1. Matt Chaput
  2. whoosh


whoosh / docs / source / searching.rst

How to search

Once you've created an index and added documents to it, you can search for those documents.

The Searcher object

To get a :class:`whoosh.searching.Searcher` object, call searcher() on your Index object:

searcher = myindex.searcher()

You'll usually want to open the searcher using a with statement so the searcher is automatically closed when you're done with it (searcher objects represent a number of open files, so if you don't explicitly close them and the system is slow to collect them, you can run out of file handles):

with ix.searcher() as searcher:

This is of course equivalent to:

    searcher = ix.searcher()

The Searcher object is the main high-level interface for reading the index. It has lots of useful methods for getting information about the index, such as lexicon(fieldname).

>>> list(searcher.lexicon("content")) [u"document", u"index", u"whoosh"]

However, the most important method on the Searcher object is :meth:`~whoosh.searching.Searcher.search`, which takes a :class:`whoosh.query.Query` object and returns a :class:`~whoosh.searching.Results` object:

from whoosh.qparser import QueryParser

qp = QueryParser("content", schema=myindex.schema)
q = qp.parse(u"hello world")

with myindex.searcher() as s:
    results = s.search(q)

By default the results contains at most the first 10 matching documents. To get more results, use the limit keyword:

results = s.search(q, limit=20)

If you want all results, use limit=None. However, setting the limit whenever possible makes searches faster because Whoosh doesn't need to examine and score every document.

Since displaying a page of results at a time is a common pattern, the search_page method lets you conveniently retrieve only the results on a given page:

results = s.search_page(q, 1)

The default page length is 10 hits. You can use the pagelen keyword argument to set a different page length:

results = s.search_page(q, 5, pagelen=20)

Results object

The :class:`~whoosh.searching.Results` object acts like a list of the matched documents. You can use it to access the stored fields of each hit document, to display to the user.

>>> # Show the best hit's stored fields >>> results[0] {"title": u"Hello World
in Python", "path": u"/a/b/c"} >>> results[0:2] [{"title": u"Hello World in
Python", "path": u"/a/b/c"}, {"title": u"Foo", "path": u"/bar"}]

By default, Searcher.search(myquery) limits the number of hits to 20, So the number of scored hits in the Results object may be less than the number of matching documents in the index.

>>> # How many documents in the entire index would have matched? >>>
len(results) 27 >>> # How many scored and sorted documents in this Results
object? >>> # This will often be less than len() if the number of hits was
limited >>> # (the default). >>> results.scored_length() 10

Calling len(Results) runs a fast (unscored) version of the query again to figure out the total number of matching documents. This is usually very fast but for large indexes it can cause a noticeable delay. If you want to avoid this delay on very large indexes, you can use the :meth:`~whoosh.searching.Results.has_exact_length`, :meth:`~whoosh.searching.Results.estimated_length`, and :meth:`~whoosh.searching.Results.estimated_min_length` methods to estimate the number of matching documents without calling len():

found = results.scored_length()
if results.has_exact_length():
    print("Scored", found, "of exactly", len(results), "documents")
    low = results.estimated_min_length()
    high = results.estimated_length()

    print("Scored", found, "of between", low, "and", "high", "documents")

Scoring and sorting


Normally the list of result documents is sorted by score. The :mod:`whoosh.scoring` module contains implementations of various scoring algorithms. The default is :class:`~whoosh.scoring.BM25F`.

You can set the scoring object to use when you create the searcher using the weighting keyword argument:

with myindex.searcher(weighting=whoosh.scoring.Cosine()) as s:

A scoring object is an object with a :meth:`~whoosh.scoring.Weighting.score` method that takes information about the term to score and returns a score as a floating point number.


See :doc:`facets`.

Highlighting snippets and More Like This

See :doc:`highlight` and :doc:`keywords` for information on these topics.

Filtering results

You can use the filter keyword argument to search() to specify a set of documents to permit in the results. The argument can be a :class:`whoosh.query.Query` object, a :class:`whoosh.searching.Results` object, or a set-like object containing document numbers. The searcher caches filters so if for example you use the same query filter with a searcher multiple times, the additional searches will be faster because the searcher will cache the results of running the filter query

You can also specify a mask keyword argument to specify a set of documents that are not permitted in the results:

with myindex.searcher() as s:
    qp = qparser.QueryParser("content", myindex.schema)
    user_q = qp.parse(query_string)

    # Only show documents in the "rendering" chapter
    allow_q = query.Term("chapter", "rendering")
    # Don't show any documents where the "tag" field contains "todo"
    restrict_q = query.Term("tag", "todo")

    results = s.search(user_q, filter=allow_q, mask=restrict_q)

(If you specify both a filter and a mask, and a matching document appears in both, the mask "wins" and the document is not permitted.)

To find out how many results were filtered out of the results, use results.filtered_count (or resultspage.results.filtered_count):

with myindex.searcher() as s:
    qp = qparser.QueryParser("content", myindex.schema)
    user_q = qp.parse(query_string)

    # Filter documents older than 7 days
    old_q = query.DateRange("created", None, datetime.now() - timedelta(days=7))
    results = s.search(user_q, mask=old_q)

    print("Filtered out %d older documents" % results.filtered_count)

Which terms from my query matched?

