You can get a list of indexes for the misclassified instances and use that with your dataset (eval_examples).

To get the list of indexes, you could do something like;

misclassified = [i for i in range(len(preds)) if preds[i] != labels[i]]

You could add that to the get_eval_report() function.