summaryrefslogtreecommitdiff
path: root/gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py
diff options
context:
space:
mode:
Diffstat (limited to 'gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py')
-rw-r--r--gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py259
1 files changed, 121 insertions, 138 deletions
diff --git a/gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py b/gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py
index e39546f1..d3f875ca 100644
--- a/gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py
+++ b/gnowsys-ndf/gnowsys_ndf/ndf/views/search_views.py
@@ -1,20 +1,24 @@
from django.shortcuts import render
from django.http import HttpResponse
-from gnowsys_ndf.ndf.models import *
from django.template import RequestContext
#from stemming.porter2 import stem
-import json
+
try:
from collections import OrderedDict
except ImportError:
# python 2.6 or earlier, use backport
from ordereddict import OrderedDict
+
+import json
import difflib
import string
import datetime
import itertools
import nltk
+from gnowsys_ndf.ndf.models import node_collection, triple_collection
+from gnowsys_ndf.ndf.models import *
+
my_doc_requirement = u'storing_orignal_doc'
reduced_doc_requirement = u'storing_reduced_doc'
to_reduce_doc_requirement = u'storing_to_be_reduced_doc'
@@ -25,10 +29,6 @@ RELATION_SEARCH = u'RELATION_SEARCH'
SEMANTIC_SEARCH = u'SEMANTIC_SEARCH'
POSSIBLE_SEARCH_TYPES = ["GSystem", "File"]
-#############
-collection = get_database()[Node.collection_name]
-#############
-
# CLASS FOR ENCODING INTO JSON - OBJECTID TO STRING CONVERSION
class Encoder(json.JSONEncoder):
def default(self, obj):
@@ -42,12 +42,12 @@ class Encoder(json.JSONEncoder):
def search_page(request, group_id):
ins_objectid = ObjectId()
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if group_ins:
group_id = str(group_ins._id)
else:
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if auth :
group_id = str(auth._id)
else:
@@ -59,8 +59,7 @@ def search_page(request, group_id):
# FUNCTION THAT RETURNS A MINIMUM COMMON CONTEXT THAT ALL SEARCH RESULTS RETURN
def getRenderableContext(group_id):
- col = get_database()[Node.collection_name]
- temp = col.Node.find({"_type":"GSystemType"}, {"name":1, "_id":0})
+ temp = node_collection.find({"_type":"GSystemType"}, {"name":1, "_id":0})
ins_objectid = ObjectId()
allGSystems = []
allGroups=[]
@@ -70,21 +69,21 @@ def getRenderableContext(group_id):
#allGroups = get_public_groups() # LIST OF ALL PUBLIC GROUPS
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if group_ins:
group_id = str(group_ins._id)
else:
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if auth :
group_id = str(auth._id)
group_ins=auth
else:
- group_ins = collection.Node.find_one({'_type': "Group","_id":ObjectId(group_id)})
+ group_ins = node_collection.find_one({'_type': "Group", "_id": ObjectId(group_id)})
if not group_ins:
- group_ins = collection.Node.find_one({'_type': "Author","_id":ObjectId(group_id)})
+ group_ins = node_collection.find_one({'_type': "Author", "_id": ObjectId(group_id)})
allGroups.append(group_ins.name)
allUsers = populate_list_of_group_members(allGroups) # LIST OF ALL USERS IN PUBLIC GROUPS
memList = populate_list_of_members() # LIST OF ALL USERS
@@ -99,12 +98,12 @@ def search_query(request, group_id):
# SCRIPT FOR CONVERTING GROUP NAME RECEIVED TO OBJECTID
ins_objectid = ObjectId()
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ group_ins = node_collection.find_one({'_type': "Group", "name": group_id})
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if group_ins:
group_id = str(group_ins._id)
else:
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if auth :
group_id = str(auth._id)
else:
@@ -116,17 +115,16 @@ def search_query(request, group_id):
# View for returning the search results according to group search
# def search_query_group(request, group_id):
-# col = get_database()[Node.collection_name]
# # SCRIPT FOR CONVERTING GROUP NAME RECEIVED TO OBJECTID
# ins_objectid = ObjectId()
# if ins_objectid.is_valid(group_id) is False :
-# group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if group_ins:
# group_id = str(group_ins._id)
# else:
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if auth :
# group_id = str(auth._id)
# else:
@@ -149,11 +147,11 @@ def results_search(request, group_id, return_only_dict = None):
ins_objectid = ObjectId()
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
+ group_ins = node_collection.find_one({'_type': "Group", "name": group_id})
if group_ins:
group_id = str(group_ins._id)
else:
- auth = collection.Node.one({'_type': 'Author', 'created_by': unicode(userid) })
