Classification

EDeN can be directly used to induce predictive estimators for graphs using the popular scikit-learn library.

The first thing to do is to load the graphs. EDeN offers an io module that can read popular formats like the gSpan graph format or the JSON serialization of networkx graphs.



In [1]:

    
from eden.io.gspan import load
pos_graphs = list(load('data/bursi.pos.gspan'))
neg_graphs = list(load('data/bursi.neg.gspan'))
graphs = pos_graphs + neg_graphs

y = [1]*len(pos_graphs) + [-1]*len(neg_graphs)
import numpy as np
y = np.array(y)

EDeN exports a vectorize function that converts a list of graphs in input to a data matrix in output.

The output format is a scipy Compressed Sparse Row matrix.



In [2]:

    
%%time
from eden.graph import vectorize

X = vectorize(graphs, complexity=2)
print 'Instances: %d Features: %d with an avg of %d features per instance' % (X.shape[0], X.shape[1],  X.getnnz()/X.shape[0])









    



Instances: 4337 Features: 65537 with an avg of 184 features per instance
CPU times: user 10.4 s, sys: 2.44 s, total: 12.9 s
Wall time: 12.1 s

Several predictive algorithms from the scikit library can process data in csr format.



In [3]:

    
%%time
from sklearn.linear_model import SGDClassifier
from sklearn.model_selection import cross_val_score

predictor = SGDClassifier(average=True, class_weight='balanced', shuffle=True, n_jobs=-1)
scores = cross_val_score(predictor, X, y, cv=10, scoring='roc_auc')
print('AUC ROC: %.4f +- %.4f' % (np.mean(scores),np.std(scores)))









    



AUC ROC: 0.9007 +- 0.0153
CPU times: user 920 ms, sys: 84.5 ms, total: 1 s
Wall time: 1.17 s