public WikidataEntity parse(String json) throws WpParseException {
JacksonTermedStatementDocument mwDoc;
try {
mwDoc = mapper.readValue(json, JacksonTermedStatementDocument.class);
} catch (IOException e) {
LOG.info("Error parsing: " + json);
throw new WpParseException(e);
}
WikidataEntity record = new WikidataEntity(mwDoc.getEntityId().getId());
// Aliases (multiple per language)
for (List<MonolingualTextValue> vlist : mwDoc.getAliases().values()) {
if (vlist.isEmpty()) continue;
if (!validLanguage(vlist.get(0).getLanguageCode())) continue;
Language lang = Language.getByLangCodeLenient(vlist.get(0).getLanguageCode());
record.getAliases().put(lang, new ArrayList<String>());
for (MonolingualTextValue v : vlist) {
record.getAliases().get(lang).add(v.getText());
}
}
// Descriptions (one per language)
for (MonolingualTextValue v : mwDoc.getDescriptions().values()) {
if (validLanguage(v.getLanguageCode())) {
Language lang = Language.getByLangCodeLenient(v.getLanguageCode());
record.getDescriptions().put(lang, v.getText());
}
}
// Labels (one per language)
for (MonolingualTextValue v : mwDoc.getLabels().values()) {
if (validLanguage(v.getLanguageCode())) {
Language lang = Language.getByLangCodeLenient(v.getLanguageCode());
record.getLabels().put(lang, v.getText());
}
}
// Claims (only for Item entities)
if (mwDoc instanceof JacksonItemDocument) {
for (List<JacksonStatement> statements :
((JacksonItemDocument) mwDoc).getJsonClaims().values()) {
for (JacksonStatement s : statements) {
record.getStatements().add(parseOneClaim(record, s));
}
}
}
return record;
}
/**
* This method takes care to not load the metric itself, and just deal in names. Once the metric
* is loaded, it has already accessed its data files.
*
* @throws ConfigurationException
*/
public void deleteDataDirectories() throws ConfigurationException {
for (String name : getSubmetrics(metricName)) {
File dir = FileUtils.getFile(srDir, name, language.getLangCode());
if (dir.exists()) {
LOG.info("deleting metric directory " + dir);
FileUtils.deleteQuietly(dir);
}
}
}
private void buildConceptsIfNecessary() throws IOException, ConfigurationException, DaoException {
boolean needsConcepts = false;
for (String name : getSubmetrics(metricName)) {
String type = getMetricType(name);
if (type.equals("sparsevector.esa") || type.equals("sparsevector.mostsimilarconcepts")) {
needsConcepts = true;
}
}
if (!needsConcepts) {
return;
}
File path =
FileUtils.getFile(
env.getConfiguration().get().getString("sr.concepts.path"),
language.getLangCode() + ".txt");
path.getParentFile().mkdirs();
// Check to see if concepts are already built
if (path.isFile() && FileUtils.readLines(path).size() > 1) {
return;
}
LOG.info("building concept file " + path.getAbsolutePath() + " for " + metricName);
SRConceptSpaceGenerator gen =
new SRConceptSpaceGenerator(
language,
env.getConfigurator().get(LocalLinkDao.class),
env.getConfigurator().get(LocalPageDao.class));
gen.writeConcepts(path);
LOG.info(
"finished creating concept file "
+ path.getAbsolutePath()
+ " with "
+ FileUtils.readLines(path).size()
+ " lines");
}
private boolean validLanguage(String langCode) {
return Language.hasLangCode(langCode) && langs.containsLanguage(langCode);
}
public static void main(String args[])
throws ConfigurationException, IOException, WikiBrainException, DaoException {
Options options = new Options();
// Number of Max Results(otherwise take from config)
options.addOption(
new DefaultOptionBuilder()
.hasArg()
.withLongOpt("max-results")
.withDescription("maximum number of results")
.create("r"));
// Specify the Datasets
options.addOption(
new DefaultOptionBuilder()
.hasArgs()
.withLongOpt("gold")
.withDescription("the set of gold standard datasets to train on")
.create("g"));
// Delete existing data models
options.addOption(
new DefaultOptionBuilder()
.hasArg()
.withLongOpt("delete")
.withDescription(
"delete all existing SR data for the metric and its submetrics (true or false, default is true)")
.create("d"));
// Specify the Metrics
options.addOption(
new DefaultOptionBuilder()
.hasArg()
.withLongOpt("metric")
.withDescription("set a local metric")
.create("m"));
// Row and column ids for most similar caches
options.addOption(
new DefaultOptionBuilder()
.hasArg()
.withLongOpt("rowids")
.withDescription("page ids for rows of cosimilarity matrices (implies -s)")
.create("p"));
options.addOption(
new DefaultOptionBuilder()
.hasArg()
.withLongOpt("colids")
.withDescription("page ids for columns of cosimilarity matrices (implies -s)")
.create("q"));
// build the cosimilarity matrix
options.addOption(
new DefaultOptionBuilder()
.withLongOpt("cosimilarity")
.withDescription("build cosimilarity matrices")
.create("s"));
// sets the mode
options.addOption(
new DefaultOptionBuilder()
.withLongOpt("mode")
.hasArg()
.withDescription("mode: similarity, mostsimilar, or both")
.create("o"));
// add option for valid most similar ids
options.addOption(
new DefaultOptionBuilder()
.withLongOpt("validMostSimilarIds")
.withDescription("Set valid most similar ids")
.create("y"));
// when building pairwise cosine and ensembles, don't rebuild already built sub-metrics.
options.addOption(
new DefaultOptionBuilder()
.withLongOpt("skip-built")
.withDescription("Don't rebuild already built bmetrics (implies -d false)")
.create("k"));
// when building pairwise cosine and ensembles, don't rebuild already built sub-metrics.
options.addOption(
new DefaultOptionBuilder()
.withLongOpt("fake")
.withDescription("Create a fake gold standard for the language.")
