Error during convert_to_tfrecord.py

[spyda90]

Hi,

I was trying the code for the first time in order to predict 3D structure of my target protein using a pre-trained model. Everything was fine until running convert_to_tfrecord.py. When I do that I obtain this error:

Traceback (most recent call last): File "../rgn-master/data_processing/convert_to_tfrecord.py", line 7, in from utils import switch ImportError: cannot import name switch

I'm not so strong in python. Can you help me in understanding if this is a bug of the script or there is a problem in my python/packages installation?

Thank you in advance,

Francesco

[alquraishi]

Thanks for bringing this to my attention. I had accidentally broken the import while shuffling files around. The new updated file should work.

[spyda90]

Thank you, it now works! I take advantage of this ticket to ask something more about running a prediction for a new sequence. I made the first for lines reported in README file, and now everything is fine. In order to use the protling.py, I'm making confusion about model, datasetname and example. My goal is to predict the 3D structure of my target protein, using a model generated (for example) by CASP7 model. Can I download the model or should I have to train it?

Thank you in advance,

Francesco

[st4cks1defl0w]

Just check out Train a new model or continue training an existing model section in the README. All pretrained models are also linked in the README.

[spyda90]

Ok, I'll try tomorrow. Thank you

[spyda90]

Hi,

I think I understood how to use the script. However, after running it, I have obtained the error you can find in the attached log file. I have also attached the configuration file I used (just the CASP7.config in the distribution, with RunName and DatasetName edited). As I read in the README file, I used the script with the "-p" and "-e weighted_test" options to predict the structure contained in the "<DatasetName>/testing" dir. I have also put the files contained in one of the "thinning" subdirectories in the main "training" dir. Could you help me solving the error?

prova.zip

[maverick0004]

@spyda90 try using -g 0, this solved the same error for me. This specifies the GPU to use while running the script. It will look something like this python protling.py RGN12/runs/CASP12/ProteinNet12Thinning90/configuration -d RGN12 -a -g 0 -p

[spyda90]

Wait a second...I have installed tensorflow and not tensorflow-gpu...could the error be due to that???

[spyda90]

Tried to install it...the error I received is pretty the same:

InvalidArgumentError (see above for traceback): No OpKernel was registered to support Op 'CudnnRNNCanonicalToParams' used by node RGN/training/layer0/fw/cudnn_lstm/cudnn_lstm/CudnnRNNCanonicalToParams (defined at /home/fgreco/WORK/Projects/TES-PJ-015-NR2F6/modeling/HomologyModelling/rgn-master/model/model.py:787) with these attrs: [rnn_mode="lstm", seed2=4493, dropout=0, seed=254, num_params=8, input_mode="linear_input", T=DT_FLOAT, direction="unidirectional"] Registered devices: [CPU, XLA_CPU] Registered kernels: [[node RGN/training/layer0/fw/cudnn_lstm/cudnn_lstm/CudnnRNNCanonicalToParams (defined at $PATH/HomologyModelling/rgn-master/model/model.py:787) ]]

Why it says "Registered devices: [CPU, XLA_CPU]"? PS: also the -g0 option did the same....

[spyda90]

To note, I'm using a GTX-1050, 2GB

[alquraishi]

GPU doesn't seem to be registered... Are you sure tensorflow-gpu is working correctly with your GPU? Can you construct and place a simple graph on the GPU?