+ auth = node_collection.one({'_type': 'Author', 'created_by': unicode(userid) })
if auth :
group_id = str(auth._id)
@@ -191,7 +189,7 @@ def results_search(request, group_id, return_only_dict = None):
# CONVERT USERNAME TO INTEGER
if user_reqd_name != "all":
#Query writtent o avoid the error due to User.Object
- auth = collection.Node.one({'_type': 'Author', 'name': user_reqd_name })
+ auth = node_collection.one({'_type': 'Author', 'name': user_reqd_name})
if auth:
user_reqd = int(auth.created_by)
@@ -199,7 +197,6 @@ def results_search(request, group_id, return_only_dict = None):
search_str_user = search_str_user.lower() # CONVERT TO LOWERCASE
search_str_noArticles = list(removeArticles(str(search_str_user))) # REMOVES ARTICLES
search_str_stemmed = list(stemWords(search_str_noArticles, search_str_user)) # STEMS THE WORDS
- col = get_database()[Node.collection_name]
#Check if the user is the super User
Access_policy=""
@@ -208,7 +205,7 @@ def results_search(request, group_id, return_only_dict = None):
else:
Access_policy=["PUBLIC"]
# GET A CURSOR ON ALL THE GSYSTEM TYPES
- all_GSystemTypes = col.Node.find({"_type":"GSystemType"}, {"_id":1})
+ all_GSystemTypes = node_collection.find({"_type":"GSystemType"}, {"_id":1})
#public_groups = get_public_groups() # GET LIST OF PUBLIC GROUPS
#public_groups = group_name_to_id(public_groups) # CONVERT GROUP NAMES TO OBJECTIDS
@@ -234,7 +231,7 @@ def results_search(request, group_id, return_only_dict = None):
if user_reqd != -1:
- exact_match = col.Node.find({'$and':[
+ exact_match = node_collection.find({'$and':[
{"member_of":{'$in':all_list}},
{"created_by":user_reqd},
{"group_set":ObjectId(group_id)},
@@ -242,7 +239,7 @@ def results_search(request, group_id, return_only_dict = None):
{"name":{"$regex":search_str_user, "$options":"i"}}]},
{"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "group_set":1, "url":1})
else:
- exact_match = col.Node.find({'$and':[
+ exact_match = node_collection.find({'$and':[
{"member_of":{'$in':all_list}},
{'$or':[{"access_policy":{"$in":Access_policy}},{'created_by':request.user.id}]},
{"group_set":ObjectId(group_id)},
@@ -280,7 +277,7 @@ def results_search(request, group_id, return_only_dict = None):
word = search_str_stemmed[c]
temp=""
if user_reqd != -1: # user_reqd = -1 => search all users, else user_reqd = pk of user
- temp = col.Node.find({'$and':[
+ temp = node_collection.find({'$and':[
{"member_of":{'$in':all_list}}, {"created_by":user_reqd},
{"group_set":ObjectId(group_id)},
{'$or':[{"access_policy":{"$in":Access_policy}},{'created_by':request.user.id}]},
@@ -288,7 +285,7 @@ def results_search(request, group_id, return_only_dict = None):
{"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "group_set":1, "url":1})
else:
# search all users in created by
- temp = col.Node.find({'$and':[
+ temp = node_collection.find({'$and':[
{"member_of":{'$in':all_list}},
{"group_set":ObjectId(group_id)},
{'$or':[{"access_policy":{"$in":Access_policy}},{'created_by':request.user.id}]},
@@ -328,7 +325,7 @@ def results_search(request, group_id, return_only_dict = None):
# EXACT MATCH OF SEARCH_USER_STR IN NAME OF GSYSTEMS OF ONE GSYSTEM TYPE
if user_reqd != -1:
- exact_match = col.Node.find({'$and':[
+ exact_match = node_collection.find({'$and':[
{"member_of":{'$in':all_list}},
{"created_by":user_reqd},
{"group_set":ObjectId(group_id)},
@@ -336,7 +333,7 @@ def results_search(request, group_id, return_only_dict = None):
{"tags":search_str_user}]},
{"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "group_set":1, "url":1})
else:
- exact_match = col.Node.find({'$and':[
+ exact_match = node_collection.find({'$and':[
{"member_of":{'$in':all_list}}, {'$or':[{"access_policy":{"$in":Access_policy}},{'created_by':request.user.id}]},
{"group_set":ObjectId(group_id)},
{"tags":search_str_user}]},
@@ -363,14 +360,14 @@ def results_search(request, group_id, return_only_dict = None):
while c < len_stemmed:
word = search_str_stemmed[c]
if user_reqd != -1:
- temp = col.Node.find({'$and':[{"tags":word},
+ temp = node_collection.find({'$and':[{"tags":word},
{"member_of":{'$in':all_list}},
{"created_by":user_reqd},
{"group_set":ObjectId(group_id)},
{'$or':[{"access_policy":{"$in":Access_policy}},{'created_by':request.user.id}]}]},
{"name":1, "_id":1, "member_of":1, "created_by":1, "group_set":1, "last_update":1, "url":1})
else:
- temp = col.Node.find({'$and':[{"tags":word},
+ temp = node_collection.find({'$and':[{"tags":word},
{"member_of":{'$in':all_list}},
{'$or':[{"access_policy":{"$in":Access_policy}},{'created_by':request.user.id}]},
{"group_set":ObjectId(group_id)}]},
@@ -409,7 +406,7 @@ def results_search(request, group_id, return_only_dict = None):
if (search_by_contents == True):
