/SFrame

SFrame: Scalable tabular and graph data-structures built for out-of-core data analysis and machine learning.

Primary LanguageC++BSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

SFrame

latest release travis build status

Scalable tabular (SFrame, SArray) and graph (SGraph) data-structures built for out-of-core data analysis.

The SFrame package provides the complete implementation of:

  • SFrame
  • SArray
  • SGraph
  • The C++ SDK surface area (gl_sframe, gl_sarray, gl_sgraph)

Introduction

The SFrame contains the open source components GraphLab Create from Turi.

Some documentation to help get started:

For more details on GraphLab Create (including documentation and tutorials) see http://turi.com.

Some of the key features of this package are.

  • A scalable column compressed disk-backed dataframe optimized for machine learning and data science needs.
  • Designed for both tabular (SFrame, SArray) as well as graph data (SGraph)
  • Support for strictly typed columns (int, float, str, datetime), weakly typed columns (schema free lists, dictionaries) as well as specialized types such as Image.
  • Uniform support for missing data.
  • Query optimization and Lazy evaluation.
  • A C++ API (gl_sarray, gl_sframe, gl_sgraph) with direct native access via the C++ SDK.
  • A Python API (SArray, SFrame, SGraph) with an indirect access via an interprocess layer.

License

The SFrame Package is licensed under a BSD license. See license file.

Installation

We release an updated version about once every 1.5 months or so. You can download the most recent version directly from pypi or using pip:

pip install -U sframe

Requirements

SFrame requires a 64-bit operating system.

Operating Systems

  • Mac OS X: 10.8+
  • Linux: Any distribution with GLIBC >= 2.11
  • Ubuntu >= 11.04
  • Debian >= 6
  • RHEL >= 6
  • SLES >= 11
  • Windows (7, 8, 10, Server 2012 R2)

Python

  • Python 2.7.x
  • Pyhton 3.4.x
  • Note: Unfortunately, Python 3.5.x is currently not supported. This is coming soon.

Build Dependencies

To simplify the development process, we use a pre-built dependency toolkit we call dato-deps which prepackages all compile-time dependencies like boost, curl, etc, some with patches which we should contribute back. On Linux, we also prepackage the entire compiler toolchain. On configuration, dato-deps is downloaded and unpacked into the deps/ directory

OS X

At least OS X 10.9. OS X 10.10 preferred.

  • On OS X: Apple XCode 6 Command Line Tools [Required]

    • "clang --version" should report at least "Apple LLVM version 6.1.0 (clang-602.0.53) (based on LLVM 3.6.0svn)"
  • ccache [Optional]

    • Not required, but highly recommended
  • cmake >= 3.2 [Optional]

    • Not required, but highly recommended

Linux

  • *nix build tools: patch, make [Required]

    • Should come with most Mac/Linux systems by default. Recent Ubuntu versions will require installing the build-essential package.
  • zlib [Required]

    • Comes with most Mac/Linux systems by default. Recent Ubuntu version will require the zlib1g-dev package.
  • ccache [Optional]

    • Not required, but highly recommended
  • cmake >= 3.2 [Optional]

    • Not required, but highly recommended

Windows

This is somewhat more involving. See the wiki.

Compiling

After you have done a git pull, cd into the SFrame directory and run the configure script:

cd SFrame
./configure

Running configure will create two sub-directories, release and debug. Select either of these modes/directories and navigate to the oss_src/unity subdirectory:

cd <repo root>/debug/oss_src/unity/python

or

cd <repo root>/release/oss_src/unity/python

Running make will build the project, according to the mode selected.

We recommend using make's parallel build feature to accelerate the compilation process. For instance:

make -j 4

Will perform up to 4 build tasks in parallel. The number of tasks to run in parallel should be roughly equal to the number of CPUs you have.

Running in the Build Tree

To avoid complicated interactions with python installed in your system, we install an entire miniconda toolchain into deps/.

In order to use the dev build that you have just compiled, some environment variables will need to be set. This can be done by sourcing a script. You'll need to pass the script either "debug" or "release" depending on the type of build you have compiled:

source <repo root>/oss_local_scripts/python_env.sh [debug | release ]

This will activate the miniconda toolchain, and you can run python directly and import sframe, etc.

Running Unit Tests

Running Python unit tests

There is a script that makes it easy to run the python unit test. You will just need to call it and pass it your build type (either "debug" or "release).

<repo root>/oss_local_scripts/run_python_test.sh [debug | release]

Alternatively, you can run nosetests directly. For instance, if I am using the debug build directory:

Activate the python environment

source <repo root>/oss_local_scripts/python_env.sh debug

Switch the build output directory and run nosetests

cd <repo root>/debug/src/unity/python
nosetests -s -v sframe/test

Running C++ Units Tests

There are also C++ tests. To compile and run them, do the following:

cd <repo root>/[debug | release]/oss_test
make
ctest

Compilation and Execution Summary

To summarize the above, to just build and run SFrame in the debug build directory:

./configure
cd debug/oss_src/unity
make -j4

# where root is the checked out SFrame directory
source <root>/oss_local_scripts/python_env.sh debug

# import sframe should work here
python

Packaging

To build an egg for your platform, run

<repo root>/oss_local_scripts/make_egg.sh

(See --help for options)

SDK

See: https://github.com/turi-code/GraphLab-Create-SDK