edu.stanford.nlp.parser.lexparser.ParentAnnotationStats Maven / Gradle / Ivy
Show all versions of stanford-parser Show documentation
package edu.stanford.nlp.parser.lexparser;
import edu.stanford.nlp.io.NumberRangeFileFilter;
import edu.stanford.nlp.ling.StringLabelFactory;
import edu.stanford.nlp.trees.*;
import edu.stanford.nlp.stats.ClassicCounter;
import edu.stanford.nlp.stats.Counters;
import edu.stanford.nlp.util.Generics;
import edu.stanford.nlp.util.Pair;
import java.io.Reader;
import java.text.NumberFormat;
import java.util.*;
/**
* See what parent annotation helps in treebank, based on support and
* KL divergence.
*
* @author Christopher Manning
* @version 2003/01/04
*/
public class ParentAnnotationStats implements TreeVisitor {
private final TreebankLanguagePack tlp;
private ParentAnnotationStats(TreebankLanguagePack tlp, boolean doTags) {
this.tlp = tlp;
this.doTags = doTags;
}
private final boolean doTags;
private Map>> nodeRules = Generics.newHashMap();
private Map,ClassicCounter>> pRules = Generics.newHashMap();
private Map,ClassicCounter>> gPRules = Generics.newHashMap();
// corresponding ones for tags
private Map>> tagNodeRules = Generics.newHashMap();
private Map,ClassicCounter>> tagPRules = Generics.newHashMap();
private Map,ClassicCounter>> tagGPRules = Generics.newHashMap();
/**
* Minimum support * KL to be included in output and as feature
*/
public static final double[] CUTOFFS = {100.0, 200.0, 500.0, 1000.0};
/**
* Minimum support of parent annotated node for grandparent to be
* studied. Just there to reduce runtime and printout size.
*/
public static final double SUPPCUTOFF = 100.0;
/**
* Does whatever one needs to do to a particular parse tree
*/
public void visitTree(Tree t) {
processTreeHelper("TOP", "TOP", t);
}
public static List kidLabels(Tree t) {
Tree[] kids = t.children();
List l = new ArrayList(kids.length);
for (Tree kid : kids) {
l.add(kid.label().value());
}
return l;
}
public void processTreeHelper(String gP, String p, Tree t) {
if (!t.isLeaf() && (doTags || !t.isPreTerminal())) { // stop at words/tags
Map>> nr;
Map,ClassicCounter>> pr;
Map,ClassicCounter>> gpr;
if (t.isPreTerminal()) {
nr = tagNodeRules;
pr = tagPRules;
gpr = tagGPRules;
} else {
nr = nodeRules;
pr = pRules;
gpr = gPRules;
}
String n = t.label().value();
if (tlp != null) {
p = tlp.basicCategory(p);
gP = tlp.basicCategory(gP);
}
List kidn = kidLabels(t);
ClassicCounter> cntr = nr.get(n);
if (cntr == null) {
cntr = new ClassicCounter>();
nr.put(n, cntr);
}
cntr.incrementCount(kidn);
List pairStr = new ArrayList(2);
pairStr.add(n);
pairStr.add(p);
cntr = pr.get(pairStr);
if (cntr == null) {
cntr = new ClassicCounter>();
pr.put(pairStr, cntr);
}
cntr.incrementCount(kidn);
List tripleStr = new ArrayList(3);
tripleStr.add(n);
tripleStr.add(p);
tripleStr.add(gP);
cntr = gpr.get(tripleStr);
if (cntr == null) {
cntr = new ClassicCounter>();
gpr.put(tripleStr, cntr);
}
cntr.incrementCount(kidn);
Tree[] kids = t.children();
for (Tree kid : kids) {
processTreeHelper(p, n, kid);
}
}
}
public void printStats() {
NumberFormat nf = NumberFormat.getNumberInstance();
nf.setMaximumFractionDigits(2);
// System.out.println("Node rules");
// System.out.println(nodeRules);
// System.out.println("Parent rules");
// System.out.println(pRules);
// System.out.println("Grandparent rules");
// System.out.println(gPRules);
// Store java code for selSplit
StringBuffer[] javaSB = new StringBuffer[CUTOFFS.length];
for (int i = 0; i < CUTOFFS.length; i++) {
javaSB[i] = new StringBuffer(" private static String[] splitters" + (i + 1) + " = new String[] {");
}
ClassicCounter> allScores = new ClassicCounter>();
// do value of parent
for (String node : nodeRules.keySet()) {
ArrayList,Double>> answers = Generics.newArrayList();
ClassicCounter> cntr = nodeRules.get(node);
double support = (cntr.totalCount());
System.out.println("Node " + node + " support is " + support);
for (Iterator> it2 = pRules.keySet().iterator(); it2.hasNext();) {
List key = it2.next();
if (key.get(0).equals(node)) { // only do it if they match
ClassicCounter> cntr2 = pRules.get(key);
double support2 = (cntr2.totalCount());
