public Dataset getDataset() throws ConfigurationException, DaoException {
if (createFakeGoldStandard) {
if (fakeGoldStandard == null) {
Corpus c =
env.getConfigurator()
.get(Corpus.class, "plain", "language", env.getDefaultLanguage().getLangCode());
try {
if (!c.exists()) c.create();
FakeDatasetCreator creator = new FakeDatasetCreator(c);
fakeGoldStandard = creator.generate(500);
} catch (IOException e) {
throw new DaoException(e);
}
}
return fakeGoldStandard;
} else {
DatasetDao dao = env.getConfigurator().get(DatasetDao.class);
List<Dataset> datasets = new ArrayList<Dataset>();
for (String name : datasetNames) {
datasets.addAll(
dao.getDatasetOrGroup(
language, name)); // throws a DaoException if language is incorrect.
}
return new Dataset(datasets); // merge all datasets together into one.
}
}
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);
}
