Fixed Struc2Vec & XNetMF

Author: ReactiveXYZ-Jackie

semb/methods/struc2vec/.gitignore

@@ -0,0 +1 @@
+random_walks.txt

semb/methods/struc2vec/algorithms.py

@@ -185,7 +185,7 @@ def generate_random_walks(num_walks,walk_length,workers,vertices):
     save_random_walks(walks)
 
 def save_random_walks(walks):
-    with open('random_walks.txt', 'w') as file:
+    with open(walk_fname, 'w') as file:
         for walk in walks:
             line = ''
             for v in walk:

semb/methods/struc2vec/algorithms_distances.py

@@ -607,7 +607,6 @@ def generate_distances_network(workers):
     logging.info('Creating distance network...')
 
     os.system("rm "+returnPickle().replace(' ', '\ ')+"weights_distances-layer-*.pickle")
-    print('!!!!!')
     print("rm "+returnPickle().replace(' ', '\ ')+"weights_distances-layer-*.pickle")
     with ProcessPoolExecutor(max_workers=1) as executor:
         job = executor.submit(generate_distances_network_part1,workers)

semb/methods/struc2vec/graph.py

@@ -229,7 +229,7 @@ def load_matfile(file_, variable_name="network", undirected=True):
 def from_networkx(G_input, undirected=True):
     G = Graph()
 
-    for idx, x in enumerate(G_input.nodes_iter()):
+    for idx, x in enumerate(G_input.nodes()):
         for y in iterkeys(G_input[x]):
             G[x].append(y)
 

semb/methods/struc2vec/method.py

@@ -6,14 +6,16 @@
 from time import time
 from . import graph
 from semb.methods import BaseMethod
+from .utils import *
+from .graph import *
 
 """
 Driver for learning Struc2Vec embedding
 """
 class Method(BaseMethod):
 
-	__PARAMS__ = dict(dim=128, walk_length=80, num_walks=10, window_size=10, util_layer=None, \
-					  iter=5, worker=1, weighted=False, directed=False, opt1=False, opt2=False, opt3=False)
+	__PARAMS__ = dict(dim=128, walk_length=80, num_walks=10, window_size=10, until_layer=None, \
+					  iter=5, workers=1, weighted=False, directed=False, opt1=False, opt2=False, opt3=False)
 
 
 	def exec_struc2vec(self):
@@ -22,7 +24,9 @@ def exec_struc2vec(self):
 			until_layer = self.params['until_layer']
 		else:
 			until_layer = None
-		G = struc2vec.Graph(self.graph, self.params['directed'],
+
+		graph = from_networkx(self.graph, not self.params['directed'])
+		G = struc2vec.Graph(graph, self.params['directed'],
 		                    self.params['workers'], untilLayer=until_layer)
 		if self.params['opt1']:
 			G.preprocess_neighbors_with_bfs_compact()
@@ -38,8 +42,10 @@ def exec_struc2vec(self):
 		G.preprocess_parameters_random_walk()
 		G.simulate_walks(self.params['num_walks'], self.params['walk_length'])
 
+		walks = LineSentence(walk_fname)
+
 	def learn_embeddings(self):
-		walks = LineSentence('random_walks.txt')
+		walks = LineSentence(walk_fname)
 		self.model = Word2Vec(walks, size=self.params['dim'], window=self.params['window_size'],
 	                	 min_count=0, hs=1, sg=1, workers=self.params['workers'], iter=self.params['iter'])
 		self.embeddings = self.model.wv

semb/methods/struc2vec/pickles/.gitignore

@@ -0,0 +1 @@
+*.pickle

semb/methods/struc2vec/pickles/.gitkeep

@@ -5,8 +5,8 @@
 from itertools import islice
 import os.path
 
-# dir_f = str(Path().resolve()) #+ "/lib/struc2vec/src"#os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
-folder_pickles = os.path.abspath(os.getcwd()) + "/lib/struc2vec/pickles/" # + '/pickles/'
+folder_pickles = os.path.join(os.path.dirname(__file__), "pickles/")
+walk_fname = os.path.join(os.path.dirname(__file__), "random_walks.txt")
 
 def returnPickle():
     return folder_pickles

semb/methods/struc2vec/utils.py

@@ -203,23 +203,3 @@ def get_representations(graph, rep_method, verbose = True):
 	if rep_method.normalize:
 		reprsn = reprsn / np.linalg.norm(reprsn, axis = 1).reshape((reprsn.shape[0],1))
 	return reprsn
-
-if __name__ == "__main__":
-	if len(sys.argv) < 2:
-		#####PUT IN YOUR GRAPH AS AN EDGELIST HERE (or pass as cmd line argument)#####  
-		#(see networkx read_edgelist() method...if networkx can read your file as an edgelist you're good!)
-		graph_file = "data/arenas_combined_edges.txt"
-	else:
-		graph_file = sys.argv[1]
-	nx_graph = nx.read_edgelist(graph_file, nodetype = int, comments="%")
-	adj_matrix = nx.adjacency_matrix(nx_graph).todense()
-	
-	graph = Graph(adj_matrix)
-	rep_method = RepMethod(max_layer = 2) #Learn representations with xNetMF.  Can adjust parameters (e.g. as in REGAL)
-	representations = get_representations(graph, rep_method)
-	print(representations.shape)
-
-
-
-
-

semb/methods/xnetmf/internal/xnetmf.py

@@ -12,7 +12,8 @@ def get_id(self):
     def train(self):
         # learn representations with xNetMF.  Can adjust parameters (e.g. as in REGAL)
         rep_method = RepMethod(max_layer=self.params['max_layer'], p=self.params['dim'], 
-                              alpha=self.params['discount'], gammastruc=self.params['gamma']) 
-        # FIXME: this doesnt look like a standard embeddings format
-        self.embeddings = get_representations(self.graph, rep_method)
-
+                              alpha=self.params['discount'], gammastruc=self.params['gamma'])
+        
+        graph = Graph(nx.adjacency_matrix(self.graph))
+        # FIXME: best way to convert/represent it in the format of the original code?
+        self.embeddings = get_representations(graph, rep_method)