我想在一些时间序列数据上运行一个 GRU 单元,以根据最后一层中的激活对它们进行聚类。我对 GRU 单元的实现做了一个小改动
def __call__(self, inputs, state, scope=None):
"""Gated recurrent unit (GRU) with nunits cells."""
with vs.variable_scope(scope or type(self).__name__): # "GRUCell"
with vs.variable_scope("Gates"): # Reset gate and update gate.
# We start with bias of 1.0 to not reset and not update.
r, u = array_ops.split(1, 2, linear([inputs, state], 2 * self._num_units, True, 1.0))
r, u = sigmoid(r), sigmoid(u)
with vs.variable_scope("Candidate"):
c = tanh(linear([inputs, r * state], self._num_units, True))
new_h = u * state + (1 - u) * c
# store the activations, everything else is the same
self.activations = [r,u,c]
return new_h, new_h
在此之后,我以以下方式连接激活,然后在调用此 GRU 单元的脚本中返回它们
@property
def activations(self):
return self._activations
@activations.setter
def activations(self, activations_array):
print "PRINT THIS"
concactivations = tf.concat(concat_dim=0, values=activations_array, name='concat_activations')
self._activations = tf.reshape(tensor=concactivations, shape=[-1], name='flatten_activations')
我以下列方式调用 GRU 单元
outputs, state = rnn.rnn(cell=cell, inputs=x, initial_state=initial_state, sequence_length=s)
其中s
是一个批处理长度数组,其中包含输入批处理的每个元素中的时间戳数。
最后我使用
fetched = sess.run(fetches=cell.activations, feed_dict=feed_dict)
执行时出现以下错误
Traceback(最近一次调用):文件“xxx.py”,第 162 行,在 fetched = sess.run(fetches=cell.activations, feed_dict=feed_dict) 文件“/xxx/local/lib/python2.7/site- packages/tensorflow/python/client/session.py”,第 315 行,运行中返回 self._run(None, fetches, feed_dict) 文件“/xxx/local/lib/python2.7/site-packages/tensorflow/python/ client/session.py",第 511 行,在 _run feed_dict_string) 文件 "/xxx/local/lib/python2.7/site-packages/tensorflow/python/client/session.py",第 564 行,在 _do_run target_list) 文件“/xxx/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”,第 588 行,在 _do_call 六.reraise(e_type, e_value, e_traceback) 文件中“/xxx/local/lib /python2.7/site-packages/tensorflow/python/client/session.py”,第 571 行,在 _do_call 返回 fn(*args) 文件“/xxx/local/lib/python2.7/site-packages/tensorflow/python/client/session.py”,第 555 行,在 _run_fn
return tf_session.TF_Run(session, feed_dict, fetch_list, target_list) tensorflow.python.pywrap_tensorflow.StatusNotOK: Invalid argument: 为 RNN/cond_396/ClusterableGRUCell/flatten_activations:0 返回的张量无效。
有人可以深入了解如何在最后一步通过可变长度序列从 GRU 单元中获取激活吗?谢谢。