.create("f"));
EnvBuilder.addStandardOptions(options);
CommandLineParser parser = new PosixParser();
CommandLine cmd;
try {
cmd = parser.parse(options, args);
} catch (ParseException e) {
System.err.println("Invalid option usage: " + e.getMessage());
new HelpFormatter().printHelp("SRBuilder", options);
return;
}
Env env = new EnvBuilder(cmd).build();
String metric = cmd.hasOption("m") ? cmd.getOptionValue("m") : null;
SRBuilder builder = new SRBuilder(env, metric);
if (cmd.hasOption("g")) {
builder.setDatasetNames(Arrays.asList(cmd.getOptionValues("g")));
}
if (cmd.hasOption("p")) {
builder.setRowIdsFromFile(cmd.getOptionValue("p"));
builder.setBuildCosimilarity(true);
}
if (cmd.hasOption("q")) {
builder.setColIdsFromFile(cmd.getOptionValue("q"));
builder.setBuildCosimilarity(true);
}
if (cmd.hasOption("y")) {
builder.setValidMostSimilarIdsFromFile(cmd.getOptionValue("y"));
}
if (cmd.hasOption("s")) {
builder.setBuildCosimilarity(true);
}
if (cmd.hasOption("k")) {
builder.setSkipBuiltMetrics(true);
builder.setDeleteExistingData(false);
}
if (cmd.hasOption("d")) {
builder.setDeleteExistingData(Boolean.valueOf(cmd.getOptionValue("d")));
}
if (cmd.hasOption("o")) {
builder.setMode(Mode.valueOf(cmd.getOptionValue("o").toUpperCase()));
}
if (cmd.hasOption("l")) {
builder.setLanguage(Language.getByLangCode(cmd.getOptionValue("l")));
}
if (cmd.hasOption("r")) {
builder.setMaxResults(Integer.valueOf(cmd.getOptionValue("r")));
}
if (cmd.hasOption("f")) {
builder.setCreateFakeGoldStandard(true);
}
builder.build();
}
private void initWord2Vec(String name) throws ConfigurationException, IOException, DaoException {
Config config = getMetricConfig(name).getConfig("generator");
File model = Word2VecGenerator.getModelFile(config.getString("modelDir"), language);
if (skipBuiltMetrics && model.isFile()) {
return;
}
if (config.hasPath("prebuilt") && config.getBoolean("prebuilt")) {
if (model.isFile()) {
return;
}
File downloadPath = new File(config.getString("binfile"));
if (!downloadPath.isFile()) {
throw new ConfigurationException(
"word2vec model "
+ downloadPath.getAbsolutePath()
+ " cannot be found."
+ " You must download it from "
+ config.getString("url")
+ " into to the wikibrain download directory.");
}
if (!config.getStringList("languages").contains(language.getLangCode())) {
throw new ConfigurationException(
"word2vec model " + downloadPath + " does not support language" + language);
}
if (downloadPath.toString().toLowerCase().endsWith("gz")) {
LOG.info("decompressing " + downloadPath + " to " + model);
File tmp = File.createTempFile("word2vec", "bin");
try {
FileUtils.deleteQuietly(tmp);
GZIPInputStream gz = new GZIPInputStream(new FileInputStream(downloadPath));
FileUtils.copyInputStreamToFile(gz, tmp);
gz.close();
model.getParentFile().mkdirs();
FileUtils.moveFile(tmp, model);
} finally {
FileUtils.deleteQuietly(tmp);
}
} else {
FileUtils.copyFile(downloadPath, model);
}
return;
}
LinkProbabilityDao lpd = env.getConfigurator().get(LinkProbabilityDao.class);
lpd.useCache(true);
if (!lpd.isBuilt()) {
lpd.build();
}
String corpusName = config.getString("corpus");
Corpus corpus = null;
if (!corpusName.equals("NONE")) {
corpus =
env.getConfigurator()
.get(Corpus.class, config.getString("corpus"), "language", language.getLangCode());
if (!corpus.exists()) {
corpus.create();
}
}
if (model.isFile()
&& (corpus == null || model.lastModified() > corpus.getCorpusFile().lastModified())) {
return;
}
if (corpus == null) {
throw new ConfigurationException(
"word2vec metric "
+ name
+ " cannot build or find model!"
+ "configuration has no corpus, but model not found at "
+ model
+ ".");
}
Word2VecTrainer trainer =
new Word2VecTrainer(env.getConfigurator().get(LocalPageDao.class), language);
if (config.hasPath("dimensions")) {
LOG.info("set number of dimensions to " + config.getInt("dimensions"));
trainer.setLayer1Size(config.getInt("dimensions"));
}
if (config.hasPath("maxWords")) {
LOG.info("set maxWords to " + config.getInt("maxWords"));
trainer.setMaxWords(config.getInt("maxWords"));
}
if (config.hasPath("window")) {
LOG.info("set window to " + config.getInt("maxWords"));
trainer.setWindow(config.getInt("window"));
}
trainer.setKeepAllArticles(true);
trainer.train(corpus.getDirectory());
trainer.save(model);
}
public synchronized SRMetric getMetric(String name) throws ConfigurationException {
return env.getConfigurator().get(SRMetric.class, name, "language", language.getLangCode());
}
