public static ConnectionCosts build(String filename) throws IOException {
FileInputStream inputStream = new FileInputStream(filename);
InputStreamReader streamReader = new InputStreamReader(inputStream);
LineNumberReader lineReader = new LineNumberReader(streamReader);
String line = lineReader.readLine();
String[] dimensions = line.split("\\s+");
assert dimensions.length == 3;
int forwardSize = Integer.parseInt(dimensions[0]);
int backwardSize = Integer.parseInt(dimensions[1]);
assert forwardSize > 0 && backwardSize > 0;
ConnectionCosts costs = new ConnectionCosts(forwardSize, backwardSize);
while ((line = lineReader.readLine()) != null) {
String[] fields = line.split("\\s+");
assert fields.length == 3;
short forwardId = Short.parseShort(fields[0]);
short backwardId = Short.parseShort(fields[1]);
short cost = Short.parseShort(fields[2]);
costs.add(forwardId, backwardId, cost);
}
return costs;
}
private void updateNode(ViterbiNode[] viterbiNodes, ViterbiNode node) {
int backwardConnectionId = node.getLeftId();
int wordCost = node.getWordCost();
int leastPathCost = DEFAULT_COST;
for (ViterbiNode leftNode : viterbiNodes) {
// If array doesn't contain any more ViterbiNodes, continue to next index
if (leftNode == null) {
return;
} else {
// cost = [total cost from BOS to previous node] + [connection cost between previous node
// and current node] + [word cost]
int pathCost =
leftNode.getPathCost()
+ costs.get(leftNode.getRightId(), backwardConnectionId)
+ wordCost;
// Add extra cost for long nodes in "Search mode".
if (mode == TokenizerBase.Mode.SEARCH || mode == TokenizerBase.Mode.EXTENDED) {
pathCost += getPenaltyCost(node);
}
// If total cost is lower than before, set current previous node as best left node (previous
// means left).
if (pathCost < leastPathCost) {
leastPathCost = pathCost;
node.setPathCost(leastPathCost);
node.setLeftNode(leftNode);
}
}
}
}