Skip to content

Latest commit

 

History

History
39 lines (26 loc) · 1.69 KB

README.md

File metadata and controls

39 lines (26 loc) · 1.69 KB
Raw

Docker

We have created two docker images for our experiments (cpu-based and gpu-based). First, you must install Docker from here.

Images

  • CPU: kbogdan/docker_dl_up:cpu This image was created from the image provided by saiprashanths. In our version TensorFlow, Keras, and other libraries were updated (e.g. Numpy). Update: nltk installed.
  • GPU: kbogdan/cuda-keras-tf:latest This image was created from here. In our version, TensorFlow and other libraries were installed (e.g. Numpy, SciPy, Cython, nltk). Update: matplotlib installed.

Once Docker is installed, you can follow the steps:

  1. Get the image, where image_name:tag is one of the above options:

    docker pull image_name:tag

  2. Create a new container.

    CPU

    docker run -it -p 8888:8888 -p 6006:6006 -v /your-local-folder:/root/sharedfolder image_name:tag bash

    GPU

    nvidia-docker run -it -p 8888:8888 -p 6006:6006 -v /your-local-folder:/root/sharedfolder image_name:tag bash

    /your-local-folder:/root/sharedfolder means that you are going to share a folder (your-local-folder) with the container. This folder in the container will be: /root/sharedfolder.

In order to change the backend of Keras to Theano you can change the property 'backend' in the file ~/.keras/keras.json from 'tensorflow' to 'theano'. Configuration of ~/.keras/keras.json in both docker images:

{ "image_dim_ordering": "th", "epsilon": 1e-07, "floatx": "float32", "backend": "tensorflow" }

TODO version of all libraries in both images