]> git.openstreetmap.org Git - nominatim.git/commitdiff
add PHP part for new ICU-base tokenizer
authorSarah Hoffmann <lonvia@denofr.de>
Sun, 2 May 2021 19:21:41 +0000 (21:21 +0200)
committerSarah Hoffmann <lonvia@denofr.de>
Wed, 5 May 2021 08:15:27 +0000 (10:15 +0200)
lib-php/tokenizer/legacy_icu_tokenizer.php [new file with mode: 0644]
nominatim/tokenizer/legacy_icu_tokenizer.py

diff --git a/lib-php/tokenizer/legacy_icu_tokenizer.php b/lib-php/tokenizer/legacy_icu_tokenizer.php
new file mode 100644 (file)
index 0000000..09cfe70
--- /dev/null
@@ -0,0 +1,238 @@
+<?php
+
+namespace Nominatim;
+
+class Tokenizer
+{
+    private $oDB;
+
+    private $oNormalizer;
+    private $oTransliterator;
+    private $aCountryRestriction;
+
+    public function __construct(&$oDB)
+    {
+        $this->oDB =& $oDB;
+        $this->oNormalizer = \Transliterator::createFromRules(CONST_Term_Normalization_Rules);
+        $this->oTransliterator = \Transliterator::createFromRules(CONST_Transliteration);
+    }
+
+    public function checkStatus()
+    {
+        $sSQL = "SELECT word_id FROM word WHERE word_token IN (' a')";
+        $iWordID = $this->oDB->getOne($sSQL);
+        if ($iWordID === false) {
+            throw new Exception('Query failed', 703);
+        }
+        if (!$iWordID) {
+            throw new Exception('No value', 704);
+        }
+    }
+
+
+    public function setCountryRestriction($aCountries)
+    {
+        $this->aCountryRestriction = $aCountries;
+    }
+
+
+    public function normalizeString($sTerm)
+    {
+        if ($this->oNormalizer === null) {
+            return $sTerm;
+        }
+
+        return $this->oNormalizer->transliterate($sTerm);
+    }
+
+    private function makeStandardWord($sTerm)
+    {
+        $sNorm = ' '.$this->oTransliterator->transliterate($sTerm).' ';
+
+        return trim(str_replace(CONST_Abbreviations[0], CONST_Abbreviations[1], $sNorm));
+    }
+
+
+    public function tokensForSpecialTerm($sTerm)
+    {
+        $aResults = array();
+
+        $sSQL = 'SELECT word_id, class, type FROM word ';
+        $sSQL .= '   WHERE word_token = \' \' || :term';
+        $sSQL .= '   AND class is not null AND class not in (\'place\')';
+
+        Debug::printVar('Term', $sTerm);
+        Debug::printSQL($sSQL);
+        $aSearchWords = $this->oDB->getAll($sSQL, array(':term' => $this->makeStandardWord($sTerm)));
+
+        Debug::printVar('Results', $aSearchWords);
+
+        foreach ($aSearchWords as $aSearchTerm) {
+            $aResults[] = new \Nominatim\Token\SpecialTerm(
+                $aSearchTerm['word_id'],
+                $aSearchTerm['class'],
+                $aSearchTerm['type'],
+                \Nominatim\Operator::TYPE
+            );
+        }
+
+        Debug::printVar('Special term tokens', $aResults);
+
+        return $aResults;
+    }
+
+
+    public function extractTokensFromPhrases(&$aPhrases)
+    {
+        $sNormQuery = '';
+        $aWordLists = array();
+        $aTokens = array();
+        foreach ($aPhrases as $iPhrase => $oPhrase) {
+            $sNormQuery .= ','.$this->normalizeString($oPhrase->getPhrase());
+            $sPhrase = $this->makeStandardWord($oPhrase->getPhrase());
+            if (strlen($sPhrase) > 0) {
+                $aWords = explode(' ', $sPhrase);
+                Tokenizer::addTokens($aTokens, $aWords);
+                $aWordLists[] = $aWords;
+            } else {
+                $aWordLists[] = array();
+            }
+        }
+
+        Debug::printVar('Tokens', $aTokens);
+        Debug::printVar('WordLists', $aWordLists);
+
+        $oValidTokens = $this->computeValidTokens($aTokens, $sNormQuery);
+
+        foreach ($aPhrases as $iPhrase => $oPhrase) {
+            $oPhrase->computeWordSets($aWordLists[$iPhrase], $oValidTokens);
+        }
+
+        return $oValidTokens;
+    }
+
+
+    private function computeValidTokens($aTokens, $sNormQuery)
+    {
+        $oValidTokens = new TokenList();
+
+        if (!empty($aTokens)) {
+            $this->addTokensFromDB($oValidTokens, $aTokens, $sNormQuery);
+
+            // Try more interpretations for Tokens that could not be matched.
+            foreach ($aTokens as $sToken) {
+                if ($sToken[0] == ' ' && !$oValidTokens->contains($sToken)) {
+                    if (preg_match('/^ ([0-9]{5}) [0-9]{4}$/', $sToken, $aData)) {
+                        // US ZIP+4 codes - merge in the 5-digit ZIP code
+                        $oValidTokens->addToken(
+                            $sToken,
+                            new Token\Postcode(null, $aData[1], 'us')
+                        );
+                    } elseif (preg_match('/^ [0-9]+$/', $sToken)) {
+                        // Unknown single word token with a number.
