source: gs3-extensions/mars-src/trunk/src/java/org/greenstone/gsdl3/util/WekaDBWrapper.java@ 36863

/*
 *    WekaDBWrapper.java
 *    Copyright (C) 2011 New Zealand Digital Library, http://www.nzdl.org
 *
 *    This program is free software; you can redistribute it and/or modify
 *    it under the terms of the GNU General Public License as published by
 *    the Free Software Foundation; either version 2 of the License, or
 *    (at your option) any later version.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU General Public License for more details.
 *
 *    You should have received a copy of the GNU General Public License
 *    along with this program; if not, write to the Free Software
 *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
package org.greenstone.gsdl3.util;

import java.io.*;
import java.util.Vector;
import java.util.Collections;
import java.util.regex.Pattern;
import java.util.regex.Matcher;

import weka.core.Attribute;
import weka.core.Instance;
import weka.core.Instances;

import org.apache.log4j.*;

import org.greenstone.gsdl3.util.WekaFindInstanceKNN;

/** Java wrapper class for access to the Weka 
 *  Devised (in the first instance) to operate as: java -jar weka.jar <arg1> <arg2> 
 *
 * Inspired by MGSearchWrapper.java
 */

public class WekaDBWrapper
{
    public final static double AV_SEGMENT_LENGTH_SECS = 6.0;
    
    /** the query result, filled in by runQuery */
    protected Vector query_result_;

    protected int offset_ = 100;
    protected int length_ = 20; // **** Unused
    
    // Approximate matching not yet utilized
    protected double radius_;   // **** Unused

    protected int max_docs_;

    protected double arousal_;
    protected double valence_;
    
    static Logger logger = Logger.getLogger (org.greenstone.gsdl3.util.WekaDBWrapper.class.getName ());

    public WekaDBWrapper() {  
    query_result_ = null;
    }

    // query param methods

    /** start point (offset) into the array of feature vectors for a track 
    - 100 by default which equals 10 seconds (assuming 0.1 frame size) */
    public void setOffset(int offset) {
    offset_ = offset;
    }
    
    /** the number of consecutive frames used in match
    - 20 by default which equals 2 seconds (assuming 0.1 frame size) */
    public void setLength(int length) {
    length_ = length;
    }

    /** distance used in approximate matching support - default is 50 */
    public void setRadius(double radius) {
    radius_ = radius;
    }
    
    public void setMaxDocs(int max_docs) {
    max_docs_ = max_docs;
    }

    public void setArousal(double arousal) {
    arousal_ = arousal;
    }
    public void setValence(double valence) {
    valence_ = valence;
    }
    
    /** returns a string with all the current query param settings */
    // the following was in MG version, do we need this in WekaDB version? // ****
    //public String getQueryParams() {}


    protected boolean addQueryResult(boolean first_entry, String doc_id, 
                     Vector<Double> rankVector, Vector<Integer> offsetVector)
    {

    if (first_entry) {
        WekaDBDocInfo wekaDB_doc_info = new WekaDBDocInfo(doc_id,rankVector,offsetVector);
        query_result_.add(wekaDB_doc_info);
        first_entry = false;
    }
    else {
        double rank = rankVector.get(0);
        int offset = offsetVector.get(0);
        WekaDBDocInfo wekaDB_doc_info = new WekaDBDocInfo(doc_id,rank,offset);
        
        query_result_.add(wekaDB_doc_info);
    }

    return first_entry;
    }


    /** actually carry out the query.
    Use the set methods to set query results.
    Writes the result to query_result.
     * - maintains state between requests as can be slow  
     * base_dir and index_path should join together to provide
     * the absolute location of the mg index files eg ..../index/dtx/demo 
     * base_dir must end with a file separator (OS dependant)
     */


    public void runQuery(String wekaDB_index_dir, String knn_model_file, 
             String assoc_index_dir, String query_string) {

    // combine index_dir with audiodb fileanem

    String full_knn_model_filename  = wekaDB_index_dir + File.separatorChar + knn_model_file;

    //String full_chr12_filename = assoc_index_dir + File.separatorChar 
    //    + query_string + File.separatorChar + "doc.chr12";

    System.err.println("**** full knn model filename  = " + full_knn_model_filename);

    // Example returned result from Weka KNN
    // => first line is the input instance ('filename+segment',Arousal,Valence)
    //    following (indented lines) nearest neighbour matches in same format
    //
    // ds_22716_5743-6,-0.549489,-0.118439
    //  ds_22716_5743-6,-0.549489,-0.118439
    //  ds_31008_6550-30,-0.549489,-0.118439
    //  ds_72651_26831-6,-0.549489,-0.118439
    //  ds_26196_9214-18,-0.549489,-0.118439


    WekaFindInstanceKNN.init(full_knn_model_filename);

    String doc_id  = query_string;
    int    segment = offset_;

    String query_doc_id_segment =  doc_id + "-" + segment;

    double query_arousal_val = arousal_;
    double query_valence_val =  valence_;

    int k_nearest_num = max_docs_;
    int expanded_k_nearest_num = max_docs_ * 5; // * internally get more matches, then sift through to arrive at the best 'max_docs_'
    
    Pattern doc_seg_re = Pattern.compile("^(\\w+)-(\\d+)$");
    //Matcher query_doc_seg_match = doc_seq_re.matcher(query_doc_id_segment);
    
    Instances nearest_instances
        = WekaFindInstanceKNN.kNearestNeighbours(query_doc_id_segment,query_arousal_val,query_valence_val,k_nearest_num);

