39a106e526e70b5d64bf9f5404311eb490df6ea5,inference_worker.py,,main,#,32

Before Change

        batch_size=server_config.batch_size,
        descriptor_path=FLAGS.descriptor)

    n = dual_net.DualNetwork(FLAGS.model, worker_config)

    outputs = {
        "features": n.sess.graph.get_tensor_by_name("unflatten_features:0"),
        "dummy": n.sess.graph.get_tensor_by_name("put_outputs:0"),
    }
    output = n.sess.run(outputs)

    for f in output["features"]:
        print("---------------------------")
        for r in range(go.N):
            row = ""
            for c in range(go.N):
                if f[r][c][0]:
                    row += " 0"
                elif f[r][c][1]:
                    row += " 1"
                else:
                    row += " ."
            print(row)


if __name__ == "__main__":
    flags.FLAGS(sys.argv, known_only=True)
    main()

After Change

    config = tf.ConfigProto()
    config.gpu_options.allow_growth = True
    sess = tf.Session(graph=tf.Graph(), config=config)
    with sess.graph.as_default():
        loop = dual_net.model_inference_worker_fn(worker_config)
        tf.train.Saver().restore(sess, FLAGS.model)

    print("running graph")
    sess.run(loop)

Instances