Examples with DiscreteFeature edu.illinois.cs.cogcomp.edison.features.DiscreteFeature used on opensource projects

Example 6 with DiscreteFeature

use of edu.illinois.cs.cogcomp.edison.features.DiscreteFeature in project cogcomp-nlp by CogComp.

the class LabelOneBefore method getFeatures.

@Override
public Set<Feature> getFeatures(Constituent c) throws EdisonException {
    String classifier;
    String prefix = "LabelOneBefore";
    TextAnnotation ta = c.getTextAnnotation();
    int start = c.getStartSpan() - 1;
    int end = c.getEndSpan() - 1;
    Set<Feature> features = new LinkedHashSet<>();
    for (int i = start; i < end; i++) {
        if (!isPOSFromCounting) {
            classifier = prefix + "_" + "POS";
            if (i >= 0) {
                TokenLabelView POSView = (TokenLabelView) ta.getView(ViewNames.POS);
                String form = ta.getToken(i);
                String tag = POSView.getLabel(i);
                features.add(new DiscreteFeature(classifier + ":" + tag + "_" + form));
            } else
                features.add(new DiscreteFeature(classifier + ":" + ""));
        } else if (isBaseLineCounting) {
            classifier = prefix + "_" + "BaselinePOS";
            if (i >= 0) {
                String form = ta.getToken(i);
                String tag = counter.tag(i, ta);
                features.add(new DiscreteFeature(classifier + ":" + tag + "_" + form));
            } else
                features.add(new DiscreteFeature(classifier + ":" + ""));
        } else {
            classifier = prefix + "_" + "MikheevPOS";
            if (i >= 0) {
                String form = ta.getToken(i);
                String tag = counter.tag(i, ta);
                features.add(new DiscreteFeature(classifier + ":" + tag + "_" + form));
            } else
                features.add(new DiscreteFeature(classifier + ":" + ""));
        }
    }
    return features;
}

Example 7 with DiscreteFeature

use of edu.illinois.cs.cogcomp.edison.features.DiscreteFeature in project cogcomp-nlp by CogComp.

the class WordConjunctionOneTwoThreeGramWindowTwo method getFeatures.

@Override
public /**
 * This feature extractor assumes that the TOKEN View has been generated in the Constituents TextAnnotation.
 * It generate a feature for a window [-2, +2] of Forms (original text) for each constituent.
 */
Set<Feature> getFeatures(Constituent c) throws EdisonException {
    TextAnnotation ta = c.getTextAnnotation();
    TOKENS = ta.getView(ViewNames.TOKENS);
    // We can assume that the constituent in this case is a Word(Token)
    int startspan = c.getStartSpan();
    int endspan = c.getEndSpan();
    // k is 3 since we need up to 3-grams
    int k = 3;
    int window = 2;
    // All our constituents are words(tokens)
    String[] forms = getWindowK(TOKENS, startspan, endspan, window);
    String id, value;
    String classifier = "WordConjunctionOneTwoThreeGramWindowTwo";
    Set<Feature> result = new LinkedHashSet<>();
    for (int j = 0; j < k; j++) {
        for (int i = 0; i < forms.length; i++) {
            // forms.length = 5, So i goes from 0 to 4, for each String in the forms array.
            StringBuilder f = new StringBuilder();
            // and three word combinations within [-2,2] window or words.
            for (int context = 0; context <= j && i + context < forms.length; context++) {
                // add a '_' between words to conjoin them together
                if (context != 0) {
                    f.append("_");
                }
                f.append(forms[i + context]);
            }
            // 2 is the center object in the array so i should go from -2 to +2 (with 0 being
            // the center)
            // j is the size of the n-gram so it goes 1 to 3
            id = classifier + ":" + ((i - window) + "_" + (j + 1));
            value = "(" + (f.toString()) + ")";
            result.add(new DiscreteFeature(id + value));
        }
    }
    return result;
}

Example 8 with DiscreteFeature

use of edu.illinois.cs.cogcomp.edison.features.DiscreteFeature in project cogcomp-nlp by CogComp.

the class POSandPositionWindowThree method getFeatures.