# FETCH ALL THE GSYSTEMS THAT HAVE BEEN MAP REDUCED.
- all_Reduced_documents = collection.Node.find({"required_for":reduced_doc_requirement}, {"content":1, "_id":0, "orignal_id":1})
+ all_Reduced_documents = node_collection.find({"required_for": reduced_doc_requirement}, {"content": 1, "_id": 0, "orignal_id": 1})
# ABOVE LINE DOES NOT RETURN ALL GSYSTEMS. IT RETURNS OBJECTS OF "ToReduceDocs" class.
for singleDoc in all_Reduced_documents:
@@ -433,7 +430,7 @@ def results_search(request, group_id, return_only_dict = None):
sorted_content_match_pairs.insert(c, pair) # SORTED INSERT (INCREASING ORDER)
for docId in sorted_content_match_pairs:
- doc = col.Node.find_one({"_id":docId['doc_id'], "access_policy":Access_policy}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "group_set":1, "url":1})
+ doc = node_collection.find_one({"_id":docId['doc_id'], "access_policy":Access_policy}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "group_set":1, "url":1})
try:
grps = doc.group_set
@@ -484,7 +481,7 @@ def results_search_group(request, group_id):
group_ins = {}
ins_objectid = ObjectId()
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})._id
+ group_ins = node_collection.find_one({'_type': "Group", "name": group_id})._id
group_id = ObjectId(group_ins)
else:
pass
@@ -541,9 +538,6 @@ def results_search_group(request, group_id):
#user = str(request.GET['author']) # GET THE VALUE OF AUTHOR FROM THE FORM
- col = get_database()[Node.collection_name] # COLLECTION NAME
-
-
# FORMAT OF THE RESULTS RETURNED
search_results_ex = {'name':[], 'tags':[], 'content':[], 'user':[]}
search_results_st = {'name':[], 'tags':[], 'content':[], 'user':[]}
@@ -556,7 +550,7 @@ def results_search_group(request, group_id):
all_ids = []
# GET A CURSOR ON ALL THE GSYSTEM TYPES
- all_GSystemTypes = col.Node.find({"_type":"GSystemType"}, {"_id":1})
+ all_GSystemTypes = node_collection.find({"_type":"GSystemType"}, {"_id":1})
len1 = all_GSystemTypes.count()
if (search_by_name == 1): # IF 1, THEN SEARCH BY NAME
@@ -567,9 +561,9 @@ def results_search_group(request, group_id):
# EXACT MATCH OF SEARCH_USER_STR IN NAME OF GSYSTEMS OF ONE GSYSTEM TYPE
if user_reqd != -1:
- exact_match = col.Node.find({'$and':[{"member_of":GSType._id},{"created_by":user_reqd}, {"access_policy":Access_policy},{"group_set":group_id},{"name":{"$regex":search_str_user, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ exact_match = node_collection.find({'$and':[{"member_of":GSType._id},{"created_by":user_reqd}, {"access_policy":Access_policy},{"group_set":group_id},{"name":{"$regex":search_str_user, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
else:
- exact_match = col.Node.find({'$and':[{"member_of":GSType._id},{ "access_policy":Access_policy}, {"group_set":group_id}, {"name":{"$regex":search_str_user, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ exact_match = node_collection.find({'$and':[{"member_of":GSType._id},{ "access_policy":Access_policy}, {"group_set":group_id}, {"name":{"$regex":search_str_user, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
# SORT THE NAMES ACCORDING TO THEIR SIMILARITY WITH THE SEARCH STRING
exact_match = list(exact_match)
@@ -591,9 +585,9 @@ def results_search_group(request, group_id):
while c < len_stemmed:
word = search_str_stemmed[c]
if user_reqd != -1:
- temp = col.Node.find({'$and':[{"member_of":GSType._id},{ "group_set":group_id},{ "created_by":user_reqd},{ "access_policy":Access_policy},{ "name":{"$regex":word, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ temp = node_collection.find({'$and':[{"member_of":GSType._id},{ "group_set":group_id},{ "created_by":user_reqd},{ "access_policy":Access_policy},{ "name":{"$regex":word, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
else:
- temp = col.Node.find({'$and':[{"member_of":GSType._id},{ "group_set":group_id}, {"access_policy":Access_policy}, {"name":{"$regex":word, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ temp = node_collection.find({'$and':[{"member_of":GSType._id},{ "group_set":group_id}, {"access_policy":Access_policy}, {"name":{"$regex":word, "$options":"i"}}]}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
#temp_sorted = sort_names_by_similarity(temp, search_str_user)
split_stem_match.append(temp)#temp_sorted)
c += 1
@@ -619,9 +613,9 @@ def results_search_group(request, group_id):
# EXACT MATCH OF SEARCH_USER_STR IN NAME OF GSYSTEMS OF ONE GSYSTEM TYPE
if user_reqd != -1:
- exact_match = col.Node.find({"member_of":GSType._id, "group_set":group_id, "created_by":user_reqd, "access_policy":Access_policy, "tags":search_str_user}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ exact_match = node_collection.find({"member_of":GSType._id, "group_set":group_id, "created_by":user_reqd, "access_policy":Access_policy, "tags":search_str_user}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
else:
