I want to use ENAS arch to inference my own dataset, so I modified the original code to use placeholder input
e.g)

 loss, lr, gn, tr_acc, _ = sess.run(run_ops, feed_dict={child_ops["X"]:images["train"][i*batch:(i+1)*batch],
                                                                           child_ops["Y"]:labels["train"][i*batch:(i+1)*batch],

and naming the last output tensor like
self.train_preds = tf.argmax(logits, axis=1, name="output")

but restoring the checkpoint file with raw tf.Session(), it keeps throwing the same digits (MNIST)

any advice for this??
Thanks