How to train/evaluate based on node index.

[ghost]

# launch a learning engine. here we split the dataset, 75% as train, 10% as validation and 15% as test.
lg = sess.learning(sub_graph, nodes=[("paper", paper_features)],
                   edges=[("paper", "cites", "paper")],
                   gen_labels=[
                       ("train", "paper", 100, (0, 75)),
                       ("val", "paper", 100, (75, 85)),
                       ("test", "paper", 100, (85, 100))
                   ])

[sighingnow]

For detail about inference after training, please refer to previous discussion here: #133