HMC sampling in tensorflow

I have been trying to implement it.

tf.contrib.bayesflow.hmc.chain requires an argument called ‘target_log_prob_fn’. I am unable to decipher what this is supposed to be.

def target_log_prob(x):
  # Standard normal
  return tf.reduce_sum(-0.5 * tf.square(x))

using this is giving me wrong values.