Java tutorial
/************************************************************************************************************************** Copyright 2012-14 Justin A. Debrabant <debrabant@cs.brown.edu> and Matteo Riondato <matteo@cs.brown.edu> Licensed 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. * Description: Driver for Hadoop implementation of parallel association rule mining. * Usage: java MRDriver <mapper id> <path to input database> <path to output local FIs> <path to output global FIs> * mapper id - specifies which Map method should be used 1 for partition mapper, 2 for binomial mapper, 3 for weighted coin flip sampler, 4 for sampler, 5 for random integer partition, * path to input database - path to file containing transactions in .dat format (1 transaction per line) * local FI output - path to directory to write local (per-reducer) FIs * global FI output - path to directory to write global FIs (combined from all local FIs) ***************************************************************************************************************************/ import java.net.URI; import java.util.Arrays; import java.util.ArrayList; import java.util.Collections; import java.util.Hashtable; import java.util.Random; import org.apache.hadoop.conf.Configured; import org.apache.hadoop.filecache.DistributedCache; import org.apache.hadoop.fs.FileStatus; import org.apache.hadoop.fs.FileSystem; import org.apache.hadoop.fs.FSDataOutputStream; import org.apache.hadoop.fs.Path; import org.apache.hadoop.io.DefaultStringifier; import org.apache.hadoop.io.DoubleWritable; import org.apache.hadoop.io.IntWritable; import org.apache.hadoop.io.LongWritable; import org.apache.hadoop.io.MapWritable; import org.apache.hadoop.io.Text; import org.apache.hadoop.mapred.Counters; import org.apache.hadoop.mapred.FileOutputFormat; import org.apache.hadoop.mapred.JobClient; import org.apache.hadoop.mapred.JobConf; import org.apache.hadoop.mapred.RunningJob; import org.apache.hadoop.mapred.SequenceFileInputFormat; import org.apache.hadoop.mapred.SequenceFileOutputFormat; import org.apache.hadoop.util.Tool; import org.apache.hadoop.util.ToolRunner; public class MRDriver extends Configured implements Tool { public final int MR_TIMEOUT_MILLI = 60000000; public static void main(String args[]) throws Exception { if (args.length != 11) { System.out.println( "usage: java MRDriver <epsilon> <delta> <minFreqPercent> <d> <datasetSize> <numSamples> <phi> <mapper id> <path to input database> " + "<path to output local FIs> <path to output global FIs>"); System.exit(1); } int res = ToolRunner.run(new MRDriver(), args); System.exit(res); } public int run(String args[]) throws Exception { FileSystem fs = null; Path samplesMapPath = null; float epsilon = Float.parseFloat(args[0]); double delta = Double.parseDouble(args[1]); int minFreqPercent = Integer.parseInt(args[2]); int d = Integer.parseInt(args[3]); int datasetSize = Integer.parseInt(args[4]); int numSamples = Integer.parseInt(args[5]); double phi = Double.parseDouble(args[6]); Random rand; /************************ Job 1 (local FIM) Configuration ************************/ JobConf conf = new JobConf(getConf()); /* * Compute the number of required "votes" for an itemsets to be * declared frequent */ // The +1 at the end is needed to ensure reqApproxNum > numsamples / 2. int reqApproxNum = (int) Math .floor((numSamples * (1 - phi)) - Math.sqrt(numSamples * (1 - phi) * 2 * Math.log(1 / delta))) + 1; int