activity.c

使用具有增量学习的监控式学习方法。包括几个不同的分类算法。

/**
 * @file
 * Activity scores for classes.
 *
 * This is not a real classifier. It is supposed to be used with a combiner.
 * It ignores the message content.
 * Folders which have not had any mails in a long time will be
 * considered less probable.
 * Keep average (or max) time between two messages for a class.
 * If the time since the last message is significantly higher than this,
 * call the class inactive.
 * Also frequency might be considered in the activity score.
 *
 * @author  Mikael Ylikoski
 * @date    2002
 */

#include <stdlib.h>
#include <string.h>
#include "activity.h"
#include "doc_classifier.h"
#include "utility.h"


/**
 * Class activity.
 */
typedef struct {
    int nom;		/**< Number of messages */
    //float average;	/**< Average delta */
    int max;		/**< Maximum delta */
    int last;		/**< Last time */
} activity;


/**
 * Activity database.
 */
struct activity_db_ {
    int time;		/**< Current time */
    activity *act;	/**< Activities */
    int noa;		/**< Number of activities */
};


/**
 * Create a new activity database.
 *
 * @param size  the initial number of classes
 * @return  The new activity database.
 */
activity_db *
activity_new_db (int size) {
    activity_db *db;

    db = my_malloc (sizeof(activity_db));
    db->act = my_calloc (size, sizeof(activity));
    db->time = 0;
    db->noa = size;
    return db;
}


/**
 * Register a new message.
 *
 * @param db     activity database to use
 * @param class  class of message
 * @return 0 if ok; != 0 otherwise.
 */
int
activity_new_msg (activity_db *db, int class) {
    activity *a;

    if (class >= db->noa) {
	a = my_realloc (db->act, (class + 1) * sizeof(activity));
	db->act = a;
	memset (&db->act[db->noa], 0,
		(class + 1 - db->noa) * sizeof(activity));
	db->noa = class + 1;
    }

    db->time++;
    db->act[class].nom++;
    if (db->time > 0 && db->act[class].last > 0)
	if (db->time - db->act[class].last > db->act[class].max)
	    db->act[class].max = db->time - db->act[class].last;
    db->act[class].last = db->time;
    return 0;
}


/**
 * Get activity score for a class.
 *
 * @param db     activity database to use
 * @param class  class to get score for
 * @return  The activity score.
 */
double
activity_score (activity_db *db, int class) {
    if (db->act[class].nom < 3)
	return 1;
    if (db->time - db->act[class].last > 5 * db->act[class].max)
	return 0;
    return 1;
}


/**
 * Get activity scores for all classes.
 *
 * @param db  activity database to use
 * @return  An array of activity scores.
 */
double *
activity_score_list (activity_db *db) {
    double *d;
    int i;

    d = my_malloc (db->noa * sizeof(double));
    for (i = 0; i < db->noa; i++)
	d[i] = activity_score (db, i);
    return d;
}


void *
activity_new_doc (const char *opts) {
    return activity_new_db (0);
}


int
activity_learn_doc (void *db, void *data, int class) {
    activity_db *adb;

    adb = (activity_db *)db;
    return activity_new_msg (adb, class);
}


double *
activity_classify_doc_score (void *db, void *data) {
    activity_db *adb;

    adb = (activity_db *)db;
    return activity_score_list (db);
}


/**
 * Keep cygwin happy.
 */
int
main (void) {
    return 0;
}


//const char *my_doc_classifier_name = "Activity";


const doc_classifier_functions my_functions = {
    .new = activity_new_doc,
    .learn = activity_learn_doc,
    .classify_score = activity_classify_doc_score
};