gumbel-softmax.py

def sample_gumbel(shape, eps=1e-20): 
  """Sample from Gumbel(0, 1)"""
  U = tf.random_uniform(shape,minval=0,maxval=1)
  return -tf.log(-tf.log(U + eps) + eps)

def gumbel_softmax_sample(logits, temperature): 
  """ Draw a sample from the Gumbel-Softmax distribution"""
  y = logits + sample_gumbel(tf.shape(logits))
  return tf.nn.softmax( y / temperature)

def gumbel_softmax(logits, temperature, hard=False):
  """Sample from the Gumbel-Softmax distribution and optionally discretize.
  Args:
    logits: [batch_size, n_class] unnormalized log-probs
    temperature: non-negative scalar
    hard: if True, take argmax, but differentiate w.r.t. soft sample y
  Returns:
    [batch_size, n_class] sample from the Gumbel-Softmax distribution.
    If hard=True, then the returned sample will be one-hot, otherwise it will
    be a probabilitiy distribution that sums to 1 across classes
  """
  y = gumbel_softmax_sample(logits, temperature)
  if hard:
    k = tf.shape(logits)[-1]
    #y_hard = tf.cast(tf.one_hot(tf.argmax(y,1),k), y.dtype)
    y_hard = tf.cast(tf.equal(y,tf.reduce_max(y,1,keep_dims=True)),y.dtype)
    y = tf.stop_gradient(y_hard - y) + y
  return y

Hi, Eric! Thanks for sharing the code. I have a question regarding line 24-26. You implemented two ways to compute the one-hot. It seems that line 26 could lead to multiple ones when tie occurs, though that is very unlikely. Which way is better? Which is tested and used in your paper?