        
    Vector expanded_query_result = new Vector();

    int nearest_instances_len = nearest_instances.size();
    
    int clamped_expanded_k_nearest_num = Math.min(expanded_k_nearest_num,nearest_instances_len);

    double pos_penalty = 0.1;
    
    for (int ei=0; ei<clamped_expanded_k_nearest_num; ei++) {
        Instance instance = nearest_instances.instance(ei);
        logger.info("\tProcessing returned instance: " + instance);
        
        String matching_doc_id_segment = instance.stringValue(0);

        
        //Pattern p = Pattern.compile("^(\\w+)-(\\d+)$");
        Matcher m = doc_seg_re.matcher(matching_doc_id_segment);
        if (m.matches()) {

        String matching_doc_id = m.group(1);
        int end_of_matching_segment_offset = Integer.parseInt(m.group(2));
        int matching_segment_offset = end_of_matching_segment_offset - (int)AV_SEGMENT_LENGTH_SECS;
        
        if (matching_doc_id.equals(doc_id)) {
            continue;
        }
        
        double matching_arousal_val = instance.value(1);
        double matching_valence_val = instance.value(2);
        
        double matching_diff = (Math.abs(query_arousal_val - matching_arousal_val)
                    + Math.abs(query_valence_val - matching_valence_val))/4.0;
        double matching_rank = 1.0 - matching_diff - (pos_penalty * (double)ei);

        logger.info("\tAdding in: matching_doc_id = " + matching_doc_id);
        WekaDBDocInfo wekaDB_doc_info = new WekaDBDocInfo(matching_doc_id,matching_rank,matching_segment_offset);
        expanded_query_result.add(wekaDB_doc_info);
        }
        else {
        logger.error("Returned AV k-nearest neighbour match '"+matching_doc_id_segment+"' could not be parsed as <doc-id>-<segment>" );
        }          
    }
    
    query_result_ = new Vector();

    int i = 0;
    while (i < k_nearest_num) {
        if (i >= expanded_query_result.size()) {
        break;
        }
        
        query_result_.add(expanded_query_result.get(i));
        i++;
    }

    Collections.sort(query_result_);    
    }
    
    public void runQueryOLD(String wekaDB_index_dir, String knn_model_file, 
             String assoc_index_dir, String query_string) {

    // combine index_dir with audiodb fileanem

    String full_knn_model_filename  = wekaDB_index_dir + File.separatorChar + knn_model_file;
    String full_chr12_filename = assoc_index_dir + File.separatorChar 
        + query_string + File.separatorChar + "doc.chr12";

    int num_matches_within_track = 6;

    // ****
    String [] cmd_array = new String[] {
        "java", "-jar", "weka.jar",
        "-d", full_knn_model_filename,
        "-Q", "nsequence",
        "-p", String.format("%d",offset_),
        "-n", String.format("%d",num_matches_within_track),
        "-l", String.format("%d",length_),
        "-r", String.format("%d",max_docs_),
        "-f", full_chr12_filename
    };

    System.err.println("**** cmd_array = " + String.join(" ", cmd_array));

    Runtime runtime = Runtime.getRuntime();
    try {
        Process wekaDB_proc = runtime.exec(cmd_array);
        //int exitVal = wekaDB_proc.waitFor();
        //System.err.println("*** exit status = " + exitVal);

        InputStream wis = wekaDB_proc.getInputStream();
        InputStreamReader wisr = new InputStreamReader(wis);
        BufferedReader wbr = new BufferedReader(wisr);

        query_result_ = new Vector();

        boolean first_entry = true;
        int line_count = 0;

        String root_doc_id = null;
        Vector<Double> rankVector = new Vector<Double>();
        Vector<Integer> offsetVector = new Vector<Integer>();

        // Example output
        //   D8 0.00105175
        //   1.69786e-16 392 392
        //   0.00113568 392 673
        //   0.00127239 392 910
        //   0.00139736 392 481
        //   0.00145331 392 303
        //   D2 0.00429758
        //   0.00403335 392 865
        //   0.00411288 392 458
        //   0.00442461 392 866
        //   0.00444272 392 864
        //   0.00447434 392 424
        // ...

        String line;
        while ((line = wbr.readLine()) != null) {
        String[] tokens = line.split("\\s+");
        line_count++;

        if (tokens.length==2) {
            // processing a top-level doc line

            if (line_count>1) {
            // struck new top-level entry => store vector vals for previous block

            first_entry = addQueryResult(first_entry,root_doc_id,rankVector,offsetVector);
            // and now reset vectors to empty to be ready for next chain of values
            rankVector = new Vector<Double>();
            offsetVector = new Vector<Integer>();
            }

            root_doc_id = tokens[0];
        }
        else {
            // should be 3 items 
            double euclidean_dist = Double.parseDouble(tokens[0]);
            int src_frame = Integer.parseInt(tokens[1]);
            int target_frame = Integer.parseInt(tokens[2]);

            // ****
            
            // enforce 1.0 as upper limit due to rounding errors 
            // in audioDB distance calculations
            double rank = Math.min(1.0 - euclidean_dist,1.0); 

            if ((line_count==2) && (src_frame==target_frame)) {
            // Found match with self
            continue;
            }

            rankVector.add(rank);
            offsetVector.add(target_frame);
        }

        }

        addQueryResult(first_entry,root_doc_id,rankVector,offsetVector);

        wbr.close();

        // sort query_result_ on 'rank' field 
        // note: compareTo() method impelemented to sort into descending order

        Collections.sort(query_result_);


    }
    catch (Exception e) {
        logger.error("Failed to execute the following command: " +  String.join(" ", cmd_array));
        e.printStackTrace();
    }

    }


    /** get the result out of the wrapper */
    public Vector getQueryResult()
    {
    return query_result_;
    }
}