I'm using a tf.contrib.estimator.DNNEstimator and I'd like to add my custom histogram to it which would be calculated from the labels and predictions.
I've seen I can very easily add a scalar metric with tf.contrib.estimator.add_metrics() but I can't see how can I add a tf.HistogramProto() for example.
I managed to do it outside the estimator train() and evaluate() cycles, but not from the inside ( with hooks, or adding metrics etc..)
Anyone knows how to do it?