double kl = Counters.klDivergence(cntr2, cntr);
System.out.println("KL(" + key + "||" + node + ") = " + nf.format(kl) + "\t" + "support(" + key + ") = " + support2);
double score = kl * support2;
answers.add(new Pair,Double>(key, new Double(score)));
allScores.setCount(key, score);
}
}
System.out.println("----");
System.out.println("Sorted descending support * KL");
Collections.sort(answers, (o1, o2) -> o2.second().compareTo(o1.second()));
for (int i = 0, size = answers.size(); i < size; i++) {
Pair p = (Pair) answers.get(i);
double psd = ((Double) p.second()).doubleValue();
System.out.println(p.first() + ": " + nf.format(psd));
if (psd >= CUTOFFS[0]) {
List lst = (List) p.first();
String nd = (String) lst.get(0);
String par = (String) lst.get(1);
for (int j = 0; j < CUTOFFS.length; j++) {
if (psd >= CUTOFFS[j]) {
javaSB[j].append("\"").append(nd).append("^");
javaSB[j].append(par).append("\", ");
}
}
}
}
System.out.println();
}
/*
// do value of parent with info gain -- yet to finish this
for (Iterator it = nodeRules.entrySet().iterator(); it.hasNext(); ) {
Map.Entry pair = (Map.Entry) it.next();
String node = (String) pair.getKey();
Counter cntr = (Counter) pair.getValue();
double support = (cntr.totalCount());
System.out.println("Node " + node + " support is " + support);
ArrayList dtrs = new ArrayList();
for (Iterator it2 = pRules.entrySet().iterator(); it2.hasNext();) {
HashMap annotated = new HashMap();
Map.Entry pair2 = (Map.Entry) it2.next();
List node2 = (List) pair2.getKey();
Counter cntr2 = (Counter) pair2.getValue();
if (node2.get(0).equals(node)) { // only do it if they match
annotated.put(node2, cntr2);
}
}
// upto
List answers = new ArrayList();
System.out.println("----");
System.out.println("Sorted descending support * KL");
Collections.sort(answers,
new Comparator() {
public int compare(Object o1, Object o2) {
Pair p1 = (Pair) o1;
Pair p2 = (Pair) o2;
Double p12 = (Double) p1.second();
Double p22 = (Double) p2.second();
return p22.compareTo(p12);
}
});
for (int i = 0, size = answers.size(); i < size; i++) {
Pair p = (Pair) answers.get(i);
double psd = ((Double) p.second()).doubleValue();
System.out.println(p.first() + ": " + nf.format(psd));
if (psd >= CUTOFFS[0]) {
List lst = (List) p.first();
String nd = (String) lst.get(0);
String par = (String) lst.get(1);
for (int j=0; j < CUTOFFS.length; j++) {
if (psd >= CUTOFFS[j]) {
javaSB[j].append("\"").append(nd).append("^");
javaSB[j].append(par).append("\", ");
}
}
}
}
System.out.println();
}
*/
// do value of grandparent
for (List node : pRules.keySet()) {
ArrayList, Double>> answers = Generics.newArrayList();
ClassicCounter> cntr = pRules.get(node);
double support = (cntr.totalCount());
if (support < SUPPCUTOFF) {
continue;
}
System.out.println("Node " + node + " support is " + support);
for (List key : gPRules.keySet()) {
if (key.get(0).equals(node.get(0)) && key.get(1).equals(node.get(1))) { // only do it if they match
ClassicCounter> cntr2 = gPRules.get(key);
double support2 = (cntr2.totalCount());
double kl = Counters.klDivergence(cntr2, cntr);
System.out.println("KL(" + key + "||" + node + ") = " + nf.format(kl) + "\t" + "support(" + key + ") = " + support2);
double score = kl * support2;
answers.add(Pair.makePair(key, new Double(score)));
allScores.setCount(key,score);
}
}
System.out.println("----");
System.out.println("Sorted descending support * KL");
Collections.sort(answers, (o1, o2) -> o2.second().compareTo(o1.second()));
for (int i = 0, size = answers.size(); i < size; i++) {
Pair p = (Pair) answers.get(i);
double psd = ((Double) p.second()).doubleValue();
System.out.println(p.first() + ": " + nf.format(psd));
if (psd >= CUTOFFS[0]) {
List lst = (List) p.first();
String nd = (String) lst.get(0);
String par = (String) lst.get(1);
String gpar = (String) lst.get(2);
for (int j = 0; j < CUTOFFS.length; j++) {
if (psd >= CUTOFFS[j]) {
javaSB[j].append("\"").append(nd).append("^");
javaSB[j].append(par).append("~");
javaSB[j].append(gpar).append("\", ");
}
}
}
}
System.out.println();
}
System.out.println();
System.out.println("All scores:");