@Override
public /**
 * This feature extractor assumes that the TOKEN View and POS View have been
 * generated in the Constituents TextAnnotation. It will use its own POS tag and well as the POS tag
 * and the SHALLOW_PARSE (Chunk) labels of the previous two tokens and return it as a discrete feature.
 */
Set<Feature> getFeatures(Constituent c) throws EdisonException {
    String classifier = "POSandPositionWindowThree";
    TextAnnotation ta = c.getTextAnnotation();
    TOKENS = ta.getView(ViewNames.TOKENS);
    POS = ta.getView(ViewNames.POS);
    // We can assume that the constituent in this case is a Word(Token) described by the LBJ
    // chunk definition
    int startspan = c.getStartSpan();
    int endspan = c.getEndSpan();
    int before = 3;
    int after = 3;
    // All our constituents are words(tokens)
    String[] tags = new String[before + after + 1];
    // three words before
    int k = -3;
    List<Constituent> wordsthreebefore = getwordskfrom(TOKENS, startspan, endspan, k);
    int i = 0;
    for (Constituent token : wordsthreebefore) {
        // Should only be one POS tag for each token
        List<String> POS_tag = POS.getLabelsCoveringSpan(token.getStartSpan(), token.getEndSpan());
        if (POS_tag.size() != 1) {
            logger.warn("Error token has more than one POS tag.");
        }
        tags[i] = POS_tag.get(0);
        i++;
    }
    tags[i] = POS.getLabelsCoveringSpan(c.getStartSpan(), c.getEndSpan()).get(0);
    i++;
    // three words after
    k = 3;
    List<Constituent> wordsthreeafter = getwordskfrom(TOKENS, startspan, endspan, k);
    for (Constituent token : wordsthreeafter) {
        // Should only be one POS tag for each token
        List<String> POS_tag = POS.getLabelsCoveringSpan(token.getStartSpan(), token.getEndSpan());
        if (POS_tag.size() != 1) {
            logger.info("Error token has more than one POS tag.");
        }
        tags[i] = POS_tag.get(0);
        i++;
    }
    Set<Feature> __result = new LinkedHashSet<Feature>();
    String __id;
    String __value;
    int contextmax = 3;
    for (int j = 0; j < contextmax; j++) {
        for (i = 0; i < tags.length; i++) {
            StringBuffer f = new StringBuffer();
            for (int context = 0; context <= j && i + context < tags.length; context++) {
                if (context != 0) {
                    f.append("_");
                }
                f.append(tags[i + context]);
            }
            __id = "" + (i + "_" + j);
            __value = "" + (f.toString());
            __result.add(new DiscreteFeature(classifier + ":" + __id + "(" + __value + ")"));
        }
    }
    return __result;
}

Example 9 with DiscreteFeature

use of edu.illinois.cs.cogcomp.edison.features.DiscreteFeature in project cogcomp-nlp by CogComp.

the class LabeledDepFeatureGenerator method POSwinConj.

private Set<Feature> POSwinConj(int head, int dep, DepInst sent, String deprel) {
    String header = "POSwin: ";
    String headleft = "Left: " + (head > 0 ? sent.strPos[head - 1] : "null") + " ";
    String headcenter = "Center: " + sent.strPos[head] + " ";
    String headright = "Right: " + (head + 1 < sent.strPos.length ? sent.strPos[head + 1] : "null") + " ";
    String depleft = "Left: " + (dep > 0 ? sent.strPos[dep - 1] : "null") + " ";
    String depcenter = "Center: " + sent.strPos[dep] + " ";
    String depright = "Right: " + (dep + 1 < sent.strPos.length ? sent.strPos[dep + 1] : "null") + " ";
    String arcdir = "Arc-dir: " + (head < dep) + " ";
    String arclength = "Arc-length: " + (head - dep) + " ";
    Set<Feature> feats = new HashSet<>();
    feats.add(new DiscreteFeature(header + headcenter + depcenter + arcdir + deprel));
    feats.add(new DiscreteFeature(header + headcenter + depcenter + arclength + deprel));
    feats.add(new DiscreteFeature(header + headleft + headcenter + headright + depleft + depcenter + depright + arcdir + deprel));
    return feats;
}

Example 10 with DiscreteFeature

use of edu.illinois.cs.cogcomp.edison.features.DiscreteFeature in project cogcomp-nlp by CogComp.

the class LabeledDepFeatureGenerator method PrefixConj.

private Set<Feature> PrefixConj(int head, int dep, DepInst sent, String deprel) {
    String header = "Prefix: ";
    String prefixhead = sent.strLemmas[head].substring(0, Math.min(sent.strLemmas[head].length(), 5)) + " ";
    String prefixdep = sent.strLemmas[dep].substring(0, Math.min(sent.strLemmas[dep].length(), 5)) + " ";
    String poshead = sent.strPos[head] + " ";
    String posdep = sent.strPos[dep] + " ";
    String arcdir = "Arc-dir: " + (head < dep) + " ";
    Set<Feature> feats = new HashSet<>();
    feats.add(new DiscreteFeature(header + prefixhead + posdep + arcdir + deprel));
    feats.add(new DiscreteFeature(header + poshead + prefixdep + arcdir + deprel));
    feats.add(new DiscreteFeature(header + prefixhead + prefixdep + arcdir + deprel));
    return feats;
}