/**
* Here we set up the PraConfig items that have to do with the input KB files. In particular, that
* means deciding which relations are known to be inverses of each other, which edges should be
* ignored because using them to predict new relations instances would consitute cheating, and
* setting the range and domain of a relation to restrict new predictions.
*
* <p>Also, if the relations have been embedded into a latent space, we perform a mapping here
* when deciding which edges to ignore. This means that each embedding of a KB graph has to have a
* different directory.
*/
public void parseKbFiles(
String directory,
String relation,
PraConfig.Builder builder,
String outputBase,
FileUtil fileUtil)
throws IOException {
// TODO(matt): allow this to be left unspecified.
Map<Integer, Integer> inverses = createInverses(directory + "inverses.tsv", builder.edgeDict);
builder.setRelationInverses(inverses);
Map<String, List<String>> embeddings = null;
if (fileUtil.fileExists(directory + "embeddings.tsv")) {
embeddings = fileUtil.readMapListFromTsvFile(directory + "embeddings.tsv");
}
List<Integer> unallowedEdges =
createUnallowedEdges(relation, inverses, embeddings, builder.edgeDict);
builder.setUnallowedEdges(unallowedEdges);
if (fileUtil.fileExists(directory + "ranges.tsv")) {
Map<String, String> ranges = fileUtil.readMapFromTsvFile(directory + "ranges.tsv");
String range = ranges.get(relation);
String fixed = range.replace("/", "_");
String cat_file = directory + "category_instances" + File.separator + fixed;
// Bhushan
cat_file = cat_file.replace(":", "_");
Set<Integer> allowedTargets = fileUtil.readIntegerSetFromFile(cat_file, builder.nodeDict);
builder.setAllowedTargets(allowedTargets);
} else {
FileWriter writer =
fileUtil.getFileWriter(outputBase + "settings.txt", true); // true -> append
writer.write("No range file found! I hope your accept policy is as you want it...\n");
System.out.println("No range file found!");
writer.close();
}
}
