diff options
| author | Peng Wu <alexepico@gmail.com> | 2011-11-24 13:02:10 +0800 |
|---|---|---|
| committer | Peng Wu <alexepico@gmail.com> | 2011-11-24 13:02:10 +0800 |
| commit | 3d731193a89fa725eec0bfce92489df0be7a272c (patch) | |
| tree | 40bb8ac2bf7afad6243d6a6f3d04127de39d920a | |
| parent | 2acda18725e41a3634e2226ef5a20a7599f5cff5 (diff) | |
| download | libpinyin-3d731193a89fa725eec0bfce92489df0be7a272c.tar.gz libpinyin-3d731193a89fa725eec0bfce92489df0be7a272c.tar.xz libpinyin-3d731193a89fa725eec0bfce92489df0be7a272c.zip | |
increase train_factor because of larger model data
| -rw-r--r-- | src/lookup/pinyin_lookup.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/lookup/pinyin_lookup.cpp b/src/lookup/pinyin_lookup.cpp index 99e2a20..2a4d191 100644 --- a/src/lookup/pinyin_lookup.cpp +++ b/src/lookup/pinyin_lookup.cpp @@ -448,7 +448,7 @@ bool PinyinLookup::train_result(PinyinKeyVector keys, CandidateConstraints const //TODO: verify the new training method. phrase_token_t last_token = sentence_start; // constraints->len + 1 == results->len - guint32 train_factor = 23; + guint32 train_factor = 23 * 5; for ( size_t i = 0; i < constraints->len; ++i){ phrase_token_t * token = &g_array_index(results, phrase_token_t, i); if ( *token == null_token ) @@ -465,7 +465,7 @@ bool PinyinLookup::train_result(PinyinKeyVector keys, CandidateConstraints const //printf("i:%d\tlast_token:%d\ttoken:%d\n", i, last_token, *token); m_phrase_index->get_phrase_item(*token, m_cache_phrase_item); m_cache_phrase_item.increase_pinyin_possibility(*m_custom, pinyin_keys + i, train_factor); - m_phrase_index->add_unigram_frequency(*token, train_factor); + m_phrase_index->add_unigram_frequency(*token, train_factor * 10); if ( last_token ){ SingleGram * system, *user; m_system_bigram->load(last_token, system); |