A minimal image with GPU version of Tensorflow, Keras and Jupyter on top of Python 3. Built with nvidia-docker, which is totally necessary to run it.

CPU version available

Before you continue with GPU version, install NVIDIA-docker with NVIDIA drivers and Docker by the guides under the link

• GPU: docker build -f Dockerfile.gpu -t kekonen/ml:gpu .
• CPU: docker build -f Dockerfile.cpu -t kekonen/ml:cpu .

nvidia-docker run -it --rm --name docker-ml -p 8888:8888 -p 6006:6006 -v ~/code/ml/kaggle:/root/shared -v /media/daniil/2E18DD98594857B9/data:/root/data kekonen/ml:gpu jupyter notebook --allow-root --ip=0.0.0.0 --port=8888 (Replace GPU with CPU)

• Jupyter Notebook: open http://localhost:8888 in your browser
• Attach to command line: docker attach docker-ml

The image contents a set of necessary installations for Tensorflow with Keras (GPU) and Jupyter Notebook. Additionally helpfull libraries for working with data.

• git
• wget
• bzip2
• python3-pip
• liblapack-dev
• libopenblas-dev
• libcupti-dev
• ca-certificates
• sudo
• locales
• fonts-liberation

• tensorflow-gpu
• keras
• jupyter
• numpy
• pandas
• setuptools
• scipy
• scikit-learn
• h5py
• matplotlib
• nltk
• Pillow
• scikit-image
• opencv-python
• sklearn