ShuhanYan / BigDL-old

Deep learning library for Apache Spark

#BigDL

A scalable deep learning library for Apache Spark.

Here's the summary of core features:

• a powerful N-dimensional array
• lots of math and data manipulating operations
• rich neural network layers
• effecient distributed numeric optimization routines on Apache Spark
• powered by MKL and MKL DNN, fast and optmized on Intel hardware platforms

##How to build ###Linux (only intel64 architecture) ####Build mvn clean package -DskipTests
Module native is skipped by default. If you want to build native module, see the Full Build.

####Full build

1. Download Intel MKL and install it in your linux box
2. Prepare MKL build environment
   source PATH_TO_MKL/bin/mklvars.sh intel64
   If Intel MKL doesn't install to default path /opt/intel, please link your libiomp5.so to project.
   ln -sf PATH_TO_INTEL_HOME/lib/intel64/libiomp5.so BIGDL_HOME/native/jni/src/main/resources/intel64
3. Full build
   mvn clean package -DskipTests -P full-build

##Example

• MNIST example
• Cifar10 example
• Imagenet example