edu.stanford.nlp.util.PriorityQueue> pq = Counters.toPriorityQueue(allScores);
while (! pq.isEmpty()) {
List key = pq.getFirst();
double score = pq.getPriority(key);
pq.removeFirst();
System.out.println(key + "\t" + score);
}
System.out.println(" // Automatically generated by ParentAnnotationStats -- preferably don't edit");
for (int i = 0; i < CUTOFFS.length; i++) {
int len = javaSB[i].length();
javaSB[i].replace(len - 2, len, "};");
System.out.println(javaSB[i]);
}
System.out.print(" public static HashSet splitters = new HashSet(Arrays.asList(");
for (int i = CUTOFFS.length; i > 0; i--) {
if (i == 1) {
System.out.print("splitters1");
} else {
System.out.print("selectiveSplit" + i + " ? splitters" + i + " : (");
}
}
// need to print extra one to close other things open
for (int i = CUTOFFS.length; i >= 0; i--) {
System.out.print(")");
}
System.out.println(";");
}
private static void getSplitters(double cutOff, Map>> nr,
Map,ClassicCounter>> pr,
Map,ClassicCounter>> gpr,
Set splitters) {
// do value of parent
for (String node : nr.keySet()) {
List,Double>> answers = new ArrayList,Double>>();
ClassicCounter> cntr = nr.get(node);
double support = (cntr.totalCount());
for (List key : pr.keySet()) {
if (key.get(0).equals(node)) { // only do it if they match
ClassicCounter> cntr2 = pr.get(key);
double support2 = cntr2.totalCount();
double kl = Counters.klDivergence(cntr2, cntr);
answers.add(new Pair, Double>(key, new Double(kl * support2)));
}
}
Collections.sort(answers, (o1, o2) -> o2.second().compareTo(o1.second()));
for (int i = 0, size = answers.size(); i < size; i++) {
Pair,Double> p = answers.get(i);
double psd = p.second().doubleValue();
if (psd >= cutOff) {
List lst = p.first();
String nd = lst.get(0);
String par = lst.get(1);
String name = nd + "^" + par;
splitters.add(name);
}
}
}
/*
// do value of parent with info gain -- yet to finish this
for (Iterator it = nr.entrySet().iterator(); it.hasNext(); ) {
Map.Entry pair = (Map.Entry) it.next();
String node = (String) pair.getKey();
Counter cntr = (Counter) pair.getValue();
double support = (cntr.totalCount());
ArrayList dtrs = new ArrayList();
for (Iterator it2 = pr.entrySet().iterator(); it2.hasNext();) {
HashMap annotated = new HashMap();
Map.Entry pair2 = (Map.Entry) it2.next();
List node2 = (List) pair2.getKey();
Counter cntr2 = (Counter) pair2.getValue();
if (node2.get(0).equals(node)) { // only do it if they match
annotated.put(node2, cntr2);
}
}
// upto
List answers = new ArrayList();
Collections.sort(answers,
new Comparator() {
public int compare(Object o1, Object o2) {
Pair p1 = (Pair) o1;
Pair p2 = (Pair) o2;
Double p12 = (Double) p1.second();
Double p22 = (Double) p2.second();
return p22.compareTo(p12);
}
});
for (int i = 0, size = answers.size(); i < size; i++) {
Pair p = (Pair) answers.get(i);
double psd = ((Double) p.second()).doubleValue();
if (psd >= cutOff) {
List lst = (List) p.first();
String nd = (String) lst.get(0);
String par = (String) lst.get(1);
String name = nd + "^" + par;
splitters.add(name);
}
}
}
*/
// do value of grandparent
for (List node : pr.keySet()) {
ArrayList,Double>> answers = Generics.newArrayList();
ClassicCounter> cntr = pr.get(node);
double support = (cntr.totalCount());
if (support < SUPPCUTOFF) {
continue;
}
for (List key : gpr.keySet()) {
if (key.get(0).equals(node.get(0)) && key.get(1).equals(node.get(1))) {
// only do it if they match
ClassicCounter> cntr2 = gpr.get(key);
double support2 = (cntr2.totalCount());
double kl = Counters.klDivergence(cntr2, cntr);
answers.add(new Pair,Double>(key, new Double(kl * support2)));
}
}
Collections.sort(answers, (o1, o2) -> o2.second().compareTo(o1.second()));
for (int i = 0, size = answers.size(); i < size; i++) {
Pair p = (Pair) answers.get(i);
double psd = ((Double) p.second()).doubleValue();
if (psd >= cutOff) {
List lst = (List) p.first();
String nd = (String) lst.get(0);
String par = (String) lst.get(1);
String gpar = (String) lst.get(2);
String name = nd + "^" + par + "~" + gpar;
splitters.add(name);
}
}
}
}
/**
* Calculate parent annotation statistics suitable for doing
* selective parent splitting in the PCFGParser inside
* FactoredParser.