- exact_match = col.Node.find({"member_of":GSType._id, "access_policy":Access_policy, "group_set":group_id, "tags":search_str_user}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ exact_match = node_collection.find({"member_of":GSType._id, "access_policy":Access_policy, "group_set":group_id, "tags":search_str_user}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
#exact_match = sort_names_by_similarity(exact_match, search_str_user)
for j in exact_match:
@@ -640,9 +634,9 @@ def results_search_group(request, group_id):
while c < len_stemmed:
word = search_str_stemmed[c]
if user_reqd != -1:
- temp = col.Node.find({"member_of":GSType._id, "group_set":group_id, "tags":word, "created_by":user_reqd, "access_policy":Access_policy}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ temp = node_collection.find({"member_of":GSType._id, "group_set":group_id, "tags":word, "created_by":user_reqd, "access_policy":Access_policy}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
else:
- temp = col.Node.find({"member_of":GSType._id, "group_set":group_id, "access_policy":Access_policy, "tags":word}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ temp = node_collection.find({"member_of":GSType._id, "group_set":group_id, "access_policy":Access_policy, "tags":word}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
#temp_sorted = sort_names_by_similarity(temp, search_str_user)
split_stem_match.append(temp)#_sorted)
@@ -664,7 +658,7 @@ def results_search_group(request, group_id):
sorted_content_match_pairs = []
if (search_by_contents == 1):
- all_Reduced_documents = collection.Node.find({"required_for":reduced_doc_requirement}, {"content":1, "_id":0, "orignal_id":1})
+ all_Reduced_documents = node_collection.find({"required_for": reduced_doc_requirement}, {"content": 1, "_id": 0, "orignal_id": 1})
##print "cursor: ", all_Reduced_documents, all_Reduced_documents.count()
for singleDoc in all_Reduced_documents:
@@ -693,7 +687,7 @@ def results_search_group(request, group_id):
#print "sorted pairs: ", sorted_content_match_pairs
for docId in sorted_content_match_pairs:
- doc = col.Node.find_one({"_id":docId['doc_id'], "group_set":group_id, "access_policy":Access_policy}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
+ doc = node_collection.find_one({"_id":docId['doc_id'], "group_set":group_id, "access_policy":Access_policy}, {"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "url":1})
if (doc != None):
doc = addType(doc)
#print "type added ", doc['created_by'], "value: ", User.objects.get(username=doc['created_by']).pk == 1
@@ -725,20 +719,19 @@ def results_search_group(request, group_id):
# ins_objectid = ObjectId()
# if ins_objectid.is_valid(group_id) is False :
-# group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if group_ins:
# group_id = str(group_ins._id)
# else:
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if auth :
# group_id = str(auth._id)
# else:
# pass
# #print "Group id is: ", group_id
-# col = get_database()[Node.collection_name]
-# temp = col.Node.find({"_type":"GSystemType"}, {"name":1, "_id":0})
+# temp = node_collection.find({"_type":"GSystemType"}, {"name":1, "_id":0})
# allGSystems = []
# for gs in temp:
@@ -756,7 +749,6 @@ def results_search_group(request, group_id):
def get_attributes(request, group_id):
- col = get_database()[Node.collection_name]
attributes = []
##print request.GET['GSystem']
@@ -771,7 +763,7 @@ def get_attributes(request, group_id):
for GSystem_name in GSystem_names:
#print GSystem_name
- test_obj = col.Node.find_one({"_type":"GSystemType", "name":GSystem_name})
+ test_obj = node_collection.find_one({"_type":"GSystemType", "name":GSystem_name})
##print test_obj.keys()
for sg_key in list_of_keys:
if sg_key in test_obj.keys():
@@ -779,7 +771,7 @@ def get_attributes(request, group_id):
attributes.append(sg_key)
#print "\n\nattr: ", attributes, "\n\n"
- temp = col.Node.one({"_type":"GSystemType", "name":GSystem_name})
+ temp = node_collection.one({"_type":"GSystemType", "name":GSystem_name})
for attr in temp.attribute_type_set:
attributes.append(attr.name)
@@ -803,10 +795,9 @@ def user_name_to_id(userNames):
def group_name_to_id(groupNames):
allGroups = []
- col = get_database()[Node.collection_name]
for gr in groupNames:
- sg_gr = col.Node.find({"_type":"Group", "name":gr})
+ sg_gr = node_collection.find({"_type":"Group", "name":gr})
for each in sg_gr:
allGroups.append(each._id)
@@ -815,8 +806,6 @@ def group_name_to_id(groupNames):
def advanced_search_results(request, group_id):
- col = get_database()[Node.collection_name]
-
# READ THE GET VALUES
search_str_user = str(request.GET['search_text']).strip()
search_groups = request.GET.getlist('Groups')
@@ -857,14 +846,14 @@ def advanced_search_results(request, group_id):
#print "attr: ", at_name
# CASE 1 -- SEARCH IN THE STRUCTURE OF THE GSYSTEM
for GSystem_name in GSystem_names:
- GSystem_obj = col.Node.one({"_type":"GSystemType", "name":GSystem_name})
+ GSystem_obj = node_collection.one({"_type":"GSystemType", "name":GSystem_name})
##print GSystem_obj
- if GSystem_obj.has_key(at_name):
+ if at_name in GSystem_obj:
if all_users != 0:
