/**
* Initialise a CTMain before usage with a first tree, going through and setting up the hash table
* etc.
*
* @param root Root of the initial tree to initialise with
* @param noOfSamples Current number of samples taken
*/
public void initialize(TreeNode root, int noOfSamples) {
// Parameter initialisation
this.noOfSamples = noOfSamples;
noOfTrees = 0;
// TaxaMap initialisation
List<TreeNode> leaves = root.getLeaves();
noOfTaxa = leaves.size();
// Hash initialisation
hashUtils = new HashUtils();
hashUtils.initialize(noOfTaxa, noOfSamples, C, seed);
hashTable = new HashTable(hashUtils.m1);
// Taxamap initialisation
taxa = new TaxaMap(noOfTaxa);
for (int i = 0; i < leaves.size(); i++) {
taxa.put(leaves.get(i).name, i);
}
leafEdgeLengths = new double[noOfTaxa];
// Adds a single star partition, once and for all.
BitSet star = new BitSet(noOfTaxa);
star.flip(0, noOfTaxa);
HashEntry entry = new HashEntry(-1, star, 0.0d);
entry.count = noOfSamples + 1;
partitions.add(entry);
// Majority threshold initialisation
updateInterestThreshold();
}
/**
* Create partitions from an input tree - recursive so will be called by many nodes, beginning
* with root but with calculations actually beginning on leaves.
*
* @param partitions Partitions in the form of entries in the hash table
* @param curInterestPercentage Current percentage of interest that we want splits to occur above
* to view in the network later
*/
private ArrayList<Cluster> constructClusters(
LinkedList<HashEntry> partitions, double curInterestPercentage) {
ArrayList<Cluster> clusters = new ArrayList<Cluster>();
// Thresholds for the current run of the cluster builder (c.f. the threshold for the partitions
// list).
double curInterestThreshold = (double) (noOfTrees * (curInterestPercentage / 100.0d));
int majInterestThreshold = (int) (noOfTrees * (50.0 / 100.0d));
for (Iterator<HashEntry> it = partitions.iterator(); it.hasNext(); ) {
HashEntry entry = it.next();
// Checks if this partition is still above threshold... partition.
// If not: remove it - O(1).
if ((double) entry.count <= curInterestThreshold) {
// correct the isMajority flag to now refer to majority not if of interest or not..
if (entry.count <= majInterestThreshold) {
entry.isMajority = false;
}
it.remove();
continue;
}
// Constructs clusters (list of TreeNode's for each set bit) from each partition.
if ((double) entry.count > curInterestThreshold) {
Cluster cluster = new Cluster();
if (entry.count > majInterestThreshold) {
cluster.isMajority = true;
}
cluster.aboveSplit = entry.partition;
cluster.noOfOccurrences = entry.count;
cluster.edgeLength = entry.edgeLengthsSum / entry.count;
for (int i = 0; i < entry.partition.size(); i++) {
if (entry.partition.get(i)) {
TreeNode node = new TreeNode(taxa.getName(i));
node.edgeLength = leafEdgeLengths[i] / noOfTrees;
cluster.add(node);
}
}
clusters.add(cluster);
}
}
// Sort by number of taxa.
// TODO: This might obviously be optimized a bit, e.g. with a PriorityQueue. - Eiriksson
Collections.sort(clusters);
return clusters;
}
/** Constructs the majorityTree. Used to make Network so remember to call this first! */
public CTree constructMajorityTree() {
int numOfNode = 0;
CTree tree = new CTree();
// Creates the clusters.
tree.clusters = constructClusters(partitions, resPercentage);
// Begins by constructing the star tree.
for (TreeNode node : tree.clusters.get(0)) {
TreeNode root = tree.getRoot(); // Retrieves the root,
node.parent = root; // Parent of this node -> root.
root.children.add(node); // Adds this node as the children of the root.
tree.nodeList.add(node); // Adds this node to the list of nodes.
assert tree.nodeList.get(0).name.equals("root");
tree.parentList.put(node.name, 0); // Adds "this node -> root" parent mapping.
}
// Constructs internal nodes for the rest of the majority bi-partitions and rewires them.
for (int z = 1; z < tree.clusters.size(); z++) {
Cluster cluster = tree.clusters.get(z);
// only take the majority ones....!
if (cluster.isMajority == true) {
// 1. Retrieves the parent of the first node in this cluster.
TreeNode parent = tree.nodeList.get(tree.parentList.get(cluster.get(0).name));
// 2. Constructs a new internal node.
