Closed kidtronnix closed 6 years ago
Found the fix, simply upgrade pip in the build.
Thank you for reporting the issue and creating PR!
The PR as it is, unfortunately, will break the Docker as the added line is not a valid Docker syntax. There should be either RUN
Docker command (e.g. RUN pip install --upgrade pip
) or the command should be rolled into previous RUN
command.
However, there is a bigger problem: after updating pip, the build script will fail with a different error:
raceback (most recent call last):
File "/usr/bin/pip", line 9, in <module>
from pip import main
ImportError: cannot import name 'main'
This is a known issue and the workaround is to avoid upgrading pip.
Ideally, all Python packages should be installed in a virtual environment (e.g. like we do in our TF container) but it would be hard to do in the current redtail
image with lots of dependencies which might not work well in virtualenvs.
So what I did instead is to do exactly what you suggested in the original post: lock matplotlib version. Not ideal, but the image is now building fine and all components seem to work (I did a clean image build and tested it). I've just pushed the fix to master.
Also, you should be using TRT 4.0, not 3.0. Something like that:
./build_redtail_image.sh /data/downloads/TensorRT-4.0.1.6.Ubuntu-16.04.4.x86_64-gnu.cuda-9.0.cudnn7.1.tar.gz
@Alexey-Kamenev I got same problem. Could you tell me how to lock matplotlib version? Thanks!
@Alexey-Kamenev I got same problem. Could you tell me how to lock matplotlib version? Thanks!
Finally, I find the master branch 'Dockerfile.kinetic' added
&& pip install 'matplotlib==2.2.2' --force-reinstall
The problem has been solved.
Closing the issue, feel free to re-open or create a new one.
Huh, sorry for the incorrect syntax. I just added the fix after trying it on my side and simply forgot to add the RUN.
The pip upgrade didn't break things for me, but locking matplotlib seems to work too :)
When I run...
I get the following error...
Any help is appreciated. I think we just need to lock down a specific
matplotlib