def reindex(self, dirty=None):
# Determine indexable attributes and their weights.
indexable = {}
for base in reversed(type.mro(type(self))):
_ = getattr(base, '_indexable', {})
if isinstance(_, dict):
indexable.update(_)
elif isinstance(_, list):
indexable.update(dict.fromkeys(_, 1.0))
# Determine if we actually need to re-index or not.
if dirty is not None and not set(indexable.keys()).intersection(set(dirty)):
return
index = DocumentIndex.objects(doc_id=self.id).first()
if index: index.delete()
# Determine the number of occurrences of each term with a per-attribute weight.
occurrences = defaultdict(float)
for attr, weight in indexable.iteritems():
value = getattr(self, attr)
if isinstance(value, basestring):
for word in lexer.strip(value):
occurrences[word] += weight
elif isinstance(value, (tuple, list, set)):
for word in lexer.strip(u' '.join(value).encode('utf8')):
occurrences[word] += weight
# Save the index and terms.
index = DocumentIndex(doc_id=str(self.id), length=len(occurrences), terms=occurrences)
index.save(safe=False)
def delete(self, safe=False):
# Delete index data, if any.
index = DocumentIndex.objects(id=self.id).first()
if index: index.delete()
# Depth-first cascading delete.
for i in self.children:
i.delete()
# Remove reference to self from parent asset.
if self.parent:
self.parent.children.remove(self)
self.parent.save()
# Actually delete this asset.
return super(Asset, self).delete(safe=safe)
def results(self, query=None):
if query is None: query = self.query
if query is None: return []
terms = keywords(' '.join(strip(query.lower())))
terms = (set(terms[0] + terms[1]), set(terms[2]))
query = dict()
aquery = dict()
for term in list(terms[0]):
if ':' in term:
terms[0].remove(term)
l, _, r = term.partition(':')
if l == 'tag':
aquery.setdefault('tags', list()).append(r)
elif l == 'kind':
aquery.setdefault('__raw__', dict())['_cls'] = {
'$regex' : r,
'$options': 'i'
}
if not terms[0] and not terms[1]:
def gen():
for record in Asset.objects(**aquery).only('title', 'description', 'path', 'acl').order_by('created'):
yield 1.0, record
return gen()
for term in terms[0]:
query['terms__%s__exists' % (term, )] = True
for term in terms[1]:
query['terms__%s__exists' % (term, )] = False
# Calculate the inverse document frequency for each term
idfs = {}
num_docs = DocumentIndex.objects.count()
for term in terms[0]:
term_docs = DocumentIndex.objects(terms__term=term).count()
idfs[term] = log_((num_docs - term_docs + 0.5) / (term_docs + 0.5))
# Get the average document length.
avg_doc_length = sum([i.length for i in DocumentIndex.objects.only('length')])/float(num_docs)
k = 2.0
b = 0.75
f = []
results = []
def compute(idfs, idx, k, b, f):
score = 0.0
for term, q in idfs.iteritems():
dividend = idx.terms[term] * (k + 1.0)
relDocSize = idx.length / avg_doc_length
divisor = q + ( 1.0 - b + b * relDocSize ) * k
termScore = (dividend / divisor) * q
score += termScore
return (score, idx.doc_id)
with futures.ThreadPoolExecutor(max_workers=5) as executor:
for idx in DocumentIndex.objects(**query):
f.append(executor.submit(compute, idfs, idx, k, b, f))
for result in futures.as_completed(f):
score, doc_id = result.result()
results.append((score, doc_id))
def iterresults():
for score, id_ in results:
yield score, Asset.objects(id=id_, **aquery).only('title', 'description', 'path', 'acl').first()
return sorted(iterresults(), lambda a, b: cmp(a[0], b[0]), reverse=True)
