Take OSB feature multiplier into account.

diff --git a/src/libstat/classifiers/bayes.c b/src/libstat/classifiers/bayes.c
index acd04f602aa854e3f975c01489d059ae536ea30c..78d112dd0ef140d1d0ab874f4bcf7f57e199eab6 100644 (file)
--- a/src/libstat/classifiers/bayes.c
+++ b/src/libstat/classifiers/bayes.c
@@ -79,6 +79,9 @@ inv_chi_square (gdouble value, gint freedom_deg)
        return MIN (1.0, sum);
 }
 
+static const double feature_weight[] = { 0, 3125, 256, 27, 4, 1 };
+
+#define PROB_COMBINE(prob, cnt, weight, assumed) (((weight) * (assumed) + (cnt) * (prob)) / ((weight) + (cnt)))
 /*
  * In this callback we calculate local probabilities for tokens
  */
@@ -90,7 +93,8 @@ bayes_classify_callback (gpointer key, gpointer value, gpointer data)
        guint i;
        struct rspamd_token_result *res;
        guint64 spam_count = 0, ham_count = 0, total_count = 0;
-       double spam_prob, spam_freq, ham_freq, bayes_spam_prob;
+       double spam_prob, spam_freq, ham_freq, bayes_spam_prob, bayes_ham_prob,
+               ham_prob, fw, w;
 
        for (i = rt->start_pos; i < rt->end_pos; i++) {
                res = &g_array_index (node->results, struct rspamd_token_result, i);
@@ -112,9 +116,13 @@ bayes_classify_callback (gpointer key, gpointer value, gpointer data)
                spam_freq = ((double)spam_count / MAX (1., (double)rt->total_spam));
                ham_freq = ((double)ham_count / MAX (1., (double)rt->total_ham));
                spam_prob = spam_freq / (spam_freq + ham_freq);
-               bayes_spam_prob = (0.5 + spam_prob * total_count) / (1. + total_count);
+               ham_prob = ham_freq / (spam_freq + ham_freq);
+               fw = feature_weight[node->window_idx % G_N_ELEMENTS (feature_weight)];
+               w = (fw * total_count) / (4.0 * (1.0 + fw * total_count));
+               bayes_spam_prob = PROB_COMBINE (spam_prob, total_count, w, 0.5);
+               bayes_ham_prob = PROB_COMBINE (ham_prob, total_count, w, 0.5);
                rt->spam_prob += log (bayes_spam_prob);
-               rt->ham_prob += log (1. - bayes_spam_prob);
+               rt->ham_prob += log (bayes_ham_prob);
                res->cl_runtime->processed_tokens ++;
        }