/**
* Assign unique ids to leaf nodes.
*
* @param graph
* @return the number of different leaf nodes
*/
private int setUniqueLeafNodeIds(DirectedGraph graph) {
int i = 0;
for (LeafNode l : graph.getLeafNodes()) {
l.setUniqueLeafId(++i);
}
return i;
}
public void run() {
try {
long startTime = System.currentTimeMillis();
logger.debug(id + "> Creating " + descFile.getName());
createDescFile();
logger.debug(id + "> Dumping features to " + featFile.getName());
dumpFeatureVectors();
logger.debug(id + "> Dumping distance matrix to " + distFile.getName());
binarySaveDistanceMatrix();
logger.debug(id + "> Calling wagon as follows:");
logger.debug(systemCall);
Process p = Runtime.getRuntime().exec(systemCall);
// collect the output
// read from error stream
StreamGobbler errorGobbler = new StreamGobbler(p.getErrorStream(), id + " err");
// read from output stream
StreamGobbler outputGobbler = new StreamGobbler(p.getInputStream(), id + " out");
// start reading from the streams
errorGobbler.start();
outputGobbler.start();
p.waitFor();
if (p.exitValue() != 0) {
finished = true;
success = false;
} else {
success = true;
logger.debug(id + "> Wagon call took " + (System.currentTimeMillis() - startTime) + " ms");
// read in the resulting CART
logger.debug(id + "> Reading CART");
BufferedReader buf = new BufferedReader(new FileReader(cartFile));
WagonCARTReader wagonReader = new WagonCARTReader(LeafType.IntAndFloatArrayLeafNode);
cart = new CART(wagonReader.load(buf, featureDefinition), featureDefinition);
buf.close();
// Fix the new cart's leaves:
// They are currently the index numbers in featureVectors;
// but what we need is the unit index numbers!
for (LeafNode leaf : cart.getLeafNodes()) {
int[] data = (int[]) leaf.getAllData();
for (int i = 0; i < data.length; i++) {
data[i] = fv[data[i]].getUnitIndex();
}
}
logger.debug(id + "> completed in " + (System.currentTimeMillis() - startTime) + " ms");
finished = true;
}
if (!Boolean.getBoolean("wagon.keepfiles")) {
featFile.delete();
distFile.delete();
}
} catch (Exception e) {
e.printStackTrace();
finished = true;
success = false;
throw new RuntimeException("Exception running wagon");
}
}
private void printLeafNodes(DirectedGraph graph, DataOutput out, PrintWriter pw)
throws IOException {
for (LeafNode leaf : graph.getLeafNodes()) {
if (leaf.getUniqueLeafId() == 0) // empty leaf, do not write
continue;
LeafType leafType = leaf.getLeafNodeType();
if (leafType == LeafType.FeatureVectorLeafNode) {
leafType = LeafType.IntArrayLeafNode;
// save feature vector leaf nodes as int array leaf nodes
}
if (out != null) {
// Leaf node type
out.writeInt(leafType.ordinal());
}
if (pw != null) {
pw.print("id" + leaf.getUniqueLeafId() + " " + leafType);
}
switch (leaf.getLeafNodeType()) {
case IntArrayLeafNode:
int data[] = ((IntArrayLeafNode) leaf).getIntData();
// Number of data points following:
if (out != null) out.writeInt(data.length);
if (pw != null) pw.print(" " + data.length);
// for each index, write the index
for (int i = 0; i < data.length; i++) {
if (out != null) out.writeInt(data[i]);
if (pw != null) pw.print(" " + data[i]);
}
break;
case FloatLeafNode:
float stddev = ((FloatLeafNode) leaf).getStDeviation();
float mean = ((FloatLeafNode) leaf).getMean();
if (out != null) {
out.writeFloat(stddev);
out.writeFloat(mean);
}
if (pw != null) {
pw.print(" 1 " + stddev + " " + mean);
}
break;
case IntAndFloatArrayLeafNode:
case StringAndFloatLeafNode:
int data1[] = ((IntAndFloatArrayLeafNode) leaf).getIntData();
float floats[] = ((IntAndFloatArrayLeafNode) leaf).getFloatData();
// Number of data points following:
if (out != null) out.writeInt(data1.length);
if (pw != null) pw.print(" " + data1.length);
// for each index, write the index and then its float
for (int i = 0; i < data1.length; i++) {
if (out != null) {
out.writeInt(data1[i]);
out.writeFloat(floats[i]);
}
if (pw != null) pw.print(" " + data1[i] + " " + floats[i]);
}
break;
case FeatureVectorLeafNode:
FeatureVector fv[] = ((FeatureVectorLeafNode) leaf).getFeatureVectors();
// Number of data points following:
if (out != null) out.writeInt(fv.length);
if (pw != null) pw.print(" " + fv.length);
// for each feature vector, write the index
for (int i = 0; i < fv.length; i++) {
if (out != null) out.writeInt(fv[i].getUnitIndex());
if (pw != null) pw.print(" " + fv[i].getUnitIndex());
}
break;
case PdfLeafNode:
throw new IllegalArgumentException("Writing of pdf leaf nodes not yet implemented");
}
if (pw != null) pw.println();
}
}