/**
* Perform an intelligent detection based on clusterLanguages()
*
* WARNING: this method is EXPERIMENTAL. It is not recommended for common
* use, and it may disappear or its functionality may change in future
* releases without notice.
*
* This compares the sample text to top the top level of clusters. If the
* sample is similar to the cluster it will drop down and compare it to the
* languages in the cluster, and so on until it hits a leaf node.
*
* this should find the language in considerably fewer compares
* (the equivalent of a binary search), however clusterLanguages() is costly
* and the loss of accuracy from this technique is significant.
*
* This method may need to be 'fuzzier' in order to become more accurate.
*
* This function could be more useful if the universe of possible languages
* was very large, however in such cases some method of Bayesian inference
* might be more helpful.
*
* @param string $str input string
*
* @return array language scores (only those compared)
* @throws Text_LanguageDetect_Exception
* @see clusterLanguages()
*/
public function clusteredSearch($str)
{
// input check
if (!Text_LanguageDetect_Parser::validateString($str)) {
return array();
}
// clusterLanguages() will return a cached result if possible
// so it's safe to call it every time
$result = $this->clusterLanguages();
$dendogram_start = $result['open_forks'];
$dendogram_data = $result['fork_data'];
$dendogram_alias = $result['name_map'];
$sample_obj = new Text_LanguageDetect_Parser($str);
$sample_obj->prepareTrigram();
$sample_obj->setPadStart(!$this->_perl_compatible);
$sample_obj->analyze();
$sample_result = $sample_obj->getTrigramRanks();
$sample_count = count($sample_result);
// input check
if ($sample_count == 0) {
return array();
}
$i = 0;
// counts the number of steps
foreach ($dendogram_start as $lang) {
if (isset($dendogram_alias[$lang])) {
$lang_key = $dendogram_alias[$lang];
} else {
$lang_key = $lang;
}
$scores[$lang] = $this->_normalize_score($this->_distance($this->_lang_db[$lang_key], $sample_result), $sample_count);
$i++;
}
if ($this->_perl_compatible) {
asort($scores);
} else {
arsort($scores);
}
$top_score = current($scores);
$top_key = key($scores);
// of starting forks, $top_key is the most similar to the sample
$cur_key = $top_key;
while (isset($dendogram_data[$cur_key])) {
$lang1 = $dendogram_data[$cur_key]['bestfit'];
$lang2 = $dendogram_data[$cur_key]['otherfit'];
foreach (array($lang1, $lang2) as $lang) {
if (isset($dendogram_alias[$lang])) {
$lang_key = $dendogram_alias[$lang];
} else {
$lang_key = $lang;
}
$scores[$lang] = $this->_normalize_score($this->_distance($this->_lang_db[$lang_key], $sample_result), $sample_count);
//todo: does not need to do same comparison again
}
$i++;
if ($scores[$lang1] > $scores[$lang2]) {
$cur_key = $lang1;
$loser_key = $lang2;
} else {
$cur_key = $lang2;
$loser_key = $lang1;
}
$diff = $scores[$cur_key] - $scores[$loser_key];
// $cur_key ({$dendogram_alias[$cur_key]}) wins
// over $loser_key ({$dendogram_alias[$loser_key]})
// with a difference of $diff
}
// found result in $i compares
// rather than sorting the result, preserve it so that you can see
// which paths the algorithm decided to take along the tree
// but sometimes the last item is only the second highest
if ($this->_perl_compatible && end($scores) > prev($scores) || !$this->_perl_compatible && end($scores) < prev($scores)) {
$real_last_score = current($scores);
$real_last_key = key($scores);
// swaps the 2nd-to-last item for the last item
unset($scores[$real_last_key]);
$scores[$real_last_key] = $real_last_score;
}
if (!$this->_perl_compatible) {
$scores = array_reverse($scores, true);
// second param requires php > 4.0.3
}
return $scores;
}
