From fea1b7b69b438fe724ae4458b148f2d3fc7d7739 Mon Sep 17 00:00:00 2001
From: Yomguithereal <guillaumeplique@gmail.com>
Date: Thu, 1 Oct 2020 12:18:41 +0200
Subject: [PATCH] Working projection

---
 .gitignore                            |  1 +
 experiments/monopartite_projection.py | 16 +++++++---
 fog/graph/projection.py               | 46 +++++++++++++++++++++------
 3 files changed, 49 insertions(+), 14 deletions(-)

diff --git a/.gitignore b/.gitignore
index 884c5c3..9190cc3 100644
--- a/.gitignore
+++ b/.gitignore
@@ -10,3 +10,4 @@ __pycache__
 
 build
 dist
+output
diff --git a/experiments/monopartite_projection.py b/experiments/monopartite_projection.py
index f53493d..d0cd949 100644
--- a/experiments/monopartite_projection.py
+++ b/experiments/monopartite_projection.py
@@ -1,8 +1,9 @@
 import csv
-from networkx import Graph
+import networkx as nx
 from fog.graph import cosine_monopartite_projection
+from experiments.utils import Timer
 
-bipartite = Graph()
+bipartite = nx.Graph()
 
 with open('./data/bipartite.csv') as f:
     reader = csv.DictReader(f)
@@ -11,8 +12,13 @@ with open('./data/bipartite.csv') as f:
         account = 'a%s' %line['account']
         url = 'u%s' % line['url']
 
-        bipartite.add_node(account, type='account')
-        bipartite.add_node(url, type='url')
+        bipartite.add_node(account, node_type='account')
+        bipartite.add_node(url, node_type='url')
         bipartite.add_edge(account, url, weight=int(line['weight']))
 
-monopartite = cosine_monopartite_projection(bipartite, 'account', part='type')
+with Timer('quadratic'):
+    monopartite = cosine_monopartite_projection(bipartite, 'account', part='node_type')
+
+print(monopartite.order(), monopartite.size())
+
+nx.write_gexf(monopartite, './output/monopartite.gexf')
diff --git a/fog/graph/projection.py b/fog/graph/projection.py
index 8b8b759..aafe8d3 100644
--- a/fog/graph/projection.py
+++ b/fog/graph/projection.py
@@ -4,16 +4,20 @@
 #
 # Miscellaneous functions related to bipartite to monopartite projections.
 #
+import math
+import networkx as nx
 from collections import defaultdict, Counter
-from networkx import Graph
+
+from fog.metrics import sparse_dot_product
 
 
 def cosine_monopartite_projection(bipartite, keep, part='bipartite', weight='weight',
                                   threshold=None):
-    monopartite = Graph()
+    monopartite = nx.Graph()
 
     vectors = defaultdict(Counter)
 
+    # TODO: what to do with nodes having no edges?
     for n1, n2, w in bipartite.edges(data=weight, default=1):
         p1 = bipartite.nodes[n1][part]
         p2 = bipartite.nodes[n2][part]
@@ -21,22 +25,46 @@ def cosine_monopartite_projection(bipartite, keep, part='bipartite', weight='wei
         assert p1 != p2, 'fog.graph.cosine_monopartite_projection: given graph is not truly bipartite.'
 
         # Swapping so n1 is from part to keep
-        if p2 == part:
+        if p2 == keep:
             n1, n2 = n2, n1
 
         vectors[n1][n2] += w
 
     norms = {}
-    inverted_index = defaultdict(list)
+    nodes = list(vectors)
+    # inverted_index = defaultdict(list)
 
     for node, vector in vectors.items():
-        monopartite.add_node(node, attr_dict=bipartite.nodes[node])
+        monopartite.add_node(node, **bipartite.nodes[node])
         s = 0
 
         for neighbor, w in vector.items():
-            s += w
-            inverted_index[neighbor].append(node)
+            s += w * w
+            # inverted_index[neighbor].append(node)
 
-        norms[node] = s
+        norms[node] = math.sqrt(s)
 
-    # print(inverted_index)
+    # Quadratic version
+    l = len(nodes)
+
+    for i, n1 in enumerate(nodes):
+        norm1 = norms[n1]
+        vector1 = vectors[n1]
+
+        for j in range(i + 1, l):
+            n2 = nodes[j]
+            norm2 = norms[n2]
+            vector2 = vectors[n2]
+
+            w = sparse_dot_product(vector1, vector2)
+
+            if w == 0:
+                continue
+
+            # NOTE: at this point, both norms should be > 0, so no need to test
+            w = w / (norm1 * norm2)
+
+            if threshold is None or w >= threshold:
+                monopartite.add_edge(n1, n2, weight=w)
+
+    return monopartite