- res = col.Node.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}, "member_of":GSystem_obj._id, at_name:{"$regex":word, "$options":"i" }}, {"name":1, "created_by":1, "last_update":1, "member_of":1 , "group_set":1, "url":1})
+ res = node_collection.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}, "member_of":GSystem_obj._id, at_name:{"$regex":word, "$options":"i" }}, {"name":1, "created_by":1, "last_update":1, "member_of":1 , "group_set":1, "url":1})
else:
- res = col.Node.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}, "created_by":{"$in":all_users}, "member_of":GSystem_obj._id, at_name:{"$regex":word, "$options":"i" }}, {"name":1, "created_by":1, "last_update":1, "member_of":1, "group_set":1, "url":1})
+ res = node_collection.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}, "created_by":{"$in":all_users}, "member_of":GSystem_obj._id, at_name:{"$regex":word, "$options":"i" }}, {"name":1, "created_by":1, "last_update":1, "member_of":1, "group_set":1, "url":1})
for obj in res:
flag = False
#print obj.name
@@ -882,19 +871,19 @@ def advanced_search_results(request, group_id):
# CASE 2 -- SEARCH THE GATTRIBUTES
try:
- attr_id = col.Node.one({"_type":"AttributeType", "name":at_name}, {"_id":1})
- res = col.Node.find({"_type":"GAttribute", "attribute_type.$id":ObjectId(attr_id._id), "object_value":{"$regex":word, "$options":"i"}}, {"name":1, "object_value":1, "subject":1})
+ attr_id = node_collection.one({"_type":"AttributeType", "name":at_name}, {"_id":1})
+ res = triple_collection.find({"_type":"GAttribute", "attribute_type.$id":ObjectId(attr_id._id), "object_value":{"$regex":word, "$options":"i"}}, {"name":1, "object_value":1, "subject":1})
#print "Sttr type: ", attr_id
for obj in res:
if all_users == 0:
- GSystem = col.Node.one({"_id":obj.subject}, {"name":1, "created_by":1, "last_update":1, "member_of":1 , "group_set":1, "url":1})
+ GSystem = node_collection.one({"_id":obj.subject}, {"name":1, "created_by":1, "last_update":1, "member_of":1 , "group_set":1, "url":1})
else:
- GSystem = col.Node.one({"_id":obj.subject, "created_by":{"$in":all_users}}, {"name":1, "created_by":1, "last_update":1, "member_of":1, "group_set":1, "url":1})
+ GSystem = node_collection.one({"_id":obj.subject, "created_by":{"$in":all_users}}, {"name":1, "created_by":1, "last_update":1, "member_of":1, "group_set":1, "url":1})
if GSystem._id not in all_ids:
#print "adding: ", GSystem._id
# THE FOLLOWING CODE MAY BE WRONG IF THE RETURNED NODE IS A MEMBER OF MORE THAN ONE GSYSTEM_TYPE
- #link_obj = collection.Node.one({"member_of":GSystem.member_of[0], "required_for":"Links"}, {"link":1})
+ #link_obj = node_collection.one({"member_of":GSystem.member_of[0], "required_for":"Links"}, {"link":1})
#print "731"
GSystem_groups = GSystem.group_set
#print "groups: ", GSystem_groups
@@ -921,8 +910,7 @@ def advanced_search_results(request, group_id):
def get_public_groups():
- col = get_database()[Node.collection_name]
- cur = col.Node.find({"_type":"Group", "group_type":"PUBLIC"}, {"name":1})
+ cur = node_collection.find({"_type": "Group", "group_type": "PUBLIC"}, {"name": 1})
allGroups = []
for obj in cur:
allGroups.append(obj.name)
@@ -933,10 +921,10 @@ def get_public_groups():
def addType(obj):
##print "received: ", obj.member_of[0]
#i = ObjectId(obj.member_of[0])
- #links = collection.Node.find({"member_of":i, "required_for":"Links"}, {"link":1})
+ #links = node_collection.find({"member_of":i, "required_for":"Links"}, {"link":1})
##print "links count", links.count(), "\n"
- auth = collection.Node.one({'_type': 'Author', 'created_by': obj.created_by})
+ auth = node_collection.one({'_type': 'Author', 'created_by': obj.created_by})
#for ob in links:
obj2 = {}
@@ -1023,14 +1011,13 @@ def populate_list_of_group_members(group_ids):
This function returns a list of users in a given list of groups.
Groups should be given as a list of group ids.
"""
- col = get_database()[Node.collection_name]
memList = []
try:
for gr in group_ids:
# THIS CODE WILL CAUSE PROBLEMS IF THERE ARE MANY GROUPS WITH THE SAME NAME
- group_id = col.Node.find_one({"_type":"Group", "name":gr}, {"_id":1})
- author_list = col.Node.one({"_type":"Group", "_id":group_id._id}, {"author_set":1, "_id":0})
+ group_id = node_collection.find_one({"_type":"Group", "name":gr}, {"_id":1})
+ author_list = node_collection.one({"_type":"Group", "_id":group_id._id}, {"author_set":1, "_id":0})
for author in author_list.author_set:
name_author = User.objects.get(pk=author).username
@@ -1060,12 +1047,12 @@ def get_node_info(request, group_id, node_name):
ins_objectid = ObjectId()
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ group_ins = node_collection.find_one({'_type': "Group", "name": group_id})
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if group_ins:
group_id = str(group_ins._id)
else:
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if auth :
group_id = str(auth._id)
else:
@@ -1073,9 +1060,7 @@ def get_node_info(request, group_id, node_name):
is_list = False
- col = get_database()[Node.collection_name]
-