You can use the terms=True keyword argument to search() to have the search record which terms in the query matched which documents:

with myindex.searcher() as s:
    results = s.seach(myquery, terms=True)

You can then get information about which terms matched from the :class:`whoosh.searching.Results` and :class:`whoosh.searching.Hit` objects:

# Was this results object created with terms=True?
if results.has_matched_terms():
    # What terms matched in the results?

    # What terms matched in each hit?
    for hit in results:

Collapsing results

Whoosh lets you eliminate all but the top N documents with the same facet key from the results. This can be useful in a few situations:

  • Eliminating duplicates at search time.
  • Restricting the number of matches per source. For example, in a web search application, you might want to show at most three matches from any website.

Whether a document should be collapsed is determined by the value of a "collapse facet". If a document has an empty collapse key, it will never be collapsed, but otherwise only the top N documents with the same collapse key will appear in the results.

See :doc:`/facets` for information on facets:

with myindex.searcher() as s:
    # Set the facet to collapse on and the maximum number of documents per
    # facet value (default is 1)
    results = s.collector(collapse="hostname", collapse_limit=3)

    # Dictionary mapping collapse keys to the number of documents that
    # were filtered out by collapsing on that key

Collapsing works with both scored and sorted results. You can use any of the facet types available in the :mod:`whoosh.sorting` module.

By default, Whoosh uses the results order (score or sort key) to determine the documents to collapse. For example, in scored results, the best scoring documents would be kept. You can optionally specify a collapse_order facet to control which documents to keep when collapsing.

For example, in a product search you could display results sorted by decreasing price, and eliminate all but the highest rated item of each product type:

from whoosh import sorting

with myindex.searcher() as s:
    price_facet = sorting.FieldFacet("price", reverse=True)
    type_facet = sorting.FieldFacet("type")
    rating_facet = sorting.FieldFacet("rating", reverse=True)

    results = s.collector(sortedby=price_facet,  # Sort by reverse price
                          collapse=type_facet,  # Collapse on product type
                          collapse_order=rating_facet  # Collapse to highest rated

The collapsing happens during the search, so it is usually more efficient than finding everything and post-processing the results. However, if the collapsing eliminates a large number of documents, collapsed search can take longer because the search has to consider more documents and remove many already-collected documents.

Since this collector must sometimes go back and remove already-collected documents, if you use it in combination with :class:`~whoosh.collectors.TermsCollector` and/or :class:`~whoosh.collectors.FacetCollector`, those collectors may contain information about documents that were filtered out of the final results by collapsing.

Time limited searches

To limit the amount of time a search can take:

from whoosh.collectors import TimeLimitCollector, TimeLimit

with myindex.searcher() as s:
    # Get a collector object
    c = s.collector(limit=None, sortedby="title_exact")
    # Wrap it in a TimeLimitedCollector and set the time limit to 10 seconds
    tlc = TimeLimitedCollector(c, timelimit=10.0)

    # Try searching
        s.search_with_collector(myquery, tlc)
    except TimeLimit:
        print("Search took too long, aborting!")

    # You can still get partial results from the collector
    results = tlc.results()

Convenience methods

The :meth:`~whoosh.searching.Searcher.document` and :meth:`~whoosh.searching.Searcher.documents` methods on the Searcher object let you retrieve the stored fields of documents matching terms you pass in keyword arguments.

This is especially useful for fields such as dates/times, identifiers, paths, and so on.

>>> list(searcher.documents(indexeddate=u"20051225")) [{"title": u"Christmas
presents"}, {"title": u"Turkey dinner report"}] >>> print
searcher.document(path=u"/a/b/c") {"title": "Document C"}

These methods have some limitations:

  • The results are not scored.
  • Multiple keywords are always AND-ed together.
  • The entire value of each keyword argument is considered a single term; you can't search for multiple terms in the same field.

Combining Results objects

It is sometimes useful to use the results of another query to influence the order of a :class:`whoosh.searching.Results` object.

For example, you might have a "best bet" field. This field contains hand-picked keywords for documents. When the user searches for those keywords, you want those documents to be placed at the top of the results list. You could try to do this by boosting the "bestbet" field tremendously, but that can have unpredictable effects on scoring. It's much easier to simply run the query twice and combine the results:

# Parse the user query
userquery = queryparser.parse(querystring)

# Get the terms searched for
termset = set()

# Formulate a "best bet" query for the terms the user
# searched for in the "content" field
bbq = Or([Term("bestbet", text) for fieldname, text
          in termset if fieldname == "content"])

# Find documents matching the searched for terms
results = s.search(bbq, limit=5)

# Find documents that match the original query
allresults = s.search(userquery, limit=10)

# Add the user query results on to the end of the "best bet"
# results. If documents appear in both result sets, push them
# to the top of the combined results.

The Results object supports the following methods:

Adds the documents in 'results' on to the end of the list of result documents.
Removes the documents in 'results' from the list of result documents.
Any result documents that also appear in 'results' are moved to the top of the list of result documents.
Any result documents that also appear in 'results' are moved to the top of the list of result documents. Then any other documents in 'results' are added on to the list of result documents.