Running examle code about macad-gym

[praveen-palanisamy]

Hi @Yiquan-lol ,
This is likely related to #27. To confirm, could you add the python package versions you are using in your env?
You can run this command: pip freeze > packages.txt and upload the packages.txt in your reply.

[Yiquan-lol]

package.txt
I created the environment from the conda-env.yml file

[praveen-palanisamy]

@Yiquan-lol : Thanks for the python environment info.
I am unfortunately not able to reproduce this error. Could you please try again with this line:
import tensorflow
added after this line:

macad-gym/src/macad_gym/carla/multi_env.py

Line 219 in 38884ac

try:

and report back if your issues is solved/not-solved?:

[Yiquan-lol]

Oh , I followed the same steps to reinstall the Conda environment and basci agent.py is running.Thank you for your previous help

[Yiquan-lol]

I found a problem: after setting up the macad-gym environment, I could run basic-agent.py, but the 'import tensorflow' would tell me that I was missing the cudatoolkit file.
When I used 'conda install cudatoolkit==8.0 condainstall cudnn', I failed to run basic_agent.py and had the same problem as before.

[praveen-palanisamy]

Thank you for the follow-on and updates!
Could you post the full error log? That'll help to see what's going wrong.

[Yiquan-lol]

After installing the MacAd-Gym environment, you can run basic_agent.py.When running "import tensorflow", "error_1" appears.
error_1.txt
After I install the cudatoolkit and cudnn, run "basic_anent" and "error_2" will appear.
error_2.txt
Can you use TensorFlow directly after installing the MacAd-Gym environment?
Did you install cudatoolkit and cudnn via "conda install"?

[praveen-palanisamy]

Thanks for the logs with clear explanation!

1. After you installed cudatoolkit==8.0, did you try adding the import tensorflow to that particular line in that file as mentioned here: #40 (comment) ?

2. Can you use TensorFlow directly after installing the MacAd-Gym environment?

   Yes. In case this detail helps, I have cuda-8.0 on my system and the following commands sets up the required env vars:

   • export PATH="$PATH:/usr/local/cuda-8.0/bin"
   • export LD_LIBRARY_PATH="/usr/local/cuda-8.0/lib64"

3. Did you install cudatoolkit and cudnn via "conda install"?

   No

[Yiquan-lol]

I did what you told me, but I still made the same error.
Can I change the tensorFlow version in conda_env.yml and reinstall the macad-yym environment to accommodate the CUDA version of my system?

[Yiquan-lol]

I used the conda directive to install the same version of TensorFlow as yours：
conda install tensorflow-gpu==1.4.1
When I run basic agent.py again, I get a new error：
error_3.txt
I don't know what's going on here

[praveen-palanisamy]

@Yiquan-lol , did you happen to get this issue resolved?
It doesn't seem to be easily reproducible and your error log points to some issues at the cpython/symbol level, specific to the GNU libc on your system and the CARLA wheel built using cpython.

[Yiquan-lol]

Oh, that's too bad.
I see that your Conda environment has Torch, so maybe I can use PyTorch for training.
Thank you for your help.

[praveen-palanisamy]

@Yiquan-lol , if you are unblocked, can this issue be closed?