README.md

RAPIDS installation for HackTalks workshop

UPDATE: Pull the Docker image again to get the complete notebooks! Slides are in the .pdf file and the notebooks are the .ipynb files.

Any environment

Requirements

• Docker CE v18+
  • Ubuntu: https://docs.docker.com/install/linux/docker-ce/ubuntu/
  • Mac: https://docs.docker.com/docker-for-mac/install/
  • Windows (didn't test): https://docs.docker.com/docker-for-windows/install/

With NVIDIA GPU

Requirements

• Ubuntu 16.04+ (didn't test other distros)

• GPU architecture: Pascal or better

• NVIDIA drivers 396+, CUDA 9.2+:

  • https://developer.nvidia.com/cuda-92-download-archive (Ubuntu 16.04)
  • https://developer.nvidia.com/cuda-downloads (Ubuntu 18.04)

  You need to reboot for the change to take effect.

• Docker CE v18+ https://docs.docker.com/install/linux/docker-ce/ubuntu/

• NVIDIA Docker v2+ https://github.com/NVIDIA/nvidia-docker#ubuntu-140416041804-debian-jessiestretch

Pulling the image and starting it

docker pull tuomars/hacktalks:rapids_with_e2e (pulls from Docker Hub)

docker run --runtime=nvidia -it --rm -p 8888:8888 -p 8787:8787 -p 8786:8786 tuomars/hacktalks:rapids_with_e2e

If the above doesn't work for some reason, try

docker pull tuomars/hacktalks:rapids_without_e2e (pulls from Docker Hub)

docker run --runtime=nvidia -it --rm -p 8888:8888 tuomars/hacktalks:rapids_without_e2e

This image doesn't contain Dask support.

Smoke test

> python
>>> import cudf
>>> import cuml

Launching a notebook

In the container,

./start_notebook.sh

Then, copy-paste the reported URL to your web browser (of the form http://localhost:8888/?token=abc123) and see the folder /notebooks.

Without NVIDIA GPU

You should be able to follow the workshop thanks to cudf and cuml being near drop-in replacements for pandas and sklearn, respectively.

Requirements

• Docker CE v18+

Pulling the image and starting it

docker pull tuomars/hacktalks:cpu (pulls from Docker Hub)

docker run -it --rm -p 8888:8888 tuomars/hacktalks:cpu

Smoke test

> python
>>> import pandas
>>> import sklearn

Launching a notebook

In the container, run

./start_notebook.sh

Then, copy-paste the reported URL to your web browser (of the form http://localhost:8888/?token=abc123) and see the folder /notebooks.

Not recommended for this workshop but possible: local installation

If you have an NVIDIA GPU, see: https://github.com/RAPIDSai and install cudf and cuml following the instructions in the repos.

Requirements

• miniconda/anaconda
• python 3.5+
• pandas (0.20.3)
• scikit-learn
• matplotlib
• notebook

Datasets used in this workshop are available at: https://drive.google.com/file/d/1rdzp7xqtTDbrDaZmreeHfPn81ZacDUtl/view?usp=sharing

Launching a notebook: ipython notebook