(true)
.withArgument(argumentBuilder.withName("target").withMaximum(1).create())
.withDescription("the name of the target variable")
.create();
Option features = builder.withLongName("features")
.withArgument(
argumentBuilder.withName("numFeatures")
.withDefault("1000")
.withMaximum(1).create())
.withDescription("the number of internal hashed features to use")
.create();
Option passes = builder.withLongName("passes")
.withArgument(
argumentBuilder.withName("passes")
.withDefault("2")
.withMaximum(1).create())
.withDescription("the number of times to pass over the input data")
.create();
Option lambda = builder.withLongName("lambda")
.withArgument(argumentBuilder.withName("lambda").withDefault("1e-4").withMaximum(1).create())
.withDescription("the amount of coefficient decay to use")
.create();
Option rate = builder.withLongName("rate")
.withArgument(argumentBuilder.withName("learningRate").withDefault("1e-3").withMaximum(1).create())
.withDescription("the learning rate")
.create();
Option noBias = builder.withLongName("noBias")
.withDescription("don't include a bias term")
.create();
Option targetCategories = builder.withLongName("categories")
.withRequired(true)
.withArgument(argumentBuilder.withName("number").withMaximum(1).create())
.withDescription("the number of target categories to be considered")
.create();
Group normalArgs = new GroupBuilder()
.withOption(help)
.withOption(quiet)
.withOption(inputFile)
.withOption(outputFile)
.withOption(target)
.withOption(targetCategories)
.withOption(predictors)
.withOption(types)
.withOption(passes)
.withOption(lambda)
.withOption(rate)
.withOption(noBias)
.withOption(features)
.create();
Parser parser = new Parser();
parser.setHelpOption(help);
parser.setHelpTrigger("--help");
parser.setGroup(normalArgs);
parser.setHelpFormatter(new HelpFormatter(" ", "", " ", 130));
CommandLine cmdLine = parser.parseAndHelp(args);
if (cmdLine == null) {
return false;
}
TrainLogistic.inputFile = getStringArgument(cmdLine, inputFile);
TrainLogistic.outputFile = getStringArgument(cmdLine, outputFile);
List<String> typeList = Lists.newArrayList();
for (Object x : cmdLine.getValues(types)) {
typeList.add(x.toString());
}
List<String> predictorList = Lists.newArrayList();
for (Object x : cmdLine.getValues(predictors)) {
predictorList.add(x.toString());
}
lmp = new LogisticModelParameters();
lmp.setTargetVariable(getStringArgument(cmdLine, target));
lmp.setMaxTargetCategories(getIntegerArgument(cmdLine, targetCategories));
lmp.setNumFeatures(getIntegerArgument(cmdLine, features));
lmp.setUseBias(!getBooleanArgument(cmdLine, noBias));
lmp.setTypeMap(predictorList, typeList);
lmp.setLambda(getDoubleArgument(cmdLine, lambda));
lmp.setLearningRate(getDoubleArgument(cmdLine, rate));
TrainLogistic.scores = getBooleanArgument(cmdLine, scores);
TrainLogistic.passes = getIntegerArgument(cmdLine, passes);
return true;
}
private static String getStringArgument(CommandLine cmdLine, Option inputFile) {
return (String) cmdLine.getValue(inputFile);
}
private static boolean getBooleanArgument(CommandLine cmdLine, Option option) {
return cmdLine.hasOption(option);
}
private static int getIntegerArgument(CommandLine cmdLine, Option features) {
return Integer.parseInt((String) cmdLine.getValue(features));
}
private static double getDoubleArgument(CommandLine cmdLine, Option op) {
return Double.parseDouble((String) cmdLine.getValue(op));
}
public static OnlineLogisticRegression getModel() {
return model;
}
public static LogisticModelParameters getParameters() {
return lmp;
}
static BufferedReader open(String inputFile) throws IOException {
InputStream in;
try {
in = Resources.getResource(inputFile).openStream();
} catch (IllegalArgumentException e) {
in = new FileInputStream(new File(inputFile));
}
return new BufferedReader(new InputStreamReader(in, Charsets.UTF_8));
}
}