public void runPra(
String kbDirectory,
String graphDirectory,
String splitsDirectory,
String parameterFile,
String topK,
String pathlength,
String isOA,
String outputBase)
throws IOException, InterruptedException, ClassNotFoundException, Exception {
outputBase = fileUtil.addDirectorySeparatorIfNecessary(outputBase);
kbDirectory = fileUtil.addDirectorySeparatorIfNecessary(kbDirectory);
graphDirectory = fileUtil.addDirectorySeparatorIfNecessary(graphDirectory);
splitsDirectory = fileUtil.addDirectorySeparatorIfNecessary(splitsDirectory);
fileUtil.mkdirOrDie(outputBase);
boolean isOnlineAug = false;
if (isOA.equalsIgnoreCase("yes")) isOnlineAug = true;
KB kb = null;
PraConfig baseConfig = null;
PraConfig.Builder baseBuilder = null;
if (isOnlineAug) {
logger.info("Initializing SVO Graph");
long initStart = System.currentTimeMillis();
kb =
OnlineAugment.init(
kbDirectory,
graphDirectory,
splitsDirectory,
outputBase,
Integer.parseInt(topK),
Integer.parseInt(pathlength));
long initEnd = System.currentTimeMillis();
logger.info("Initialization took " + (initEnd - initStart) / 1000.00 + " seconds");
// ImportDriver svo = new ImportDriver();
// svo.readSVOGraph();
// kb.setSVONodeDict(svo.getSVONodeDict());
// kb.setSVOAdjList(svo.getSVOAdjList());
} else {
baseBuilder = new PraConfig.Builder();
parseGraphFiles(graphDirectory, baseBuilder);
baseBuilder.setFromParamFile(fileUtil.getBufferedReader(parameterFile));
// This call potentially uses the edge dictionary that's set in
// parseGraphFiles - this MUST be
// called after parseGraphFiles, or things will break with really
// weird
// errors. TODO(matt): I
// really should write a test for this...
Map<String, String> nodeNames = null;
if (fileUtil.fileExists(kbDirectory + "node_names.tsv")) {
nodeNames = fileUtil.readMapFromTsvFile(kbDirectory + "node_names.tsv", true);
}
Outputter outputter = new Outputter(baseBuilder.nodeDict, baseBuilder.edgeDict, nodeNames);
baseBuilder.setOutputter(outputter);
baseConfig = baseBuilder.build();
Utils.deleteShards(graphDirectory + "graph_chi");
GraphCreator gc = new GraphCreator(outputBase, false);
gc.shardGraph(graphDirectory + "graph_chi/edges.tsv", 2);
}
long start = System.currentTimeMillis();
FileWriter writer = fileUtil.getFileWriter(outputBase + "settings.txt");
writer.write("KB used: " + kbDirectory + "\n");
writer.write("Graph used: " + graphDirectory + "\n");
writer.write("Splits used: " + splitsDirectory + "\n");
writer.write("Parameter file used: " + parameterFile + "\n");
writer.write("Parameters:\n");
fileUtil.copyLines(fileUtil.getBufferedReader(parameterFile), writer);
writer.write("End of parameters\n");
writer.close();
String relationsFile = splitsDirectory + "relations_to_run.tsv";
String line;
BufferedReader reader = fileUtil.getBufferedReader(relationsFile);
while ((line = reader.readLine()) != null) {
String relation = line;
long startTrainTime = System.currentTimeMillis();
if (isOnlineAug) {
// *******************
// Online Augmentation - Training Time
logger.info("Augmenting during training time");
// Augment during training time
// The test code modifies the directory
kb.setOutputDir(outputBase);
kb = Corpus.startTrainAugmentation(kb, relation, true);
// Bhushan, shard the graph. Num shards fixed at 2
Utils.deleteShards(graphDirectory + "graph_chi");
GraphCreator gc = new GraphCreator(outputBase, false);
gc.shardGraph(graphDirectory + "graph_chi/edges.tsv", 2);
// *******************
/* Reread all the graph files */
baseBuilder = new PraConfig.Builder();
parseGraphFiles(graphDirectory, baseBuilder);
baseBuilder.setFromParamFile(fileUtil.getBufferedReader(parameterFile));
Map<String, String> nodeNames = null;
if (fileUtil.fileExists(kbDirectory + "node_names.tsv")) {
nodeNames = fileUtil.readMapFromTsvFile(kbDirectory + "node_names.tsv", true);
}
Outputter outputter = new Outputter(baseBuilder.nodeDict, baseBuilder.edgeDict, nodeNames);
baseBuilder.setOutputter(outputter);
baseConfig = baseBuilder.build();
/* Finished Rereading the graph files */
}
PraConfig.Builder builder = new PraConfig.Builder(baseConfig);
logger.info("\n\n\n\nRunning PRA for relation " + relation);
boolean doCrossValidation = false;
parseKbFiles(kbDirectory, relation, builder, outputBase, fileUtil);
String outdir = fileUtil.addDirectorySeparatorIfNecessary(outputBase + relation);
fileUtil.mkdirs(outdir);
builder.setOutputBase(outdir);
initializeSplit(
splitsDirectory, kbDirectory, relation, builder, new DatasetFactory(), fileUtil);
PraConfig config = builder.build();
if (config.allData != null) {
doCrossValidation = true;
}
// Run PRA
if (doCrossValidation) {
new PraTrainAndTester().crossValidate(config, kb, isOnlineAug, relation, startTrainTime);
} else {
new PraTrainAndTester().trainAndTest(config, kb, isOnlineAug, relation, startTrainTime);
}
}
long end = System.currentTimeMillis();
long millis = end - start;
int seconds = (int) (millis / 1000);
int minutes = seconds / 60;
seconds = seconds - minutes * 60;
BufferedWriter out = new BufferedWriter(new FileWriter(outputBase + "/timings.txt", true));
out.write("Took " + minutes + " minutes and " + seconds + " seconds\n");
out.flush();
out.close();
// kb.closeDB();
System.out.println("Took " + minutes + " minutes and " + seconds + " seconds");
writer.close();
}
