python - Can we feed a value without defining tf.placeholder? -
here code https://github.com/tensorflow/tensorflow/blob/r0.10/tensorflow/models/rnn/ptb/ptb_word_lm.py,i wonder why can feed model by:
cost, state, _ = session.run([m.cost, m.final_state, eval_op], {m.input_data: x, m.targets: y, m.initial_state: state}) because initial_state isn't tf.placeholder,so how can feed it?
in code, defines class. , defines self._initial_state = cell.zero_state(batch_size, data_type()), state = self._initial_state, , (cell_output, state) = cell(inputs[:, time_step, :], state). after that, self._final_state = state. what's more, defines function in class:
@property def final_state(self): return self._final_state
and here comes
state = m.initial_state.eval() cost, state, _ = session.run([m.cost, m.final_state, eval_op], {m.input_data: x, m.targets: y, m.initial_state: state}) and have ran code locally, quite have difference without state in feeddict.
can help?
because initial_state isn't
tf.placeholder,so how can feed it?
placeholder tensorflow tensor; can feed tensor using feed_dict mechanism,
and have ran code locally, quite have difference without state in feed_dict.
here if don't feed learned state previous batch, state initialized 0 next batch, hence there degradation of results.
Comments
Post a Comment