A bunch of DevOps files.
- [ ] create a CI environment to check that those environments work correctly, by testing against
h5py. (seems that I can't test
matplotlibdue to lack of test data; and I believe testing that could be very fragile since it's a display issue).
Things to remember for Docker images.
if possible, specify versions of numpy and scipy (I think specifying numpy is enough) for Docker containers. I found that tests on them are pretty fragile and I simply want the container to run some really stable verisons which won't fail when running the following lines.
python -c 'import numpy as np;np.test()'
python -c 'import scipy as sp;sp.test()'
Note that it seems it's required to install
build-essential before you can pass all
MATLAB related ansible scripts
ansible/roles/matlab/files, you should have
R2014a_UNIX_original.tar.gz to make the
matlab Ansible role really work. They can be generated from the corresponding ISO files I obtained from somewhere like tpb... These ISO file are available in Yimeng's 128G Flash Drive (TODO: move them to somewhere permanent in lab server)
How to build Bazel and TensorFlow
cd to relevant directory containing
Dockerfile, and then run command like
sudo docker build -t leelabcnbc/bazel:0.4.4-cuda8.0-cudnn5-centos6 . # finally, push it to dockerhub sudo docker push leelabcnbc/bazel:0.4.4-cuda8.0-cudnn5-centos6
for TF, run command like
sudo docker build -t leelabcnbc/tensorflow:0.12.1-centos6-py27-gpu . # finally, push it to dockerhub sudo docker push leelabcnbc/tensorflow:0.12.1-centos6-py27-gpu
then run command like
sudo docker run leelabcnbc/tensorflow:0.12.1-centos6-py27-gpu
to know the path of the wheel file, and then
sudo docker ps -a to get the name of container.
finally run command like
sudo docker cp musing_hawking:/tmp/tensorflow_pkg/tensorflow-0.12.1-cp27-cp27mu-linux_x86_64.whl ~
to get the wheel file.