001    package org.maltparser.ml.lib;
002    
003    import java.io.Serializable;
004    
005    import org.maltparser.core.helper.HashMap;
006    
007    /**
008     * The purpose of the feature map is to map MaltParser's column based features together with the symbol code from the symbol table to 
009     * unique indices suitable for liblinear and libsvm.  A feature column position are combined together with the symbol code in a 
010     * 64-bit key (Long), where 16 bits are reserved for the position and 48 bits are reserved for the symbol code.  
011     * 
012     * @author Johan Hall
013     *
014     */
015    public class FeatureMap  implements Serializable {
016            private static final long serialVersionUID = 7526471155622776147L;
017            private HashMap<Long,Integer> map;
018            private int featureCounter;
019            
020            /**
021             * Creates a feature map and sets the feature counter to 1
022             */
023            public FeatureMap() {
024                    map = new HashMap<Long, Integer>();
025                    this.featureCounter = 1;
026            }
027            
028            /**
029             * Adds a mapping from a combination of the position in the column-based feature vector and the symbol code to 
030             * an index value suitable for liblinear and libsvm.
031             * 
032             * @param featurePosition a position in the column-based feature vector
033             * @param code a symbol code
034             * @return the index value 
035             */
036            public int addIndex(int featurePosition, int code) {
037                    long key = ((((long)featurePosition) << 48) | (long)code);
038                    Integer index = map.get(key);
039                    if (index == null) {
040                            index = featureCounter++;
041                            map.put(key, index);
042                    }
043                    return index.intValue();
044            }
045            
046            /**
047             * Return
048             * 
049             * @param featurePosition the position in the column-based feature vector
050             * @param code the symbol code suitable for liblinear and libsvm
051             * @return the index value if it exists, otherwise -1
052             */
053            public int getIndex(int featurePosition, int code) {
054                    long key = ((((long)featurePosition) << 48) | (long)code);
055                    Integer index = map.get(key);
056                    return (index == null)?-1:index;
057            }
058            
059            
060            public int addIndex(int featurePosition, int code1, int code2) {
061                    long key = ((((long)featurePosition) << 48) | (((long)code1) << 24) | (long)code2);
062                    Integer index = map.get(key);
063                    if (index == null) {
064                            index = featureCounter++;
065                            map.put(key, index);
066                    }
067                    return index.intValue();
068            }
069            
070            public int getIndex(int featurePosition, int code1, int code2) {
071                    long key = ((((long)featurePosition) << 48) | (((long)code1) << 24) | (long)code2);
072                    Integer index = map.get(key);
073                    return (index == null)?-1:index;
074            }
075                    
076            /**
077             * @return the size of the map
078             */
079            public int size() {
080                    return map.size();
081            }
082            
083            /**
084             * @return the current value of the feature counter.
085             */
086            public int getFeatureCounter() {
087                    return featureCounter;
088            }
089    }