sampleSize = (int) Math.ceil((2 / Math.pow(epsilon, 2)) * (d + Math.log(1 / phi))); //System.out.println("reducersNum: " + numSamples + " reqApproxNum: " + reqApproxNum); conf.setInt("PARMM.reducersNum", numSamples); conf.setInt("PARMM.datasetSize", datasetSize); conf.setInt("PARMM.minFreqPercent", minFreqPercent); conf.setInt("PARMM.sampleSize", sampleSize); conf.setFloat("PARMM.epsilon", epsilon); // Set the number of reducers equal to the number of samples, to // maximize parallelism. Required by our Partitioner. conf.setNumReduceTasks(numSamples); // XXX: why do we disable the speculative execution? MR conf.setBoolean("mapred.reduce.tasks.speculative.execution", false); conf.setInt("mapred.task.timeout", MR_TIMEOUT_MILLI); /* * Enable compression of map output. * * We do it for this job and not for the aggregation one because * each mapper there only print out one record for each itemset, * so there isn't much to compress, I'd say. MR * * In Amazon MapReduce compression of the map output seems to be * happen by default and the Snappy codec is used, which is * extremely fast. */ conf.setBoolean("mapred.compress.map.output", true); //conf.setMapOutputCompressorClass(com.hadoop.compression.lzo.LzoCodec.class); conf.setJarByClass(MRDriver.class); conf.setMapOutputKeyClass(IntWritable.class); conf.setMapOutputValueClass(Text.class); conf.setOutputKeyClass(Text.class); conf.setOutputValueClass(DoubleWritable.class); conf.setInputFormat(SequenceFileInputFormat.class); // We write the collections found in a reducers as a SequenceFile conf.setOutputFormat(SequenceFileOutputFormat.class); SequenceFileOutputFormat.setOutputPath(conf, new Path(args[9])); // set the mapper class based on command line option switch (Integer.parseInt(args[7])) { case 1: System.out.println("running partition mapper..."); SequenceFileInputFormat.addInputPath(conf, new Path(args[8])); conf.setMapperClass(PartitionMapper.class); break; case 2: System.out.println("running binomial mapper..."); SequenceFileInputFormat.addInputPath(conf, new Path(args[8])); conf.setMapperClass(BinomialSamplerMapper.class); break; case 3: System.out.println("running coin mapper..."); SequenceFileInputFormat.addInputPath(conf, new Path(args[8])); conf.setMapperClass(CoinFlipSamplerMapper.class); case 4: System.out.println("running sampler mapper..."); SequenceFileInputFormat.addInputPath(conf, new Path(args[8])); conf.setMapperClass(InputSamplerMapper.class); // create a random sample of size T*m rand = new Random(); long sampling_start_time = System.nanoTime(); int[] samples = new int[numSamples * sampleSize]; for (int i = 0; i < numSamples * sampleSize; i++) { samples[i] = rand.nextInt(datasetSize); } // for each key in the sample, create a list of all T samples to which this key belongs Hashtable<LongWritable, ArrayList<IntWritable>> hashTable = new Hashtable<LongWritable, ArrayList<IntWritable>>(); for (int i = 0; i < numSamples * sampleSize; i++) { ArrayList<IntWritable> sampleIDs = null; LongWritable key = new LongWritable(samples[i]); if (hashTable.containsKey(key)) sampleIDs = hashTable.get(key); else sampleIDs = new ArrayList<IntWritable>(); sampleIDs.add(new IntWritable(i % numSamples)); hashTable.put(key, sampleIDs); } /* * Convert the Hastable to a MapWritable which we will * write to HDFS and distribute to all Mappers using * DistributedCache */ MapWritable map = new MapWritable(); for (LongWritable key : hashTable.keySet()) { ArrayList<IntWritable> sampleIDs = hashTable.get(key); IntArrayWritable sampleIDsIAW = new IntArrayWritable(); sampleIDsIAW.set(sampleIDs.toArray(new IntWritable[sampleIDs.size()])); map.put(key, sampleIDsIAW); } fs = FileSystem.get(URI.create("samplesMap.ser"), conf); samplesMapPath = new Path("samplesMap.ser"); FSDataOutputStream out = fs.create(samplesMapPath, true); map.write(out); out.sync(); out.close(); DistributedCache.addCacheFile(new URI(fs.getWorkingDirectory() + "/samplesMap.ser#samplesMap.ser"), conf); // stop the sampling timer long sampling_end_time = System.nanoTime(); long sampling_runtime = (sampling_end_time - sampling_start_time) / 1000000; System.out.println("sampling runtime (milliseconds): " + sampling_runtime); break; // end switch case case 5: System.out.println("running random integer partition mapper..."); conf.setInputFormat(WholeSplitInputFormat.class); Path inputFilePath = new Path(args[8]); WholeSplitInputFormat.addInputPath(conf, inputFilePath); conf.setMapperClass(RandIntPartSamplerMapper.class); // Compute number of map tasks. fs = inputFilePath.getFileSystem(conf); FileStatus inputFileStatus = fs.getFileStatus(inputFilePath); long len = inputFileStatus.getLen(); long blockSize = inputFileStatus.getBlockSize(); conf.setLong("mapred.min.split.size", blockSize); conf.setLong("mapred.max.split.size", blockSize); int mapTasksNum = ((int) (len / blockSize)) + 1; conf.setNumMapTasks(mapTasksNum); //System.out.println("len: " + len + " blockSize: " // + blockSize + " mapTasksNum: " + mapTasksNum); // Extract random integer partition of total sample // size into up to mapTasksNum partitions. // XXX I'm not sure this is a correct way to do // it. rand = new Random(); IntWritable[][] toSampleArr = new IntWritable[mapTasksNum][numSamples]; for (int j = 0; j < numSamples; j++) { IntWritable[] tempToSampleArr = new IntWritable[mapTasksNum]; int sum = 0; int i; for (i = 0; i < mapTasksNum - 1; i++) { int size = rand.nextInt(sampleSize - sum); tempToSampleArr[i] = new IntWritable(size); sum += size; if (sum > numSamples * sampleSize) { System.out.println("Something went wrong generating the sample Sizes"); System.exit(1); } if (sum == sampleSize) { break; } } if (i == mapTasksNum - 1) { tempToSampleArr[i] = new IntWritable(sampleSize - sum); } else { for (; i < mapTasksNum; i++) { tempToSampleArr[i] = new IntWritable(0); } } Collections.shuffle(Arrays.asList(tempToSampleArr)); for (i = 0; i < mapTasksNum; i++) { toSampleArr[i][j] = tempToSampleArr[i]; } } for (int i = 0; i < mapTasksNum; i++) { DefaultStringifier.storeArray(conf, toSampleArr[i], "PARMM.toSampleArr_" + i); } break; default: System.err.println("Wrong Mapper ID. Can only be in [1,5]"); System.exit(1); break; } /* * We don't use the default hash partitioner because we want to * maximize the parallelism. That's why we also fix the number * of reducers. */ conf.setPartitionerClass(FIMPartitioner.class); conf.setReducerClass(FIMReducer.class); /************************ Job 2 (aggregation) Configuration ************************/ JobConf confAggr = new JobConf(getConf()); confAggr.setInt("PARMM.reducersNum", numSamples); confAggr.setInt("PARMM.reqApproxNum", reqApproxNum); confAggr.setInt("PARMM.sampleSize", sampleSize); confAggr.setFloat("PARMM.epsilon", epsilon); // XXX: Why do we disable speculative execution? MR