rich search workon

[wolnelektury.git] / apps / search / index.py

diff --git a/apps/search/index.py b/apps/search/index.py
index 318e540..42a271e 100644 (file)
--- a/apps/search/index.py
+++ b/apps/search/index.py
@@ -1,14 +1,22 @@
-
+# -*- coding: utf-8 -*-

 from django.conf import settings
 from django.conf import settings

-from lucene import SimpleFSDirectory, IndexWriter, File, Field, NumericField, \
-    Version, Document, JavaError, IndexSearcher, QueryParser, Term, PerFieldAnalyzerWrapper, \
-    SimpleAnalyzer, PolishAnalyzer, ArrayList, KeywordAnalyzer, NumericRangeQuery
+from lucene import SimpleFSDirectory, IndexWriter, File, Field, \
+    NumericField, Version, Document, JavaError, IndexSearcher, \
+    QueryParser, Term, PerFieldAnalyzerWrapper, \
+    SimpleAnalyzer, PolishAnalyzer, ArrayList, \
+    KeywordAnalyzer, NumericRangeQuery, BooleanQuery, \
+    BlockJoinQuery, BlockJoinCollector, TermsFilter, \
+    HashSet, BooleanClause, Term, CharTermAttribute, \
+    PhraseQuery, StringReader, TermQuery

     # KeywordAnalyzer
     # KeywordAnalyzer

+import sys

 import os
 import errno
 from librarian import dcparser
 from librarian.parser import WLDocument
 import os
 import errno
 from librarian import dcparser
 from librarian.parser import WLDocument

-from catalogue.models import Book
+import catalogue.models
+from multiprocessing.pool import ThreadPool
+import atexit

 
 
 class WLAnalyzer(PerFieldAnalyzerWrapper):
 
 
 class WLAnalyzer(PerFieldAnalyzerWrapper):


@@ -28,6 +36,7 @@ class WLAnalyzer(PerFieldAnalyzerWrapper):
         self.addAnalyzer("source_name", simple)
         self.addAnalyzer("publisher", simple)
         self.addAnalyzer("author", simple)
         self.addAnalyzer("source_name", simple)
         self.addAnalyzer("publisher", simple)
         self.addAnalyzer("author", simple)

+        self.addAnalyzer("is_book", keyword)

 
         #self.addanalyzer("fragment_anchor", keyword)
 
 
         #self.addanalyzer("fragment_anchor", keyword)
 


@@ -57,7 +66,8 @@ class Index(IndexStore):
     def open(self, analyzer=None):
         if self.index:
             raise Exception("Index is already opened")
     def open(self, analyzer=None):
         if self.index:
             raise Exception("Index is already opened")

-        self.index = IndexWriter(self.store, self.analyzer, IndexWriter.MaxFieldLength.LIMITED)
+        self.index = IndexWriter(self.store, self.analyzer,\
+                                 IndexWriter.MaxFieldLength.LIMITED)

         return self.index
 
     def close(self):
         return self.index
 
     def close(self):


@@ -68,29 +78,20 @@ class Index(IndexStore):
     def remove_book(self, book):
         q = NumericRangeQuery.newIntRange("book_id", book.id, book.id, True,True)
         self.index.deleteDocuments(q)
     def remove_book(self, book):
         q = NumericRangeQuery.newIntRange("book_id", book.id, book.id, True,True)
         self.index.deleteDocuments(q)

-        
+

     def index_book(self, book, overwrite=True):
         if overwrite:
             self.remove_book(book)
     def index_book(self, book, overwrite=True):
         if overwrite:
             self.remove_book(book)

+            

 
         doc = self.extract_metadata(book)
         parts = self.extract_content(book)
         block = ArrayList().of_(Document)
 
         doc = self.extract_metadata(book)
         parts = self.extract_content(book)
         block = ArrayList().of_(Document)

-        
-        try:
-            self.index.addDocument(doc)
-            for p in parts:
-                self.index.addDocument(p)                
-        except JavaError as e:
-            import nose.tools; nose.tools.set_trace()
-
-            #block.add(p)
-            #self.index.addDocuments(block)
-            
-            #        import nose.tools; nose.tools.set_trace()
-            #block.add(doc)

 
 

-        #        self.index.addDocuments(block)
+        for p in parts:
+            block.add(p)
+        block.add(doc)
+        self.index.addDocuments(block)

 
     master_tags = [
         'opowiadanie',
 
     master_tags = [
         'opowiadanie',


@@ -101,6 +102,8 @@ class Index(IndexStore):
         'wywiad'
         ]
 
         'wywiad'
         ]
 