- list_of_nodes = col.Node.find({"name":node_name}, {"_id":1})
+ list_of_nodes = node_collection.find({"name": node_name}, {"_id": 1})
if list_of_nodes.count() > 1:
list_of_nodes.rewind()
@@ -1088,7 +1073,7 @@ def get_node_info(request, group_id, node_name):
list_nodes = json.dumps(list_nodes, cls=Encoder)
return render(request, 'ndf/node_details.html', {'is_list':1, 'list_nodes':list_nodes, 'all_fields':'', 'groupid':group_id})
else:
- sg_node = col.Node.one({"name":node_name})
+ sg_node = node_collection.one({"name": node_name})
GSTypes = sg_node.member_of
attrs = []
@@ -1098,10 +1083,10 @@ def get_node_info(request, group_id, node_name):
results['last_update'] = str(sg_node.last_update.date())
for GSType in GSTypes:
- obj = col.Node.one({"_id":ObjectId(GSType)})
+ obj = node_collection.one({"_id": ObjectId(GSType)})
for attr in obj.attribute_type_set:
- custom_attrs = col.Node.find({"_type":"GAttribute", "subject":ObjectId(sg_node._id), "attribute_type.$id":ObjectId(attr._id)}, {"name":1, "object_value":1})
+ custom_attrs = triple_collection.find({"_type": "GAttribute", "subject": ObjectId(sg_node._id), "attribute_type.$id": ObjectId(attr._id)}, {"name":1, "object_value":1})
for sg_attr in custom_attrs:
temp = sg_attr.name
i1 = temp.index('--') + 3
@@ -1122,19 +1107,18 @@ def get_node_info2(request, group_id, node_id):
"""
ins_objectid = ObjectId()
if ins_objectid.is_valid(group_id) is False :
- group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if group_ins:
group_id = str(group_ins._id)
else:
- auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+ auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
if auth :
group_id = str(auth._id)
else:
pass
- col = get_database()[Node.collection_name]
- sg_node = col.Node.one({"_id":ObjectId(node_id)})
+ sg_node = node_collection.one({"_id": ObjectId(node_id)})
GSTypes = sg_node.member_of
attrs = []
@@ -1144,10 +1128,10 @@ def get_node_info2(request, group_id, node_id):
results['last_update'] = str(sg_node.last_update.date())
for GSType in GSTypes:
- obj = col.Node.one({"_id":ObjectId(GSType)})
+ obj = node_collection.one({"_id": ObjectId(GSType)})
for attr in obj.attribute_type_set:
- custom_attrs = col.Node.find({"_type":"GAttribute", "subject":ObjectId(sg_node._id), "attribute_type.$id":ObjectId(attr._id)}, {"name":1, "object_value":1})
+ custom_attrs = triple_collection.find({"_type": "GAttribute", "subject": ObjectId(sg_node._id), "attribute_type.$id": ObjectId(attr._id)}, {"name":1, "object_value":1})
for sg_attr in custom_attrs:
temp = sg_attr.name
i1 = temp.index('--') + 3
@@ -1171,11 +1155,11 @@ def get_relations_for_autoSuggest(request, group_id):
ins = [] # HOLDS LIST OF SUGGESTIONS TO RETURN
# FIND NAMES OF ALL GSYSTEMTYPES OR RELATIONTYPES THAT START WITH WHAT HAS BEEN TYPED BY THE USER
- instances = col.Node.find({ "$or": [{ "_type": "RelationType"}, {"_type": "GSystemType"}], "name":{'$regex':"^"+x, "$options":"i"}}, {"name":1})
+ instances = node_collection.find({ "$or": [{ "_type": "RelationType"}, {"_type": "GSystemType"}], "name":{'$regex':"^"+x, "$options":"i"}}, {"name":1})
for inst in instances:
ins.append(prefix + ' ' + inst.name)
- instances = col.Node.find({"_type": "GSystem", "name":{'$regex':"^"+x, "$options":"i"}}, {"name":1})
+ instances = node_collection.find({"_type": "GSystem", "name":{'$regex':"^"+x, "$options":"i"}}, {"name":1})
for inst in instances:
ins.append(prefix + ' ' + inst.name)
@@ -1188,12 +1172,12 @@ def get_relations_for_autoSuggest(request, group_id):
# # GET VALUE FROM TEXT BOX
# ins_objectid = ObjectId()
# if ins_objectid.is_valid(group_id) is False :
-# group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if group_ins:
# group_id = str(group_ins._id)
# else:
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if auth :
# group_id = str(auth._id)
# else:
@@ -1221,8 +1205,7 @@ def ra_search_results(request, group_id):
"""
sq = str(request.GET['search_text']).strip() # SEARCH QUERY
- col = get_database()[Node.collection_name]
- relations = col.Node.find({ "_type": "RelationType"}, {"name":1, "inverse_name":1})
+ relations = node_collection.find({"_type": "RelationType"}, {"name": 1, "inverse_name": 1})
CASE_TWO_THRESHOLD = 0.6
GSYSTEM_MIN_THRESHOLD_CASE1 = 0.69
@@ -1275,9 +1258,9 @@ def ra_search_results(request, group_id):
c += 1
try:
- relationType_obj = col.Node.one({"_type":"RelationType", "$or": [ {"name":max_match_rel}, {"inverse_name":max_match_rel} ] }, {"_id":1, "name":1})
+ relationType_obj = node_collection.one({"_type": "RelationType", "$or": [{"name": max_match_rel}, {"inverse_name": max_match_rel} ] }, {"_id": 1, "name": 1})
- GRelation_objs = col.Node.find({"_type":"GRelation", "relation_type.$id":relationType_obj._id})
+ GRelation_objs = triple_collection.find({"_type": "GRelation", "relation_type.$id": relationType_obj._id})
#subjects = []
#right_subjects = []
@@ -1288,14 +1271,14 @@ def ra_search_results(request, group_id):
right_sub = name_Grel[name_Grel.rfind('--')+3:]
if difflib.SequenceMatcher(None, GSystem_name, left_sub).ratio() > GSYSTEM_MIN_THRESHOLD_CASE1:
