CODEWITHSUNDEEP
pythontensorflow
WangQL
WangQL

Reputation: 21

ValueError: Attempt to have a second RNNCell use the weights of a variable scope that already has weights

I run ptb_word_lm.py provided by Tensorflow 1.0, but it shows this message:

ValueError: Attempt to have a second RNNCell use the weights of a variable scope that already has weights: 'Model/RNN/multi_rnn_cell/cell_0/basic_lstm_cell'; and the cell was not constructed as BasicLSTMCell(..., reuse=True). To share the weights of an RNNCell, simply reuse it in your second calculation, or create a new one with the argument reuse=True.

Then I modify the code, add 'reuse=True' to BasicLSTMCell, but it show this message:

ValueError: Variable Model/RNN/multi_rnn_cell/cell_0/basic_lstm_cell/weights does not exist, or was not created with tf.get_variable(). Did you mean to set reuse=None in VarScope?

How could I solve this problems?

Upvotes: 2

Views: 3092

Answers (5)

prateek agrawal
prateek agrawal

Reputation: 463

The following worked for me:

with tf.variable_scope('rnn'):
    outputs, final_state = tf.nn.dynamic_rnn(lstm_cell, X_in, initial_state=init_state, time_major=False, scope='rnn')

Upvotes: 1

Li Yu
Li Yu

Reputation: 37

You can trying to add scope='lstmrnn' in your tf.nn.dynamic_rnn() function.

Upvotes: 0

Hossein
Hossein

Reputation: 2111

Just add the following line on top of your code:

tf.reset_default_graph()

Upvotes: 0

wades
wades

Reputation: 36

It's modify lstm_cell() as follows:

def lstm_cell():
    if 'reuse' in inspect.signature(tf.contrib.rnn.BasicLSTMCell.__init__).parameters:
        return tf.contrib.rnn.BasicLSTMCell(size, forget_bias=0.0,
                                     state_is_tuple=True, 
                                     reuse=tf.get_variable_scope().reuse)
    else:
        return tf.contrib.rnn.BasicLSTMCell(
                                     size, forget_bias=0.0, state_is_tuple=True)

My Environment:

Windows 10_x64

tensorflow-gpu: 1.1.0

test is OK.

Upvotes: 0

se zk
se zk

Reputation: 11

Adding reuse = tf.get_variable_scope().reuse to BasicLSTMCell is OK to me.

Upvotes: 1

Related Questions