+    skip_header_tags = ['autor_utworu', 'nazwa_utworu']
+

     def create_book_doc(self, book):
         """
         Create a lucene document connected to the book
     def create_book_doc(self, book):
         """
         Create a lucene document connected to the book


@@ -117,6 +120,7 @@ class Index(IndexStore):
         doc = self.create_book_doc(book)
         doc.add(Field("slug", book.slug, Field.Store.NO, Field.Index.ANALYZED_NO_NORMS))
         doc.add(Field("tags", ','.join([t.name for t in book.tags]), Field.Store.NO, Field.Index.ANALYZED))
         doc = self.create_book_doc(book)
         doc.add(Field("slug", book.slug, Field.Store.NO, Field.Index.ANALYZED_NO_NORMS))
         doc.add(Field("tags", ','.join([t.name for t in book.tags]), Field.Store.NO, Field.Index.ANALYZED))

+        doc.add(Field("is_book", 'true', Field.Store.NO, Field.Index.NOT_ANALYZED))

 
         # validator, name
         for field in dcparser.BookInfo.FIELDS:
 
         # validator, name
         for field in dcparser.BookInfo.FIELDS:


@@ -150,6 +154,7 @@ class Index(IndexStore):
             if master.tag in self.master_tags:
                 return master
 
             if master.tag in self.master_tags:
                 return master
 

+    

     def extract_content(self, book):
         wld = WLDocument.from_file(book.xml_file.path)
         root = wld.edoc.getroot()
     def extract_content(self, book):
         wld = WLDocument.from_file(book.xml_file.path)
         root = wld.edoc.getroot()


@@ -159,11 +164,16 @@ class Index(IndexStore):
         # header_index - the 0-indexed position of header element.
         # content
         master = self.get_master(root)
         # header_index - the 0-indexed position of header element.
         # content
         master = self.get_master(root)

+        if master is None:
+            return []
+        

         header_docs = []
         for header, position in zip(list(master), range(len(master))):
         header_docs = []
         for header, position in zip(list(master), range(len(master))):

-            print("header %s @%d" % (header, position))
+            if header.tag in self.skip_header_tags:
+                continue

             doc = self.create_book_doc(book)
             doc.add(NumericField("header_index", Field.Store.YES, True).setIntValue(position))
             doc = self.create_book_doc(book)
             doc.add(NumericField("header_index", Field.Store.YES, True).setIntValue(position))

+            doc.add(Field("header_type", header.tag, Field.Store.YES, Field.Index.NOT_ANALYZED))

             content = u' '.join([t for t in header.itertext()])
             doc.add(Field("content", content, Field.Store.NO, Field.Index.ANALYZED))
             header_docs.append(doc)
             content = u' '.join([t for t in header.itertext()])
             doc.add(Field("content", content, Field.Store.NO, Field.Index.ANALYZED))
             header_docs.append(doc)


@@ -183,7 +193,6 @@ class Index(IndexStore):
         # will contain (framgent id -> { content: [], themes: [] }
         fragments = {}
         for start, end in walker(master):
         # will contain (framgent id -> { content: [], themes: [] }
         fragments = {}
         for start, end in walker(master):

-            print("%s %s" % (start, end))

             if start is not None and start.tag == 'begin':
                 fid = start.attrib['id'][1:]
                 fragments[fid] = {'content': [], 'themes': []}
             if start is not None and start.tag == 'begin':
                 fid = start.attrib['id'][1:]
                 fragments[fid] = {'content': [], 'themes': []}


@@ -194,14 +203,26 @@ class Index(IndexStore):
                 fragments[fid]['content'].append(start.tail)
             elif start is not None and start.tag == 'end':
                 fid = start.attrib['id'][1:]
                 fragments[fid]['content'].append(start.tail)
             elif start is not None and start.tag == 'end':
                 fid = start.attrib['id'][1:]

+                if fid not in fragments:
+                    continue  # a broken <end> node, skip it

                 frag = fragments[fid]
                 del fragments[fid]
                 frag = fragments[fid]
                 del fragments[fid]

-                print("Fragment %s complete, themes: %s contents: %s" % (fid, frag['themes'], frag['content']))
+
+                def jstr(l):
+                    return u' '.join(map(
+                        lambda x: x == None and u'(none)' or unicode(x),
+                        l))

 
                 doc = self.create_book_doc(book)
 
                 doc = self.create_book_doc(book)

