You can download and run the notebooks on your Jupyter-enabled machine.
- Ubuntu 14.04 LTS
- NVIDIA GPU (recommended GTX 970 or higher)
- CUDA 7.5 (installed under /usr/local/cuda)
- cuDNN v4 (optional)
- Python 2.7 / 3.4
- Chainer v1.20.0.1
- All version after v1.5.0 should work
- Additional python packages
- pyparsing == '2.1.10'
- pydot == '1.2.3'
- jupyter
- matplotlib
Copy .ipynb files, data.py, and image directory into a jupyter-managed directory.
$ jupyter notebook
This repository also provide helper scripts to install CUDA, Chainer, and Jupyter on Ubuntu 14.04.
CUDA libraries will be installed into /usr/local/cuda/
.
$ ./install-cuda.sh
Reboot is needed before proceeding. You can also install cuDNN by yourself at this point.
Python packages will be installed under $HOME/.local/
with pip install --user
$ ./install-chainer.sh
$ ./install-jupyter.sh
Make sure that the environmental variables are loaded.
$ source ~/.bash_profile
$ jupyter notebook
The access URL is https://<ip address>:8888/
. The password is "chainer".