Building MLeap From Source
MLeap is hosted on Github as a public project. Building MLeap is very straightforward and has very few dependencies. Here are instructions for building from source.
Install Simple Build Tool, which is used to build many Scala projects.
Compiling Core MLeap
The core of MLeap includes every submodule except for Tensorflow integration. We do not include Tensorflow in the core build because its dependencies can be difficult to install.
Clone the Github Repository
git clone https://github.com/combust/mleap.git cd mleap
Initialize Git Submodules
MLeap depends on a git submodule for all of the protobuf definitions, so we need to initialize and update them.
git submodule init git submodule update
There are many submodules that make up the MLeap project. All of them are aggregated to the root SBT project. This means running a command from SBT will cause the command to be run on all subprojects.
Run the Tests
MLeap has exensive testing, including full parity tests between MLeap and Spark transformers.
Compiling Tensorflow Support
mleap-tensorflow submodule does not happen
automatically. Instead, we first need to compile Tensorflow and install
the Tensorflow Java jar to our local maven2 repository.
Tensorflow has a great set of instructions for compiling and installing.
Build Tensorflow MLeap Module
Run Tensorflow Integration Tests
TENSORFLOW_JNI=/path/to/tensorflow/library/folder/java sbt mleap-tensorflow/test