mariel-pettee / choreography Goto Github PK

Hello, and thank you for sharing this nice project.
I have been trying to setup the environment for a few days now, and I've encountered a series of issues. I will try to describe them, and if any of those ring any bell to anyone... I would love to hear about your thoughts.

After installing all requirements to the letter (as far as I can tell), when I run the command
python rnn.py rnn_test --cells 64 64 64 64 --n_mixes 25 --look_back 128 --batch_size 128 --n_epochs 10 --lr 1e-4 --use_pca
(I had to remove the final True argument as it ws returning an error) I get

File "/home/julien/choreography/functions/functions.py", line 19, in setup_gpus
    gpus = tf.config.experimental.list_physical_devices('GPU')
AttributeError: module 'tensorflow' has no attribute 'config'

From the research I then did, it seems that people face this when using tf 2, where the config attr was moved. So I double, triple checked the version of tf my conda environment was using and it is most definitely v 1.13.1, which should not have this issue.

>>> print(tf.__version__)
1.13.1

I tried various tf versions (1.13.1 & 2, 1.14, 1.15, 2.0.0, 2.1.0). At some point I even had a situation where manuall typing

python import tensorflow as tf
tf.list_physical_devices

was actually returning my GPU information, but where the main code would still give me this attribute error (I couldn't reproduce the situation).

Do you have any suggestions of things I should try? I tried this on both Linux Mint 19.1 and Windows 10, with RTX 2070 Max-Q, python 3.7.6 mostly.

P.S: Would you share the code with which you produce the npy files (from csv files I am assuming)? I am currently experimenting with a fine motion capture system and I would like to produce and train on my own dataset.

Thank you very much in advance !

This occured after command pip install -r requirements.txt
Im running python 3.7 - does this make a difference?