String nodeName = "int" + Integer.toString(numOfNode);
TreeNode internalNode = new TreeNode(nodeName);
internalNode.addProperty("noOfOccurrences", cluster.noOfOccurrences);
internalNode.edgeLength = cluster.edgeLength;
internalNode.parent = parent;
// 3. Insert the new node into the node list.
tree.nodeList.add(internalNode);
assert tree.nodeList.get(tree.nodeList.size() - 1).name.equals(internalNode.name);
tree.parentList.put(nodeName, tree.nodeList.size() - 1);
// update the clusters node references...
// (use a method of storing edges that referenced the positions in the Tree's nodelists for
// reference later.)
tree.clusters.get(z).nodeRefA = tree.parentList.get(cluster.get(0).name);
tree.clusters.get(z).nodeRefB = (tree.nodeList.size() - 1);
for (TreeNode node : cluster) {
// 4. Makes this node the child of the new internal node.
node.parent = internalNode;
assert node.parent.name.equals(tree.nodeList.get(tree.nodeList.size() - 1).name);
tree.parentList.put(node.name, tree.nodeList.size() - 1);
internalNode.children.add(node);
// 5. Delete the moved node(s) from the parent's children.
// TODO: optimize? probably not.
for (int i = 0; i < parent.children.size(); i++) {
if (parent.children.get(i).name.equals(node.name)) {
parent.children.remove(i);
break;
}
}
}
// Wires up the internal node.
parent.children.add(internalNode);
numOfNode++;
}
}
return tree;
}
/**
* Create partitions from an input tree - recursive so will be called by many nodes, beginning
* with root but with calculations actually beginning on leaves.
*
* @param node Node in tree with above split to add to the hash table
*/
private BitSet createPartitions(TreeNode node) {
if (node.isLeaf()) { // Leaf node.
int index = taxa.get(node.name);
// leaf simply has it's keys stored in the hash utilities
node.addProperty("tableHashKey", hashUtils.a1[index]);
node.addProperty("bucketHashKey", hashUtils.a2[index]);
// Updates the edge length array.
leafEdgeLengths[index] += node.edgeLength;
// Create a new partition from scratch to represent it
BitSet partition = new BitSet(noOfTaxa);
partition.set(index);
assert partition.cardinality() == 1 : "There should be exactly a single bit set.";
return partition;
} else { // An internal node: Traverses the tree in post order.
BitSet partition = new BitSet(noOfTaxa);
// Get the node's partition representation from its children
List<TreeNode> children = node.children;
for (TreeNode child : children) {
partition.or(createPartitions(child));
}
// if this node is NOT to the right side of the root then add it... (Avoid adding splits twice
// for CNetworks)
if (node.parent != root || root.getRight() != node) {
noOfPartitions++;
long tableKey = 0;
long bucketKey = 0;
long tableKey2 = 0;
long bucketKey2 = 0;
// Calculate the hash keys for this partition
for (TreeNode child : children) {
tableKey += child.getIntProperty("tableHashKey");
bucketKey += child.getIntProperty("bucketHashKey");
}
// if first is one then we need to store the flipped version so we store each split in one
// representation only.
if (partition.get(0) == true) {
// copy to a new partition that is the flipped version
BitSet partitionF = new BitSet(noOfTaxa);
for (int l = 0; l < noOfTaxa; l++) {
if (partition.get(l) == false) partitionF.set(l);
}
// calculate the hash keys for the flipped partition...
for (int k = 0; k < noOfTaxa; k++) {
if (partitionF.get(k) == true) {
tableKey2 += hashUtils.a1[k];
bucketKey2 += hashUtils.a2[k];
}
}
// store the properties in the node
node.addProperty("tableHashKey", (int) (tableKey2 % hashUtils.m1));
node.addProperty("bucketHashKey", (int) (bucketKey2 % hashUtils.m2));
if (noOfPartitions < noOfTaxa - 2) { // Avoids the addition of the star partition
hashTable.put(
partitionF,
node.edgeLength,
node.getIntProperty("tableHashKey"),
node.getIntProperty("bucketHashKey"),
interestThreshold,
partitions);
}
// remember to still return the original partition with its appropriate keys...
node.addProperty("tableHashKey", (int) (tableKey % hashUtils.m1));
node.addProperty("bucketHashKey", (int) (bucketKey % hashUtils.m2));
return partition;
}
// if first is zero then simply add the partition, recursively calculating the hash
else {
node.addProperty("tableHashKey", (int) (tableKey % hashUtils.m1));
node.addProperty("bucketHashKey", (int) (bucketKey % hashUtils.m2));
if (noOfPartitions < noOfTaxa - 2) { // Avoids the addition of the star partition
hashTable.put(
partition,
node.edgeLength,
node.getIntProperty("tableHashKey"),
node.getIntProperty("bucketHashKey"),
interestThreshold,
partitions);
}
return partition;
}
}
// still return the partition even if node was not added as was right of root...
return partition;
}
}
