Problems in tensorflow 1.2.0

[tyashima]

Hi, I'd like to run your rnn_gan.py, but probably because I'm using the latest version of tensorflow,
I have a problem. The script doesn't work.
I modified several lines because in tensorflow 1.2.0 they change the ordering, notation, and so on.
Also, since we don't have tf.nn.bidirectional_rnn in tensorflow 1.2.0, I tried other bidirectional rnn.
Then I got some errors, (mostly ValueErrors, rank incorrespondence, different input shape, etc...)

I'm using Ubuntu 14.04 with GeForce GTX TITAN X (12GB).
Is it because of the different version of tensorflow? or Am I missing some points?
Also, could you tell me which version of tensorflow you're using?

Thanks.

[olofmogren]

Hi Tiyashima, The code needs some modifications to be able to run on tf 1.2. I have already done most of them but they need some testing before push. Feel free to try and do it yourself. Olof

…

-- Phone: +46-703-969624 Web:  http://mogren.one/ Email: mogren@chalmers.se, olof@mogren.one

-------- Ursprungligt meddelande -------- Ämne: [olofmogren/c-rnn-gan] Problems in tensorflow 1.2.0 (#3) Från: tyashima <notifications@github.com> Till: olofmogren/c-rnn-gan <c-rnn-gan@noreply.github.com> Kopia: Subscribed <subscribed@noreply.github.com>

Hi, I'd like to run your rnn_gan.py, but probably because I'm using the latest version of tensorflow, I have a problem. The script doesn't work. I modified several lines because in tensorflow 1.2.0 they change the ordering, notation, and so on. Also, since we don't have `tf.nn.bidirectional_rnn` in tensorflow 1.2.0, I tried other bidirectional rnn. Then I got some errors, (mostly ValueErrors, rank incorrespondence, different input shape, etc...) I'm using Ubuntu 14.04 with GeForce GTX TITAN X (12GB). Is it because of the different version of tensorflow? or Am I missing some points? Also, could you tell me which version of tensorflow you're using? Thanks. -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: #3

[tyashima]

Thank you for your reply.

As I said before, I modified the script and I thought it would work, but somehow I got an error.
so let me clarify what is happening to me.

When I type this
python rnn_gan_mod.py --model small --datadir midis --traindir traindata --select_validation_percentage 20 --select_test_percentage 20

I got this error (and some outputs for your reference).

num_song_features:4
num_meta_features:155
songlength: 4
(20, 4, 4)
(20, 4, 4)
self._input_songdata shape (20, 4, 4)
generated data shape (20, 4)
Traceback (most recent call last):
File "rnn_gan_mod.py", line 864, in
tf.app.run()
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/app.py", line 48, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "rnn_gan_mod.py", line 684, in main
m = RNNGAN(is_training=True, num_song_features=num_song_features, num_meta_features=num_meta_features)
File "rnn_gan_mod.py", line 367, in init
self.real_d,self.real_d_features = self.discriminator(songdata_inputs, is_training, msg='real')
File "rnn_gan_mod.py", line 433, in discriminator
outputs, state_fw, state_bw = tf.nn.bidirectional_dynamic_rnn(cell_fw, cell_bw, inputs, initial_state_fw=self._initial_state_fw, initial_state_bw=self.initial_state_bw)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/rnn.py", line 375, in bidirectional_dynamic_rnn
time_major=time_major, scope=fw_scope)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/rnn.py", line 574, in dynamic_rnn
dtype=dtype)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/rnn.py", line 637, in dynamic_rnn_loop
for input in flat_input)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/rnn.py", line 637, in
for input in flat_input)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_shape.py", line 649, in with_rank_at_least
raise ValueError("Shape %s must have rank at least %d" % (self, rank))
ValueError: Shape (4, 20) must have rank at least 3

Did you encounter this sort of errors before?
and do you have any idea why it happens?
Any comments would be appreciated, thanks.

[JihadFenix]

Same errors for me too. Really interested for any pointers adapting your code to tf 1.2.

[kiwiloveskiwis]

Same problem here

[Ddedalous]

same problem here.anyone have any fixes?

[RFirstman]

I changed tf.nn.bidirectional_dynamic_rnn to tf.nn.static_bidirectional_rnn, and was able to get past the error posted above. In addition, a few lines above this, there are two statements that contain tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * FLAGS.num_layers_d, state_is_tuple=True). An array of lstm_cell is being made, but each of these refers to the same cell. So, I fixed it by making a simple function called getcell() that sets up the LSTM as usual, but I call it FLAGS.num_layers_d times to make that many distinct cells. So, those statements should now read tf.nn.rnn_cell.MultiRNNCell([getcell() for _ in range(FLAGS.num_layers_d)], state_is_tuple=True)

[olofmogren]

Please create a pull request, @RFirstman!

[bbrito]

@RFirstman could you share your fixes?

[RFirstman]

My fixes were only band-aid solutions, unfortunately. I was unable to get it to work following my above advice. However, I did find a fork that does fix the issues and uses TensorFlow 1.4.

https://github.com/dhruvsharma1992/Music-generation-cRNN-GAN

[bbrito]

Ok cool! I will try it! Thank you so much!

[olofmogren]

I have now incorporated the changes made by Dhruv Sharma into the original repo. It now runs and trains on TF 1.x.