def process(inStream, outStream,
fields={"id": "id", "text": "text"},
limits={"clusters": 10, "top_documents": 10}):
all = {}
text_field = fields["text"]
key_field = fields["id"]
max_clusters = limits["clusters"]
max_top_docs = limits["top_documents"]
c = Corpus()
for line in inStream:
data = line.split('\t', 1)[1]
doc = json.loads(data.decode("utf8"))
key = doc[key_field]
all[key] = doc
text = c.add((key, doc[text_field]), key=key)
clusters = c.cluster()
results = []
for c in clusters[:max_clusters]:
tophits = [c.primary]
tophits += [hit["object"] for hit in c.similars[:max_top_docs-1]]
topdocs = []
for (key, text) in tophits:
topdocs.append(all[key])
results.append({"top_documents": topdocs})
json.dump({"clusters": results}, outStream)
def test_cluster():
c = Corpus(similarity=0.1)
for doc in docs:
c.add(doc)
groups = c.cluster()
eq_(len(groups), 2)
eq_(len(groups[0].similars), 1)
eq_(len(groups[1].similars), 1)
def test_cluster():
c = Corpus(similarity=0.1)
for doc in docs:
c.add(doc)
groups = c.cluster()
eq_(len(groups), 2)
eq_(len(groups[0].similars), 1)
eq_(len(groups[1].similars), 1)
def cluster_by_platform(qs, feeling, version):
qs = qs.filter(version=version)
# We need to create corpii for each platform and manually inspect each
# opinion and put it in the right platform bucket.
seen = {}
for platform in OS_USAGE:
c = Corpus(similarity=SIM_THRESHOLD, stopwords=STOPWORDS)
seen = {}
for op in qs.filter(os=platform.short):
if op.description in seen:
continue
# filter short descriptions
if len(op.description) < 15:
continue
seen[op.description] = 1
c.add(op, str=op.description, key=op.id)
result = c.cluster()
if result:
cluster_type, created = ClusterType.objects.get_or_create(
feeling=feeling,
version=version,
platform=platform.short,
frequency='weekly')
# Remove the old cluster_groups
Cluster.objects.filter(type=cluster_type).delete()
# Store the clusters into groups
for group in result:
cluster = Cluster(type=cluster_type)
cluster.num_opinions = len(group.similars) + 1
cluster.pivot = group.primary
cluster.save()
for s in group.similars:
ClusterItem(
cluster=cluster,
opinion=s['object'],
score=s['similarity']).save()
def cluster_queryset(qs):
seen = {}
c = Corpus(similarity=SIM_THRESHOLD, stopwords=STOPWORDS)
for op in qs:
if op.description in seen:
continue
# filter short descriptions
if len(op.description) < 15:
continue
seen[op.description] = 1
c.add(op, str=op.description, key=op.id)
return c.cluster()
def cluster_queryset(qs):
seen = {}
c = Corpus(similarity=SIM_THRESHOLD, stopwords=STOPWORDS)
for op in qs:
if op.description in seen:
continue
# filter short descriptions
if len(op.description) < 15:
continue
seen[op.description] = 1
c.add(op, str=op.description, key=op.id)
return c.cluster()
def process(inStream,
outStream,
fields={
"id": "id",
"text": "text"
},
limits={
"clusters": 10,
"top_documents": 10
}):
all = {}
text_field = fields["text"]
key_field = fields["id"]
max_clusters = limits["clusters"]
max_top_docs = limits["top_documents"]
c = Corpus()
for line in inStream:
data = line.split('\t', 1)[1]
doc = json.loads(data.decode("utf8"))
key = doc[key_field]
all[key] = doc
text = c.add((key, doc[text_field]), key=key)
clusters = c.cluster()
results = []
for c in clusters[:max_clusters]:
tophits = [c.primary]
tophits += [hit["object"] for hit in c.similars[:max_top_docs - 1]]
topdocs = []
for (key, text) in tophits:
topdocs.append(all[key])
results.append({"top_documents": topdocs})
json.dump({"clusters": results}, outStream)