- reqd_obj = col.Node.one({"_id":gr_obj.right_subject, "access_policy":"PUBLIC"}, {"_id":1, "name":1, "created_by":1, "last_update":1, "url":1})
+ reqd_obj = node_collection.one({"_id":gr_obj.right_subject, "access_policy":"PUBLIC"}, {"_id":1, "name":1, "created_by":1, "last_update":1, "url":1})
#subjects.append(gr_obj.subject)
#right_subjects.append(gr_obj.name)
reqd_obj = addType(reqd_obj)
result_members.append(reqd_obj)
#print "match %: ", difflib.SequenceMatcher(None, GSystem_name, left_sub).ratio()
elif difflib.SequenceMatcher(None, GSystem_name, right_sub).ratio() > GSYSTEM_MIN_THRESHOLD_CASE1:
- reqd_obj = col.Node.one({"_id":gr_obj.subject, "access_policy":"PUBLIC"}, {"_id":1, "name":1, "created_by":1, "last_update":1, "url":1})
+ reqd_obj = node_collection.one({"_id":gr_obj.subject, "access_policy":"PUBLIC"}, {"_id":1, "name":1, "created_by":1, "last_update":1, "url":1})
#subjects.append(gr_obj.subject)
#right_subjects.append(gr_obj.name)
reqd_obj = addType(reqd_obj)
@@ -1309,7 +1292,7 @@ def ra_search_results(request, group_id):
if max_match < CASE_TWO_THRESHOLD:
# LOOK FOR GSYSTEM_TYPE IN SEARCH QUERY
- GSystemTypes = col.Node.find({"_type":"GSystemType"}, {"name":1})
+ GSystemTypes = node_collection.find({"_type": "GSystemType"}, {"name": 1})
sorted_gst = {}
max_length = 0
@@ -1352,7 +1335,7 @@ def ra_search_results(request, group_id):
sorted_gs = {}
max_length = 0
- GSystems_sq = col.Node.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}}, {"name":1})
+ GSystems_sq = node_collection.find({"_type": {"$in": POSSIBLE_SEARCH_TYPES}}, {"name": 1})
for gs in GSystems_sq:
length = len(gs.name.split())
@@ -1386,13 +1369,13 @@ def ra_search_results(request, group_id):
c -= 1
try:
- GStype = col.Node.one({"name":max_match_rel_gst}, {"_id":1})
- GS_sq = col.Node.find_one({"name":"GSystem", "name":max_match_rel_gs}, {"_id":1})
+ GStype = node_collection.one({"name": max_match_rel_gst}, {"_id": 1})
+ GS_sq = node_collection.find_one({"name": "GSystem", "name": max_match_rel_gs}, {"_id": 1})
- allMembers = col.Node.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}, "member_of":GStype._id}, {"_id":1, "name":1})
+ allMembers = node_collection.find({"_type": {"$in": POSSIBLE_SEARCH_TYPES}, "member_of": GStype._id}, {"_id": 1, "name": 1})
# LOOK FOR GRELATIONS WITH BOTH GSYSTEMS ON EITHER SIDE
- relations = col.Node.find({"_type":"GRelation", "right_subject":GS_sq._id}, {"subject":1})
+ relations = triple_collection.find({"_type": "GRelation", "right_subject": GS_sq._id}, {"subject": 1})
relations.rewind()
for rel in relations:
@@ -1403,7 +1386,7 @@ def ra_search_results(request, group_id):
result_members.append(member)
allMembers.rewind()
- relations = col.Node.find({"_type":"GRelation", "subject":GS_sq._id}, {"right_subject":1})
+ relations = triple_collection.find({"_type": "GRelation", "subject": GS_sq._id}, {"right_subject": 1})
for rel in relations:
for member in allMembers:
if rel.right_subject == member._id:
@@ -1418,7 +1401,7 @@ def ra_search_results(request, group_id):
sorted_gs = {}
max_length = 0
- GSystems_sq = col.Node.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}}, {"name":1})
+ GSystems_sq = node_collection.find({"_type": {"$in": POSSIBLE_SEARCH_TYPES}}, {"name": 1})
for gs in GSystems_sq:
length = len(gs.name.split())
@@ -1452,7 +1435,7 @@ def ra_search_results(request, group_id):
c += 1
# LOOK FOR GSYSTEM_TYPE IN SEARCH QUERY
- GSystemTypes = col.Node.find({"_type":"GSystemType"}, {"name":1})
+ GSystemTypes = node_collection.find({"_type": "GSystemType"}, {"name": 1})
sorted_gst = {}
max_length = 0
@@ -1489,13 +1472,13 @@ def ra_search_results(request, group_id):
c -= 1
try:
- GStype = col.Node.one({"name":max_match_rel_gst}, {"_id":1})
- GS_sq = col.Node.find_one({"name":"GSystem", "name":max_match_rel_gs}, {"_id":1})
+ GStype = node_collection.one({"name": max_match_rel_gst}, {"_id": 1})
+ GS_sq = node_collection.find_one({"name": "GSystem", "name": max_match_rel_gs}, {"_id": 1})
- allMembers = col.Node.find({"_type":{"$in":POSSIBLE_SEARCH_TYPES}, "member_of":GStype._id}, {"_id":1, "name":1, "created_by":1, "last_update":1, "url":1})
+ allMembers = node_collection.find({"_type": {"$in": POSSIBLE_SEARCH_TYPES}, "member_of": GStype._id}, {"_id": 1, "name": 1, "created_by": 1, "last_update": 1, "url": 1})
# LOOK FOR GRELATIONS WITH BOTH GSYSTEMS ON EITHER SIDE
- relations = col.Node.find({"_type":"GRelation", "right_subject":GS_sq._id}, {"subject":1})
+ relations = triple_collection.find({"_type": "GRelation", "right_subject": GS_sq._id}, {"subject": 1})
relations.rewind()
for rel in relations:
@@ -1505,7 +1488,7 @@ def ra_search_results(request, group_id):
member = addType(member)
result_members.append(member)
- relations = col.Node.find({"_type":"GRelation", "subject":GS_sq._id}, {"right_subject":1})
+ relations = triple_collection.find({"_type": "GRelation", "subject": GS_sq._id}, {"right_subject": 1})
for rel in relations:
allMembers.rewind()
for member in allMembers:
