m_infap.c

/* 
   Copyright (c) 2008 - Chris Buckley. 

   Permission is granted for use and modification of this file for
   research, non-commercial purposes. 
*/

#include "common.h"
#include "sysfunc.h"
#include "trec_eval.h"
#include "functions.h"
#include "trec_format.h"

static int
te_calc_infap(const EPI * epi, const REL_INFO * rel_info,
              const RESULTS * results, const TREC_MEAS * tm, TREC_EVAL * eval);

/* See trec_eval.h for definition of TREC_MEAS */
TREC_MEAS te_meas_infAP = { "infAP",
    "    Inferred AP\n\
    A measure that allows sampling of judgement pool: Qrels/results divided\n\
    into unpooled, pooled_but_unjudged, pooled_judged_rel,pooled_judged_nonrel.\n\
    My intuition of infAP:\n\
    Assume a judgment pool with a random subset that has been judged.\n\
    Calculate P at rel doc using only the judged higher retrieved docs,\n\
    then average in 0's from higher docs that were not in the judgment pool.\n\
    (Those in the pool but not judged are ignored, since they are assumed\n\
    to be relevant in the same proportion as those judged.)\n\
    Cite:    'Estimating Average Precision with Incomplete and Imperfect\n\
    Judgments', Emine Yilmaz and Javed A. Aslam. CIKM \n",
    te_init_meas_s_double,
    te_calc_infap,
    te_acc_meas_s,
    te_calc_avg_meas_s,
    te_print_single_meas_s_double,
    te_print_final_meas_s_double,
    NULL, -1
};

static int
te_calc_infap(const EPI * epi, const REL_INFO * rel_info,
              const RESULTS * results, const TREC_MEAS * tm, TREC_EVAL * eval)
{
    RES_RELS res_rels;
    long j;
    long nonrel_so_far, rel_so_far, pool_unjudged_so_far;
    double inf_ap = 0.0;

    if (UNDEF == te_form_res_rels(epi, rel_info, results, &res_rels))
        return (UNDEF);

    nonrel_so_far = 0;
    rel_so_far = 0;
    pool_unjudged_so_far = 0;
    for (j = 0; j < res_rels.num_ret; j++) {
        if (res_rels.results_rel_list[j] == RELVALUE_NONPOOL)
            /* document not in pool. Skip */
            continue;
        if (res_rels.results_rel_list[j] == RELVALUE_UNJUDGED) {
            /* document in pool but unjudged. */
            pool_unjudged_so_far++;
            continue;
        }

        if (res_rels.results_rel_list[j] >= 0 &&
            res_rels.results_rel_list[j] < epi->relevance_level) {
            nonrel_so_far++;
        } else {
            /* Judged Rel doc */
            rel_so_far++;
            /* Special case nonrel_so_far == 0 to avoid division by 0 */
            /* inf_ap */
            if (0 == j)
                inf_ap += 1.0;
            else {
                double fj = (double) j;
                inf_ap += 1.0 / (fj + 1.0) +
                    (fj / (fj + 1.0)) *
                    ((rel_so_far - 1 + nonrel_so_far +
                      pool_unjudged_so_far) / fj) * ((rel_so_far - 1 +
                                                      INFAP_EPSILON) /
                                                     (rel_so_far - 1 +
                                                      nonrel_so_far +
                                                      2 * INFAP_EPSILON));
            }
        }
    }
    if (res_rels.num_rel) {
        inf_ap /= res_rels.num_rel;
    }
    eval->values[tm->eval_index].value = inf_ap;

    return (1);
}