cuDF - GPU DataFrames

  

The RAPIDS cuDF library is a GPU DataFrame manipulation library based on Apache Arrow that accelerates loading, filtering, and manipulation of data for model training data preparation. The RAPIDS GPU DataFrame provides a pandas-like API that will be familiar to data scientists, so they can now build GPU-accelerated workflows more easily.

Quick Start

Please see the Demo Docker Repository, choosing a tag based on the NVIDIA CUDA version you’re running. This provides a ready to run Docker container with example notebooks and data, showcasing how you can utilize cuDF.

Install cuDF

Conda

It is easy to install cuDF using conda. You can get a minimal conda installation with Miniconda or get the full installation with Anaconda.

Install and update cuDF using the conda command:

conda install -c nvidia -c rapidsai -c numba -c conda-forge -c defaults cudf=0.4.0

Note: This conda installation only applies to Linux and Python versions 3.5/3.6.

Pip

Support is coming soon, please use conda for the time being.

Development Setup

The following instructions are for developers and contributors to cuDF OSS development. These instructions are tested on Linux Ubuntu 16.04 & 18.04. Use these instructions to build cuDF from source and contribute to its development. Other operatings systems may be compatible, but are not currently tested.

Get libcudf Dependencies

Compiler requirements:

• gcc version 5.4
• nvcc version 9.2
• cmake version 3.12

CUDA/GPU requirements:

• CUDA 9.2+
• NVIDIA driver 396.44+
• Pascal architecture or better

You can obtain CUDA from https://developer.nvidia.com/cuda-downloads

Since cmake will download and build Apache Arrow (version 0.7.1 or 0.8+) you may need to install Boost C++ (version 1.58+) before running cmake:

# Install Boost C++ for Ubuntu 16.04/18.04
$ sudo apt-get install libboost-all-dev

or

# Install Boost C++ for Conda
$ conda install -c conda-forge boost

Script to build cuDF from source

Build from Source

To install cuDF from source, ensure the dependencies are met and follow the steps below:

• Clone the repository and submodules

git clone --recurse-submodules https://github.com/rapidsai/cudf.git
cd cudf

• Create the conda development environment cudf_dev

# create the conda environment (assuming in base `cudf` directory)
conda env create --name cudf_dev --file conda/environments/dev_py35.yml
# activate the environment
source activate cudf_dev

• Build and install libcudf. CMake depends on the nvcc executable being on your path or defined in $CUDACXX.

$ cd cpp                                            # navigate to C/C++ CUDA source root directory
$ mkdir build                                       # make a build directory
$ cd build                                          # enter the build directory
$ cmake .. -DCMAKE_INSTALL_PREFIX=/install/path     # configure cmake ... use $CONDA_PREFIX if you're using Anaconda
$ make -j                                           # compile the libraries librmm.so, libcudf.so ... '-j' will start a parallel job using the number of physical cores available on your system
$ make install                                      # install the libraries librmm.so, libcudf.so to '/install/path'

• To run tests (Optional):

$ make test

• Build and install cffi bindings:

$ make python_cffi                                  # build CFFI bindings for librmm.so, libcudf.so
$ make install_python                               # build & install CFFI python bindings. Depends on cffi package from PyPi or Conda
$ py.test -v                                        # optional, run python tests on low-level python bindings

• 4. Build the cudf python package, in the python folder:

$ cd ../../python
$ python setup.py build_ext --inplace

• You will also need the following environment variables, including $CUDA_HOME.

NUMBAPRO_NVVM=$CUDA_HOME/nvvm/lib64/libnvvm.so
NUMBAPRO_LIBDEVICE=$CUDA_HOME/nvvm/libdevice

• To run Python tests (Optional):

$ py.test -v                                        # run python tests on cudf python bindings

• Finally, install the Python package to your Python path:

$ python setup.py install                           # install cudf python bindings

Done! You are ready to develop for the cuDF OSS project.

Automated Build in Docker Container

A Dockerfile is provided with a preconfigured conda environment for building and installing cuDF from source based off of the master branch.

Prerequisites

• Install nvidia-docker2 for Docker + GPU support
• Verify NVIDIA driver is 396.44 or higher
• Ensure CUDA 9.2+ is installed

Usage

From cudf project root run the following, to build with defaults:

$ docker build --tag cudf .

After the container is built run the container:

$ docker run --runtime=nvidia -it cudf bash

Activate the conda environment cudf to use the newly built cuDF and libcudf libraries:

root@3f689ba9c842:/# source activate cudf
(cudf) root@3f689ba9c842:/# python -c "import cudf"
(cudf) root@3f689ba9c842:/#

Customizing the Build

Several build arguments are available to customize the build process of the container. These are spcified by using the Docker build-arg flag. Below is a list of the available arguments and their purpose:

Build Argument	Default Value	Other Value(s)	Purpose
CUDA_VERSION	9.2	10.0	set CUDA version
LINUX_VERSION	ubuntu16.04	ubuntu18.04	set Ubuntu version
CC & CXX	5	7	set gcc/g++ version; NOTE: gcc7 requires Ubuntu 18.04
CUDF_REPO	This repo	Forks of cuDF	set git URL to use for git clone
CUDF_BRANCH	master	Any branch name	set git branch to checkout of CUDF_REPO
NUMBA_VERSION	0.40.0	Not supported	set numba version
NUMPY_VERSION	1.14.3	Not supported	set numpy version
PANDAS_VERSION	0.20.3	Not supported	set pandas version
PYARROW_VERSION	0.10.0	0.8.0+	set pyarrow version
PYTHON_VERSION	3.5	3.6	set python version

---

Open GPU Data Science

The RAPIDS suite of open source software libraries aim to enable execution of end-to-end data science and analytics pipelines entirely on GPUs. It relies on NVIDIA® CUDA® primitives for low-level compute optimization, but exposing that GPU parallelism and high-bandwidth memory speed through user-friendly Python interfaces.

Apache Arrow on GPU

The GPU version of Apache Arrow is a common API that enables efficient interchange of tabular data between processes running on the GPU. End-to-end computation on the GPU avoids unnecessary copying and converting of data off the GPU, reducing compute time and cost for high-performance analytics common in artificial intelligence workloads. As the name implies, cuDF uses the Apache Arrow columnar data format on the GPU. Currently, a subset of the features in Apache Arrow are supported.