confAggr.setBoolean("mapred.reduce.tasks.speculative.execution", false); confAggr.setInt("mapred.task.timeout", MR_TIMEOUT_MILLI); confAggr.setJarByClass(MRDriver.class); confAggr.setMapOutputKeyClass(Text.class); confAggr.setMapOutputValueClass(DoubleWritable.class); confAggr.setOutputKeyClass(Text.class); confAggr.setOutputValueClass(Text.class); confAggr.setMapperClass(AggregateMapper.class); confAggr.setReducerClass(AggregateReducer.class); confAggr.setInputFormat(CombineSequenceFileInputFormat.class); SequenceFileInputFormat.addInputPath(confAggr, new Path(args[9])); FileOutputFormat.setOutputPath(confAggr, new Path(args[10])); long FIMjob_start_time = System.currentTimeMillis(); RunningJob FIMjob = JobClient.runJob(conf); long FIMjob_end_time = System.currentTimeMillis(); RunningJob aggregateJob = JobClient.runJob(confAggr); long aggrJob_end_time = System.currentTimeMillis(); long FIMjob_runtime = FIMjob_end_time - FIMjob_start_time; long aggrJob_runtime = aggrJob_end_time - FIMjob_end_time; if (args[7].equals("4")) { // Remove samplesMap file fs.delete(samplesMapPath, false); } Counters counters = FIMjob.getCounters(); Counters.Group FIMMapperStartTimesCounters = counters.getGroup("FIMMapperStart"); long[] FIMMapperStartTimes = new long[FIMMapperStartTimesCounters.size()]; int i = 0; for (Counters.Counter counter : FIMMapperStartTimesCounters) { FIMMapperStartTimes[i++] = counter.getCounter(); } Counters.Group FIMMapperEndTimesCounters = counters.getGroup("FIMMapperEnd"); long[] FIMMapperEndTimes = new long[FIMMapperEndTimesCounters.size()]; i = 0; for (Counters.Counter counter : FIMMapperEndTimesCounters) { FIMMapperEndTimes[i++] = counter.getCounter(); } Counters.Group FIMReducerStartTimesCounters = counters.getGroup("FIMReducerStart"); long[] FIMReducerStartTimes = new long[FIMReducerStartTimesCounters.size()]; i = 0; for (Counters.Counter counter : FIMReducerStartTimesCounters) { FIMReducerStartTimes[i++] = counter.getCounter(); } Counters.Group FIMReducerEndTimesCounters = counters.getGroup("FIMReducerEnd"); long[] FIMReducerEndTimes = new long[FIMReducerEndTimesCounters.size()]; i = 0; for (Counters.Counter counter : FIMReducerEndTimesCounters) { FIMReducerEndTimes[i++] = counter.getCounter(); } Counters countersAggr = aggregateJob.getCounters(); Counters.Group AggregateMapperStartTimesCounters = countersAggr.getGroup("AggregateMapperStart"); long[] AggregateMapperStartTimes = new long[AggregateMapperStartTimesCounters.size()]; i = 0; for (Counters.Counter counter : AggregateMapperStartTimesCounters) { AggregateMapperStartTimes[i++] = counter.getCounter(); } Counters.Group AggregateMapperEndTimesCounters = countersAggr.getGroup("AggregateMapperEnd"); long[] AggregateMapperEndTimes = new long[AggregateMapperEndTimesCounters.size()]; i = 0; for (Counters.Counter counter : AggregateMapperEndTimesCounters) { AggregateMapperEndTimes[i++] = counter.getCounter(); } Counters.Group AggregateReducerStartTimesCounters = countersAggr.getGroup("AggregateReducerStart"); long[] AggregateReducerStartTimes = new long[AggregateReducerStartTimesCounters.size()]; i = 0; for (Counters.Counter counter : AggregateReducerStartTimesCounters) { AggregateReducerStartTimes[i++] = counter.getCounter(); } Counters.Group AggregateReducerEndTimesCounters = countersAggr.getGroup("AggregateReducerEnd"); long[] AggregateReducerEndTimes = new long[AggregateReducerEndTimesCounters.size()]; i = 0; for (Counters.Counter counter : AggregateReducerEndTimesCounters) { AggregateReducerEndTimes[i++] = counter.getCounter(); } long FIMMapperStartMin = FIMMapperStartTimes[0]; for (long l : FIMMapperStartTimes) { if (l < FIMMapperStartMin) { FIMMapperStartMin = l; } } long FIMMapperEndMax = FIMMapperEndTimes[0]; for (long l : FIMMapperEndTimes) { if (l > FIMMapperEndMax) { FIMMapperEndMax = l; } } System.out.println("FIM job setup time (milliseconds): " + (FIMMapperStartMin - FIMjob_start_time)); System.out.println("FIMMapper total runtime (milliseconds): " + (FIMMapperEndMax - FIMMapperStartMin)); long[] FIMMapperRunTimes = new long[FIMMapperStartTimes.length]; long FIMMapperRunTimesSum = 0; for (int l = 0; l < FIMMapperStartTimes.length; l++) { FIMMapperRunTimes[l] = FIMMapperEndTimes[l] - FIMMapperStartTimes[l]; FIMMapperRunTimesSum += FIMMapperRunTimes[l]; } System.out.println("FIMMapper average task runtime (milliseconds): " + FIMMapperRunTimesSum / FIMMapperStartTimes.length); long FIMMapperRunTimesMin = FIMMapperRunTimes[0]; long FIMMapperRunTimesMax = FIMMapperRunTimes[0]; for (long l : FIMMapperRunTimes) { if (l < FIMMapperRunTimesMin) { FIMMapperRunTimesMin = l; } if (l > FIMMapperRunTimesMax) { FIMMapperRunTimesMax = l; } } System.out.println("FIMMapper minimum task runtime (milliseconds): " + FIMMapperRunTimesMin); System.out.println("FIMMapper maximum task runtime (milliseconds): " + FIMMapperRunTimesMax); long FIMReducerStartMin = FIMReducerStartTimes[0]; for (long l : FIMReducerStartTimes) { if (l < FIMReducerStartMin) { FIMReducerStartMin = l; } } long FIMReducerEndMax = FIMReducerEndTimes[0]; for (long l : FIMReducerEndTimes) { if (l > FIMReducerEndMax) { FIMReducerEndMax = l; } } System.out .println("FIM job shuffle phase runtime (milliseconds): " + (FIMReducerStartMin - FIMMapperEndMax)); System.out.println("FIMReducer total runtime (milliseconds): " + (FIMReducerEndMax - FIMReducerStartMin)); long[] FIMReducerRunTimes = new long[FIMReducerStartTimes.length]; long FIMReducerRunTimesSum = 0; for (int l = 0; l < FIMReducerStartTimes.length; l++) { FIMReducerRunTimes[l] = FIMReducerEndTimes[l] - FIMReducerStartTimes[l]; FIMReducerRunTimesSum += FIMReducerRunTimes[l]; } System.out.println("FIMReducer average task runtime (milliseconds): " + FIMReducerRunTimesSum / FIMReducerStartTimes.length); long FIMReducerRunTimesMin = FIMReducerRunTimes[0]; long FIMReducerRunTimesMax = FIMReducerRunTimes[0]; for (long l : FIMReducerRunTimes) { if (l < FIMReducerRunTimesMin) { FIMReducerRunTimesMin = l; } if (l > FIMReducerRunTimesMax) { FIMReducerRunTimesMax = l; } } System.out.println("FIMReducer minimum task runtime (milliseconds): " + FIMReducerRunTimesMin); System.out.println("FIMReducer maximum task runtime (milliseconds): " + FIMReducerRunTimesMax); System.out.println("FIM job cooldown time (milliseconds): " + (FIMjob_end_time - FIMReducerEndMax)); long AggregateMapperStartMin = AggregateMapperStartTimes[0]; for (long l : AggregateMapperStartTimes) { if (l < AggregateMapperStartMin) { AggregateMapperStartMin = l; } } long AggregateMapperEndMax = AggregateMapperEndTimes[0]; for (long l : AggregateMapperEndTimes) { if (l > AggregateMapperEndMax) { AggregateMapperEndMax = l; } } System.out.println( "Aggregation job setup time (milliseconds): " + (AggregateMapperStartMin - FIMjob_end_time)); System.out.println("AggregateMapper total runtime (milliseconds): " + (AggregateMapperEndMax - AggregateMapperStartMin)); long[] AggregateMapperRunTimes = new long[AggregateMapperStartTimes.length]; long AggregateMapperRunTimesSum = 0; for (int l = 0; l < AggregateMapperStartTimes.length; l++) { AggregateMapperRunTimes[l] = AggregateMapperEndTimes[l] - AggregateMapperStartTimes[l]; AggregateMapperRunTimesSum += AggregateMapperRunTimes[l]; } System.out.println("AggregateMapper average task runtime (milliseconds): " + AggregateMapperRunTimesSum / AggregateMapperStartTimes.length); long AggregateMapperRunTimesMin = AggregateMapperRunTimes[0]; long AggregateMapperRunTimesMax = AggregateMapperRunTimes[0]; for (long l : AggregateMapperRunTimes) { if (l < AggregateMapperRunTimesMin) { AggregateMapperRunTimesMin = l; } if (l > AggregateMapperRunTimesMax) { AggregateMapperRunTimesMax = l; } } System.out.println("AggregateMapper minimum task runtime (milliseconds): " + AggregateMapperRunTimesMin); System.out.println("AggregateMapper maximum task runtime (milliseconds): " + AggregateMapperRunTimesMax); long AggregateReducerStartMin = AggregateReducerStartTimes[0]; for (long l : AggregateReducerStartTimes) { if (l < AggregateReducerStartMin) { AggregateReducerStartMin = l; } } long AggregateReducerEndMax = AggregateReducerEndTimes[0]; for (long l : AggregateReducerEndTimes) { if (l > AggregateReducerEndMax) { AggregateReducerEndMax = l; } } System.out.println("Aggregate job round shuffle phase runtime (milliseconds): " + (AggregateReducerStartMin - AggregateMapperEndMax)); System.out.println("AggregateReducer total runtime (milliseconds): " + (AggregateReducerEndMax - AggregateReducerStartMin)); long[] AggregateReducerRunTimes = new long[AggregateReducerStartTimes.length]; long AggregateReducerRunTimesSum = 0; for (int l = 0; l < AggregateReducerStartTimes.length; l++) { AggregateReducerRunTimes[l] = AggregateReducerEndTimes[l] - AggregateReducerStartTimes[l]; AggregateReducerRunTimesSum += AggregateReducerRunTimes[l]; } System.out.println("AggregateReducer average task runtime (milliseconds): " + AggregateReducerRunTimesSum / AggregateReducerStartTimes.length); long AggregateReducerRunTimesMin = AggregateReducerRunTimes[0]; long AggregateReducerRunTimesMax = AggregateReducerRunTimes[0]; for (long l : AggregateReducerRunTimes) { if (l < AggregateReducerRunTimesMin) { AggregateReducerRunTimesMin = l; } if (l > AggregateReducerRunTimesMax) { AggregateReducerRunTimesMax = l; } } System.out.println("AggregateReducer minimum task runtime (milliseconds): " + AggregateReducerRunTimesMin); System.out.println("AggregateReducer maximum task runtime (milliseconds): " + AggregateReducerRunTimesMax); System.out.println( "Aggregation job cooldown time (milliseconds): " + (aggrJob_end_time - AggregateReducerEndMax)); System.out .println("total runtime (all inclusive) (milliseconds): " + (aggrJob_end_time - FIMjob_start_time)); System.out.println("total runtime (no FIM job setup, no aggregation job cooldown) (milliseconds): " + (AggregateReducerEndMax - FIMMapperStartMin)); System.out.println("total runtime (no setups, no cooldowns) (milliseconds): " + (FIMReducerEndMax - FIMMapperStartMin + AggregateReducerEndMax - AggregateMapperStartMin)); System.out.println("FIM job runtime (including setup and cooldown) (milliseconds): " + FIMjob_runtime); System.out.println("FIM job runtime (no setup, no cooldown) (milliseconds): " + (FIMReducerEndMax - FIMMapperStartMin)); System.out.println( "Aggregation job runtime (including setup and cooldown) (milliseconds): " + aggrJob_runtime); System.out.println("Aggregation job runtime (no setup, no cooldown) (milliseconds): " + (AggregateReducerEndMax - AggregateMapperStartMin)); return 0; } }