evaluation.py

import pickle as pkl
import glob
from sklearn import svm
from sklearn.metrics import f1_score
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import cross_val_score
import networkx as nx

def load_labels(path):
	labels = []
	for i in range(len(glob.glob("%s/*.graphml" % (path)))):
		G = nx.read_graphml("%s/%s.graphml" % (path, i))
		labels.append(G.graph("label"))
	return labels

"""
  Loads embeddings from pickle files and returns it.
   
  :param path: path to the folder containing all pickle files. These files are usually generated by the SG2V or run_sgcn scripts.
  :return List of embeddings
"""
def load_embeddings(path):
	embeddings = []
	for i in range(len(glob.glob("%s/*.pkl" % (path)))):
		with open("%s/%s.pkl" % (path, i), "rb") as f:
			emb = pkl.load(f)
		embeddings.append(emb)
	return embeddings

"""
  Performs the evaluation.
   
  :param embeddings: list of embeddings
  :param labels: List of labels
  :return None
"""
def evaluation(embeddings, labels):
	binary_classifier_model = svm.SVC(class_weight="balanced")
	scores = cross_val_score(binary_classifier_model, embeddings, labels, cv=3, scoring='f1_micro')

	print ("Micro F-measure: %0.4f" % (scores.mean()))


def evaluate_SG2V():
	for method in ["g2v", "sg2vn", "sg2vsb"]:
		embeddings = load_embeddings("out/SG2V/%s" % (method))
		labels = load_labels("data/%s" % (method))
		evaluation(embeddings, labels)