-                doc.add(Field("fragment_anchor", fid, Field.Store.YES, Field.Index.NOT_ANALYZED))
-                doc.add(Field("content", u' '.join(filter(lambda s: s is not None, frag['content'])), Field.Store.NO, Field.Index.ANALYZED))
-                doc.add(Field("themes", u' '.join(frag['themes']), Field.Store.NO, Field.Index.ANALYZED))
+                doc.add(Field("fragment_anchor", fid,
+                              Field.Store.YES, Field.Index.NOT_ANALYZED))
+                doc.add(Field("content",
+                              u' '.join(filter(lambda s: s is not None, frag['content'])),
+                              Field.Store.NO, Field.Index.ANALYZED))
+                doc.add(Field("themes",
+                              u' '.join(filter(lambda s: s is not None, frag['themes'])),
+                              Field.Store.NO, Field.Index.ANALYZED))
+

                 fragment_docs.append(doc)
             elif start is not None:
                 for frag in fragments.values():
                 fragment_docs.append(doc)
             elif start is not None:
                 for frag in fragments.values():


@@ -220,18 +241,86 @@ class Index(IndexStore):
         self.close()
 
 
         self.close()
 
 

+class ReusableIndex(Index):
+    """
+    Works like index, but does not close/optimize Lucene index
+    until program exit (uses atexit hook).
+    This is usefull for importbooks command.
+
+    if you cannot rely on atexit, use ReusableIndex.close_reusable() yourself.
+    """
+    index = None
+    pool = None
+    pool_jobs = None
+
+    def open(self, analyzer=None, threads=4):
+        if ReusableIndex.index is not None:
+            self.index = ReusableIndex.index
+        else:
+            ReusableIndex.pool = ThreadPool(threads)
+            ReusableIndex.pool_jobs = []
+            Index.open(self, analyzer)
+            ReusableIndex.index = self.index
+            atexit.register(ReusableIndex.close_reusable)
+
+    def index_book(self, *args, **kw):
+        job = ReusableIndex.pool.apply_async(Index.index_book, (self,)+ args, kw)
+        ReusableIndex.pool_jobs.append(job)
+
+    @staticmethod
+    def close_reusable():
+        if ReusableIndex.index is not None:
+            for job in ReusableIndex.pool_jobs:
+                job.wait()
+            ReusableIndex.pool.close()
+
+            ReusableIndex.index.optimize()
+            ReusableIndex.index.close()
+            ReusableIndex.index = None
+
+    def close(self):
+        pass
+
+

 class Search(IndexStore):
     def __init__(self, default_field="content"):
         IndexStore.__init__(self)
         self.analyzer = PolishAnalyzer(Version.LUCENE_34)
         ## self.analyzer = WLAnalyzer()
         self.searcher = IndexSearcher(self.store, True)
 class Search(IndexStore):
     def __init__(self, default_field="content"):
         IndexStore.__init__(self)
         self.analyzer = PolishAnalyzer(Version.LUCENE_34)
         ## self.analyzer = WLAnalyzer()
         self.searcher = IndexSearcher(self.store, True)

-        self.parser = QueryParser(Version.LUCENE_34, default_field, self.analyzer)
+        self.parser = QueryParser(Version.LUCENE_34, default_field,
+                                  self.analyzer)
+
+        self.parent_filter = TermsFilter()
+        self.parent_filter.addTerm(Term("is_book", "true"))

 
     def query(self, query):
         return self.parser.parse(query)
 
 
     def query(self, query):
         return self.parser.parse(query)
 

-    def search(self, query, max_results=50):
+    def wrapjoins(self, query, fields=[]):
+        """
+        This functions modifies the query in a recursive way,
+        so Term and Phrase Queries contained, which match
+        provided fields are wrapped in a BlockJoinQuery,
+        and so delegated to children documents.
+        """
+        if BooleanQuery.instance_(query):
+            qs = BooleanQuery.cast_(query)
+            for clause in qs:
+                clause = BooleanClause.cast_(clause)
+                clause.setQuery(self.wrapjoins(clause.getQuery(), fields))
+            return qs
+        else:
+            termset = HashSet()
+            query.extractTerms(termset)
+            for t in termset:
+                t = Term.cast_(t)
+                if t.field() not in fields:
+                    return query
+            return BlockJoinQuery(query, self.parent_filter,
+                                  BlockJoinQuery.ScoreMode.Total)
+
+    def simple_search(self, query, max_results=50):

         """Returns (books, total_hits)
         """
 
         """Returns (books, total_hits)
         """
 


@@ -239,6 +328,128 @@ class Search(IndexStore):
         bks = []
         for found in tops.scoreDocs:
             doc = self.searcher.doc(found.doc)
         bks = []
         for found in tops.scoreDocs:
             doc = self.searcher.doc(found.doc)

