我试图恢复一个预先训练的LSTM模型，并用它来评估新数据。但是它一直说张量不是该图的元素。我尝试了许多解决方案，但都没有结果。
def main():
graph = tf.Graph()
with graph.as_default():
x = tf.placeholder(tf.float32, shape=[None, 8], name='Input')
y = tf.placeholder(tf.float32, shape=[None, _NUM_CLASSES], name='Output')
global_step = tf.Variable(initial_value=0, trainable=False, name='global_step')
keep_prob_ = tf.placeholder(tf.float32, name='keep')
learning_rate_ = tf.placeholder(tf.float32, shape=[], name='learning_rate')
with graph.as_default():
lstm_in = tf.layers.dense(x, lstm_size, activation=None)
lstm_in = tf.split(lstm_in, 1, 0)
lstm = tf.contrib.rnn.BasicLSTMCell(lstm_size)
drop = tf.contrib.rnn.DropoutWrapper(lstm, output_keep_prob=keep_prob_)
cell = tf.contrib.rnn.MultiRNNCell([drop] * lstm_layers)
initial_state = cell.zero_state(_BATCH_SIZE, tf.float32)
with graph.as_default():
outputs, final_state = tf.contrib.rnn.static_rnn(cell, lstm_in, dtype=tf.float32,
initial_state = initial_state)
softmax = tf.layers.dense(outputs[-1], units=_NUM_CLASSES, name="logits")
y_pred_cls = tf.argmax(softmax, axis=1)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=softmax, labels=y))
tf.summary.scalar("Loss",loss)
train_op = tf.train.AdamOptimizer(learning_rate_)
gradients = train_op.compute_gradients(loss)
capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in gradients]
optimizer = train_op.apply_gradients(capped_gradients)
correct_prediction = tf.equal(y_pred_cls, tf.argmax(y, axis=1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
tf.summary.scalar("Accuracy/train", accuracy)
with graph.as_default():
saver = tf.train.Saver()
merged = tf.summary.merge_all()
sess = tf.Session(graph=graph)
print("Trying to restore last checkpoint ...")
last_chk_path = tf.train.latest_checkpoint(checkpoint_dir=_SAVE_PATH)
saver.restore(sess, save_path=last_chk_path)
def get_emg_array(msg):
global counter, predicted_class_last, array
val_state = sess.run(cell.zero_state(_BATCH_SIZE, tf.float32))
emg_data = numpy.array([list(msg.data)])
predicted_class = sess.run(y_pred_cls, feed_dict={x: emg_data, keep_prob_: 1.0, learning_rate_: learning_rate})
sess.close()
if __name__ == '__main__':
main()
[错误] [1558437001.632465]：错误的回调：回溯（最近一次调用最近）：文件 “ /opt/ros/kinetic/lib/python2.7/dist-packages/rospy/topics.py”，第 750，在_invoke_callback中 cb（msg）文件“ /home/luke/bionic_hand_ws/src/bionic_hand/scripts/EMG_LSTM.py”，行 114，在get_emg_array中 val_state = sess.run（cell.zero_state（_BATCH_SIZE，tf.float32））文件 “ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 929行，正在运行 run_metadata_ptr）文件“ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， _run中的第1137行 self._graph，提取，feed_dict_tensor，feed_handles = feed_handles）文件 “ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第471行，在 init 中 self._fetch_mapper = _FetchMapper.for_fetch（fetches）文件“ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第261行，在for_fetch中 返回_ListFetchMapper（fetch）文件“ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第370行，在 init self._mappers = [_FetchMapper.for_fetch（fetch）用于在提取中进行提取]文件 “ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第261行，在for_fetch中 返回_ListFetchMapper（fetch）文件“ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第370行，在 init self._mappers = [_FetchMapper.for_fetch（fetch）用于在提取中进行提取]文件 “ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第271行，在for_fetch中 返回_ElementFetchMapper（fetches，contraction_fn）文件“ /home/luke/venv2/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”， 第307行，在 init '张量。 （％s）'％（fetch，str（e）））ValueError：无法将获取参数解释为张量。 （张量 Tensor（“ MultiRNNCellZeroState_78 / DropoutWrapperZeroState / BasicLSTMCellZeroState / zeros：0”， shape =（128，192），dtype = float32）不是此图的元素。）