@@ -1659,18 +1642,18 @@ def map_reduce(x,mapper,reducer):
"""
def perform_map_reduce(request,group_id):
#This function shall perform map reduce on all the objects which are present in the ToReduce() class Collection
- all_instances = list(collection.ToReduce.find({'required_for':'map_reduce_to_reduce'}))
+ all_instances = list(node_collection.collection.ToReduce.find({'required_for':'map_reduce_to_reduce'}))
for particular_instance in all_instances:
#print particular_instance._id,'\n'
particular_instance_id = particular_instance.id_of_document_to_reduce
#Now Pick up a node from the Node Collection class
- orignal_node = collection.Node.find_one({"_id":particular_instance_id})
- map_reduce_node = collection.MyReduce.find_one({'required_for':'map_reduce_reduced','orignal_doc_id':particular_instance_id})
+ orignal_node = node_collection.find_one({"_id":particular_instance_id})
+ map_reduce_node = node_collection.find_one({'required_for':'map_reduce_reduced','orignal_doc_id':particular_instance_id})
if map_reduce_node:
map_reduce_node.content_org = dict(map_reduce(orignal_node.content_org,mapper,reducer))
map_reduce_node.save()
else:
- z = collection.MyReduce()
+ z = node_collection.collection.MyReduce()
z.content_org = dict(map_reduce(orignal_node.content_org,mapper,reducer))
z.orignal_doc_id = particular_instance_id
z.required_for = u'map_reduce_reduced'
@@ -1685,22 +1668,22 @@ def perform_map_reduce(request,group_id):
#connection.register([ReducedDocs])
#connection.register([ToReduceDocs])
- dltr=list(collection.ToReduceDocs.find({'required_for':to_reduce_doc_requirement})) #document_list_to_reduce
+ dltr=list(node_collection.find({'required_for':to_reduce_doc_requirement})) #document_list_to_reduce
for doc in dltr:
doc_id = doc.doc_id
#print "DOC ID LN 1504::",doc_id
- #orignal_doc = collection.MyDocs.find_one({"_id":doc_id,'required_for':my_doc_requirement})
- orignal_doc = collection.Node.find_one({"_id":doc_id})
+ #orignal_doc = node_collection.find_one({"_id": doc_id,'required_for': my_doc_requirement})
+ orignal_doc = node_collection.find_one({"_id": doc_id})
content_dict = dict(map_reduce(orignal_doc.content_org,mapper,reducer))
- dord = collection.ReducedDocs.find_one({"orignal_id":doc_id,'required_for':reduced_doc_requirement}) #doc of reduced docs
+ dord = node_collection.find_one({"orignal_id": doc_id, 'required_for': reduced_doc_requirement}) #doc of reduced docs
if dord:
dord.content=content_dict
dord.is_indexed = False
dord.save()
else:
- new_doc = collection.ReducedDocs()
+ new_doc = node_collection.collection.ReducedDocs()
new_doc.content = content_dict
new_doc.orignal_id = doc_id
new_doc.required_for = reduced_doc_requirement
@@ -1738,14 +1721,14 @@ def td_doc():
#This is the list of documents which contains the indexed words
- lod = collection.IndexedWordList.find({'required_for':indexed_word_list_requirement}) #list_of_documents_cursor
+ lod = node_collection.find({'required_for': indexed_word_list_requirement}) #list_of_documents_cursor
"""
What does indexing mean?
In our scenario,indexing simply means to store the number if occurances of a particular word in each and every document.
"""
- mrd = collection.ReducedDocs.find({'required_for':reduced_doc_requirement}) #map_reduced_documents
+ mrd = node_collection.find({'required_for': reduced_doc_requirement}) #map_reduced_documents
mrdl = list(mrd)
@@ -1777,7 +1760,7 @@ def generate_big_dict():
#This function will generate a big dictionary i.e. it will simply go and combine all the dictionaries together
#connection.register([IndexedWordList])
- lod = collection.IndexedWordList.find({'required_for':indexed_word_list_requirement})
+ lod = node_collection.find({'required_for': indexed_word_list_requirement})
lodl = list(lod)
prefs = {} #prefs ==> Preferences
@@ -1907,12 +1890,12 @@ def recommend(prefs,term,similarity = sim_distance):
# def cf_search(request, group_id):
# ins_objectid = ObjectId()
# if ins_objectid.is_valid(group_id) is False :
-# group_ins = collection.Node.find_one({'_type': "Group","name": group_id})
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# group_ins = node_collection.find_one({'_type': "Group","name": group_id})
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if group_ins:
# group_id = str(group_ins._id)
# else:
-# auth = collection.Node.one({'_type': 'Author', 'name': unicode(request.user.username) })
+# auth = node_collection.one({'_type': 'Author', 'name': unicode(request.user.username) })
# if auth :
# group_id = str(auth._id)
# else:
@@ -1949,7 +1932,7 @@ def get_nearby_words(request,group_id):
result_array = []
for (relevance,each_id) in final_ranking_list:
- obj = collection.Node.find_one({"_id":ObjectId(each_id)},{"name":1, "_id":1, "member_of":1, "created_by":1, "last_update":1, "group_set":1, "url":1})
+ obj = node_collection.find_one({"_id": ObjectId(each_id)}, {"name": 1, "_id": 1, "member_of": 1, "created_by": 1, "last_update": 1, "group_set": 1, "url": 1})
obj = addType(obj)
obj["relevance"]=relevance
result_array.append(obj)