Public Repository

Last pushed: a year ago
Short Description
Machine Learning (GPU) Python3.4 (0.9), with frills like keras, opencv, pandas, seaborn, and bokeh
Full Description

It's a jupyter style docker image, so it should start a server automatically, meaning you should be able to go to localhost:8888 immediately in a web browser and start doing stuff.

To be clear, this is just the official tensorflow:gpu jupyter image upgraded to python3 with a couple of the data science libraries I like stuck on. Now that I know a bit more about tensorflow, I've added keras and openCV for easier prototyping and computer vision stuff. However, note that in particular the addition of openCV adds 6GB for a total of 11GB, but at least it's all CUDA optimized.

With the recent upgrade to 0.9, or somewhere along the way (I'm pulling/building master), running cuda with docker has been abstracted away to some extent with https://github.com/NVIDIA/nvidia-docker. Follow the instructions there to run it. Also, I suggest linking a folder so you don't have to copy data back and forth from containers if you are working locally. ie -v "~/notebooks:/notebooks"

sample code I use to start the notebook:

nvidia-docker run -it --net host -p 8888:8888 -v "/home/thomas/notebooks:/notebooks" thomasekeller/tensorflow-py3-frills

Changed the build to use bazel so it should work fine now, tensorflow imports without error in the notebook on my laptop now.
Dockerfile heavily inspired by grahama's https://hub.docker.com/r/grahama/tf/

Docker Pull Command
Owner
thomasekeller