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

Example 1 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature 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;
}

Example 2 with Feature

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

the class LabeledDepFeatureGenerator method ChunkConj.

private Set<Feature> ChunkConj(int head, int dep, DepInst sent, String deprel) {
    String header = "POSChunk: ";
    String chunkhead = sent.strChunk[head] + " ";
    String poshead = sent.strPos[head] + " ";
    String chunkdep = sent.strChunk[dep] + " ";
    String posdep = sent.strPos[dep] + " ";
    String arcdir = "Arc-dir: " + (head < dep) + " ";
    String arclength = "Arc-length " + (head - dep) + " ";
    Set<Feature> feats = new HashSet<>();
    feats.add(new DiscreteFeature(header + chunkhead + chunkdep + arcdir + deprel));
    feats.add(new DiscreteFeature(header + chunkhead + chunkdep + arclength + deprel));
    feats.add(new DiscreteFeature(header + chunkhead + poshead + chunkdep + posdep + arcdir + deprel));
    feats.add(new DiscreteFeature(header + chunkhead + poshead + chunkdep + posdep + arclength + deprel));
    return feats;
}

Example 3 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature 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 4 with Feature

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

the class POSWindowTwo method getFeatures.

@Override
public /**
     * This feature extractor assumes that the TOKEN View, POS View have been
     * generated in the Constituents TextAnnotation. It will use its own POS tag of the
     * two context words before and after the constituent.
     *
     **/
Set<Feature> getFeatures(Constituent c) throws EdisonException {
    TextAnnotation ta = c.getTextAnnotation();
    View TOKENS = null, POS = null;
    try {
        TOKENS = ta.getView(ViewNames.TOKENS);
        POS = ta.getView(ViewNames.POS);
    } catch (Exception e) {
        e.printStackTrace();
    }
    // 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();
    // All our constituents are words(tokens)
    // words two before & after
    int k = 2;
    String[] tags = getwindowtagskfrom(TOKENS, POS, startspan, endspan, k);
    String classifier = "POSWindowTwo";
    String __id, __value;
    Set<Feature> __result = new LinkedHashSet<Feature>();
    for (int i = 0; i < tags.length; i++) {
        if (tags[i] == null) {
            continue;
        } else {
            __id = classifier + ":" + i;
            __value = "(" + tags[i] + ")";
            logger.info(__id + __value);
            __result.add(new DiscreteFeature(__id + __value));
        }
    }
    return __result;
}

Example 5 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature 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;
}