def __init__(self, batch_size=1):
self._state_feeds = {}
self._state_fetches = []
self._state_feed_names = []
self._batch_size = batch_size
self._graph = tf.get_default_graph()
# Store the feeds and fetches for recurrent states.
statesaver = bookkeeper.recurrent_state()
for state in six.itervalues(statesaver.GetStateDescriptors()):
shape = [d.size for d in state['feed_shape'].dim]
if shape[0] == 0:
shape[0] = batch_size
feed_name = state['feed_op'].name
self._state_feed_names.append(feed_name)
self._state_fetches.append(state['fetch_name'])
def __init__(self, batch_size=1):
self._state_feeds = {}
self._state_fetches = []
self._state_feed_names = []
self._batch_size = batch_size
self._graph = tf.get_default_graph()
# Store the feeds and fetches for recurrent states.
statesaver = bookkeeper.recurrent_state()
for state in six.itervalues(statesaver.GetStateDescriptors()):
shape = [d.size for d in state['feed_shape'].dim]
if shape[0] == 0:
shape[0] = batch_size
feed_name = state['feed_op'].name
self._state_feed_names.append(feed_name)
self._state_fetches.append(state['fetch_name'])