* Usage: java edu.stanford.nlp.parser.lexparser.ParentAnnotationStats
* [-tags] treebankPath
*
* @param args One argument: path to the Treebank
*/
public static void main(String[] args) {
boolean doTags = false;
if (args.length < 1) {
System.out.println("Usage: java edu.stanford.nlp.parser.lexparser.ParentAnnotationStats [-tags] treebankPath");
} else {
int i = 0;
boolean useCutOff = false;
double cutOff = 0.0;
while (args[i].startsWith("-")) {
if (args[i].equals("-tags")) {
doTags = true;
i++;
} else if (args[i].equals("-cutOff") && i + 1 < args.length) {
useCutOff = true;
cutOff = Double.parseDouble(args[i + 1]);
i += 2;
} else {
System.err.println("Unknown option: " + args[i]);
i++;
}
}
Treebank treebank = new DiskTreebank(in -> new PennTreeReader(in, new LabeledScoredTreeFactory(new StringLabelFactory()), new BobChrisTreeNormalizer()));
treebank.loadPath(args[i]);
if (useCutOff) {
Set splitters = getSplitCategories(treebank, doTags, 0, cutOff, cutOff, null);
System.out.println(splitters);
} else {
ParentAnnotationStats pas = new ParentAnnotationStats(null, doTags);
treebank.apply(pas);
pas.printStats();
}
}
}
/**
* Call this method to get a String array of categories to split on.
* It calculates parent annotation statistics suitable for doing
* selective parent splitting in the PCFGParser inside
* FactoredParser.
* If tlp is non-null tlp.basicCategory() will be called on parent and
* grandparent nodes.
* This version just defaults some parameters.
* Implementation note: This method is not designed for concurrent
* invocation: it uses static state variables.
*/
public static Set getSplitCategories(Treebank t, double cutOff, TreebankLanguagePack tlp) {
return getSplitCategories(t, true, 0, cutOff, cutOff, tlp);
}
/**
* Call this method to get a String array of categories to split on.
* It calculates parent annotation statistics suitable for doing
* selective parent splitting in the PCFGParser inside
* FactoredParser.
* If tlp is non-null tlp.basicCategory() will be called on parent and
* grandparent nodes.
* Implementation note: This method is not designed for concurrent
* invocation: it uses static state variables.
*/
public static Set getSplitCategories(Treebank t, boolean doTags, int algorithm, double phrasalCutOff, double tagCutOff, TreebankLanguagePack tlp) {
ParentAnnotationStats pas = new ParentAnnotationStats(tlp, doTags);
t.apply(pas);
Set splitters = Generics.newHashSet();
pas.getSplitters(phrasalCutOff, pas.nodeRules, pas.pRules, pas.gPRules, splitters);
pas.getSplitters(tagCutOff, pas.tagNodeRules, pas.tagPRules, pas.tagGPRules, splitters);
return splitters;
}
/**
* This is hardwired to calculate the split categories from English
* Penn Treebank sections 2-21 with a default cutoff of 300 (as used
* in ACL03PCFG). It was added to upgrading of code in cases where no
* Treebank was available, and the pre-stored list was being used).
*/
public static Set getEnglishSplitCategories(String treebankRoot) {
TreebankLangParserParams tlpParams = new EnglishTreebankParserParams();
Treebank trees = tlpParams.memoryTreebank();
trees.loadPath(treebankRoot, new NumberRangeFileFilter(200, 2199, true));
return getSplitCategories(trees, 300.0, tlpParams.treebankLanguagePack());
}
}