-            bks.append(Book.objects.get(id=doc.get("book_id")))
+            bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))
+        return (bks, tops.totalHits)
+
+    def search(self, query, max_results=50):
+        query = self.query(query)
+        query = self.wrapjoins(query, ["content", "themes"])
+
+        tops = self.searcher.search(query, max_results)
+        bks = []
+        for found in tops.scoreDocs:
+            doc = self.searcher.doc(found.doc)
+            bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))
+        return (bks, tops.totalHits)
+
+    def bsearch(self, query, max_results=50):
+        q = self.query(query)
+        bjq = BlockJoinQuery(q, self.parent_filter, BlockJoinQuery.ScoreMode.Avg)
+
+        tops = self.searcher.search(bjq, max_results)
+        bks = []
+        for found in tops.scoreDocs:
+            doc = self.searcher.doc(found.doc)
+            bks.append(catalogue.models.Book.objects.get(id=doc.get("book_id")))

         return (bks, tops.totalHits)
 
         return (bks, tops.totalHits)
 

+# TokenStream tokenStream = analyzer.tokenStream(fieldName, reader);
+# OffsetAttribute offsetAttribute = tokenStream.getAttribute(OffsetAttribute.class);
+# CharTermAttribute charTermAttribute = tokenStream.getAttribute(CharTermAttribute.class);
+
+# while (tokenStream.incrementToken()) {
+#     int startOffset = offsetAttribute.startOffset();
+#     int endOffset = offsetAttribute.endOffset();
+#     String term = charTermAttribute.toString();
+# }
+
+
+class MultiSearch(Search):
+    """Class capable of IMDb-like searching"""
+    def get_tokens(self, queryreader):
+        if isinstance(queryreader, str):
+            queryreader = StringReader(queryreader)
+        queryreader.reset()
+        tokens = self.analyzer.reusableTokenStream('content', queryreader)
+        toks = []
+        while tokens.incrementToken():
+            cta = tokens.getAttribute(CharTermAttribute.class_)
+            toks.append(cta.toString())
+        return toks
+
+    def make_phrase(self, tokens, field='content', joined=False, slop=2):
+        phrase = PhraseQuery()
+        phrase.setSlop(slop)
+        for t in tokens:
+            term = Term(field, t)
+            phrase.add(term)
+        if joined:
+            phrase = self.content_query(phrase)
+        return phrase
+
+    def make_term_query(self, tokens, field='content', modal=BooleanClause.Occur.SHOULD, joined=False):
+        q = BooleanQuery()
+        for t in tokens:
+            term = Term(field, t)
+            q.add(BooleanClause(TermQuery(term), modal))
+        if joined:
+            self.content_query(q)
+        return q
+
+    def content_query(self, query):
+        return BlockJoinQuery(query, self.parent_filter,
+                              BlockJoinQuery.ScoreMode.Total)
+
+    def multisearch(self, query, max_results=50):
+        """
+        Search strategy:
+        - (phrase) OR -> content
+                      -> title
+                      -> author
+        - (keywords)  -> author
+                      -> motyw
+                      -> tags
+                      -> content
+        """
+        queryreader = StringReader(query)
+        tokens = self.get_tokens(queryreader)
+
+        top_level = BooleanQuery()
+        Should = BooleanClause.Occur.SHOULD
+
+        phrase_level = BooleanQuery()
+        phrase_level.setBoost(1.3)
+
+        p_content = self.make_phrase(tokens, joined=True)
+        p_title = self.make_phrase(tokens, 'title')
+        p_author = self.make_phrase(tokens, 'author')
+
+        phrase_level.add(BooleanClause(p_content, Should))
+        phrase_level.add(BooleanClause(p_title, Should))
+        phrase_level.add(BooleanClause(p_author, Should))
+
+        kw_level = BooleanQuery()
+
+        kw_level.add(self.make_term_query(tokens, 'author'), Should)
+        kw_level.add(self.make_term_query(tokens, 'themes', joined=True), Should)
+        kw_level.add(self.make_term_query(tokens, 'tags'), Should)
+        kw_level.add(self.make_term_query(tokens, joined=True), Should)
+
+        top_level.add(BooleanClause(phrase_level, Should))
+        top_level.add(BooleanClause(kw_level, Should))
+
+        print self.do_search(phrase_level)
+        print self.do_search(kw_level)
+        print self.do_search(top_level)
+
+    def do_search(self, query, max_results=50):
+        tops = self.searcher.search(query, max_results)
+        #tops = self.searcher.search(p_content, max_results)
+
+        bks = []
+        for found in tops.scoreDocs:
+            doc = self.searcher.doc(found.doc)
+            b = catalogue.models.Book.objects.get(id=doc.get("book_id"))
+            bks.append(b)
+            print "%s (%d) -> %f" % (b, b.id, found.score)
+        return (bks, tops.totalHits)