+                        // Assume it is a house number.
+                        $oValidTokens->addToken(
+                            $sToken,
+                            new Token\HouseNumber(null, trim($sToken))
+                        );
+                    }
+                }
+            }
+        }
+
+        return $oValidTokens;
+    }
+
+
+    private function addTokensFromDB(&$oValidTokens, $aTokens, $sNormQuery)
+    {
+        // Check which tokens we have, get the ID numbers
+        $sSQL = 'SELECT word_id, word_token, word, class, type, country_code,';
+        $sSQL .= ' operator, coalesce(search_name_count, 0) as count';
+        $sSQL .= ' FROM word WHERE word_token in (';
+        $sSQL .= join(',', $this->oDB->getDBQuotedList($aTokens)).')';
+
+        Debug::printSQL($sSQL);
+
+        $aDBWords = $this->oDB->getAll($sSQL, null, 'Could not get word tokens.');
+
+        foreach ($aDBWords as $aWord) {
+            $oToken = null;
+            $iId = (int) $aWord['word_id'];
+
+            if ($aWord['class']) {
+                // Special terms need to appear in their normalized form.
+                // (postcodes are not normalized in the word table)
+                $sNormWord = $this->normalizeString($aWord['word']);
+                if ($aWord['word'] && strpos($sNormQuery, $sNormWord) === false) {
+                    continue;
+                }
+
+                if ($aWord['class'] == 'place' && $aWord['type'] == 'house') {
+                    $oToken = new Token\HouseNumber($iId, trim($aWord['word_token']));
+                } elseif ($aWord['class'] == 'place' && $aWord['type'] == 'postcode') {
+                    if ($aWord['word']
+                        && pg_escape_string($aWord['word']) == $aWord['word']
+                    ) {
+                        $oToken = new Token\Postcode(
+                            $iId,
+                            $aWord['word'],
+                            $aWord['country_code']
+                        );
+                    }
+                } else {
+                    // near and in operator the same at the moment
+                    $oToken = new Token\SpecialTerm(
+                        $iId,
+                        $aWord['class'],
+                        $aWord['type'],
+                        $aWord['operator'] ? Operator::NEAR : Operator::NONE
+                    );
+                }
+            } elseif ($aWord['country_code']) {
+                // Filter country tokens that do not match restricted countries.
+                if (!$this->aCountryRestriction
+                    || in_array($aWord['country_code'], $this->aCountryRestriction)
+                ) {
+                    $oToken = new Token\Country($iId, $aWord['country_code']);
+                }
+            } else {
+                $oToken = new Token\Word(
+                    $iId,
+                    $aWord['word_token'][0] != ' ',
+                    (int) $aWord['count'],
+                    substr_count($aWord['word_token'], ' ')
+                );
+            }
+
+            if ($oToken) {
+                $oValidTokens->addToken($aWord['word_token'], $oToken);
+            }
+        }
+    }
+
+
+    /**
+     * Add the tokens from this phrase to the given list of tokens.
+     *
+     * @param string[] $aTokens List of tokens to append.
+     *
+     * @return void
+     */
+    private static function addTokens(&$aTokens, $aWords)
+    {
+        $iNumWords = count($aWords);
+
+        for ($i = 0; $i < $iNumWords; $i++) {
+            $sPhrase = $aWords[$i];
+            $aTokens[' '.$sPhrase] = ' '.$sPhrase;
+            $aTokens[$sPhrase] = $sPhrase;
+
+            for ($j = $i + 1; $j < $iNumWords; $j++) {
+                $sPhrase .= ' '.$aWords[$j];
+                $aTokens[' '.$sPhrase] = ' '.$sPhrase;
+                $aTokens[$sPhrase] = $sPhrase;
+            }
+        }
+    }
+}
index 09d4059e907ea39b3ecfbbec52e8661f26879fce..064b395c6a8a88d7d1f22230fbd463852241f3b9 100644 (file)
@@ -133,15 +133,18 @@ class LegacyICUTokenizer:
     def _install_php(self, config):
         """ Install the php script for the tokenizer.
         """
+        abbr_inverse = list(zip(*self.abbreviations))
         php_file = self.data_dir / "tokenizer.php"
         php_file.write_text(dedent("""\
             <?php
             @define('CONST_Max_Word_Frequency', {1.MAX_WORD_FREQUENCY});
             @define('CONST_Term_Normalization_Rules', "{0.normalization}");
-            @define('CONST_Transliteration'. "{0.transliteration}");
-            # XXX abreviations
+            @define('CONST_Transliteration', "{0.transliteration}");
+            @define('CONST_Abbreviations', array(array('{2}'), array('{3}')));
             require_once('{1.lib_dir.php}/tokenizer/legacy_icu_tokenizer.php');
-            """.format(self, config)))
+            """.format(self, config,
+                       "','".join(abbr_inverse[0]),
+                       "','".join(abbr_inverse[1]))))
 
 
     def _save_config(self, config):