Hi Jeff,
I have created a Dockerfile for NuPIC a few weeks ago, but it was a little
hacky. Today I updated the Dockerfile in Github
(https://github.com/Allanino/docker-nupic/blob/master/Dockerfile) so that it
installs the Python dependendies from the requeriments.txt
(https://github.com/numenta/nupic/blob/master/external/common/requirements.txt).
After that change, the hacks are no more necessary. The Docker container pass
all the unit tests for NuPIC and run fine the hotgym example.
As a testimony of Docker usage, recently I finished a project that uses NuPIC
to catch anomalies in servers response times. I used Docker to put NuPIC with
all my Python scripts into a single container (including a script to deliver a
static html with real-time graphs of the results). To deploy the project, I
only started an Amazon EC2 instance running CoreOS and executed a single
command inside it to start the container.
I have worked with Docker only for a month, but if you need any help maybe I
could give you a hand, so feel free to ask.
Best regards,
Allan
Em Quinta-feira, 14 de Novembro de 2013 14:10,
"[email protected]" <[email protected]> escreveu:
Hi Marek -
Yes, you can manage whatever you like in a VM. That is actually the point
of Docker - that you don't have to. As we have seen, keeping the
environment up to date with changes in the codebase can be a challenge as
the core developers plow ahead. With Docker, the environment dependencies
are part of the package (but don't conflict with your host system
packages). So, let's say that the team decides that they need to upgrade to
the latest version of numpy in order to do something new on NuPIC. Other
developers pull down the latest changes to NuPIC, and now their code no
longer works, because they don't have the proper version of numpy. With
Docker, they can pull down the changes to NuPIC plus the numpy update all
at once using a git-like interface. They don't have to download an entirely
new VM, or install numpy on their own. Just what you need. Kind of the best
of both worlds (somewhere between a VM, and managing all of your own
packages natively).
All that being said, I haven't actually spent much time with Docker, and I
am still fiddling with it, checking it out. If other people have tried
working with Docker and have stories to tell, please share. Here is a video
with some guys talking about it that kind of helps describe them: Linux
Action Show! <http://www.jupiterbroadcasting.com/37396/> Skip to minute
0:47:00
Also, there is the Docker website: https://www.docker.io/
- Jeff
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
