Open prashantsingh905 opened 7 years ago
you should use this : cell = tf.contrib.rnn.LSTMCell(self.num_input, state_is_tuple=False) outputs, state = tf.contrib.rnn.static_rnn(cell, [_X [step] ], state, dtype=tf.float32)
This stackoverflow answer might help you to run ROLO with the newer versions of tensorflow http://stackoverflow.com/questions/41789133/c-state-and-m-state-in-tensorflow-lstm
Could anyone run the pretrained model? I am using tf1.1. which version of tf supports tf.nn.rnn?
@suryaprakaz Hello, do you run ROLO successfully? And do you run MOLO successfully?
@sunshinezhihuo,I run the ROLO successfully on the pretrained model,but I can't reproduce the result of the autor.The result run on the pretrained model is worse..
There seem to be more and more places where the code needs to be modified to be compatible with updates to tensorflow. Since nobody is maintaining the ROLO code anymore, does anybody know which tensorflow version I should use so that ROLO works out of the box?
Edit: Since the last update in this repo says:
Guanghan update for tensorflow 0.11.0
Latest commit 6612007 on Nov 2, 2016
and the oldest available version of tensorflow-gpu in pip is 0.12.rc0, I tried that (along with CUDA-8.0) ... but it didn't work.
There were errors with the import, then
TypeError: 'Tensor' object is not iterable.
on the line
outputs, state = tf.nn.rnn(cell, [_X[step]], state)
Incorporating the suggestions from @masoumeh1 changed the error message, usually to some form of
AttirbuteError: 'module' object has no attribute 'static_rnn'
i am trying to train ROLO_step6_train_20_exp1.py after converting it with tensorflow upgrade script. But still i am facing this issue and i have raw_rnn and dynamic_rnn . but with both there is also a error.Please let me know the fix if anybody trained this on v1.0.1
File "updated1_ROLO_step6_train_20_exp1.py", line 117, in LSTM_single outputs, state = tf.nn.rnn(cell, [_X[step]], state) AttributeError: 'module' object has no attribute rnn