The parameter server is a distributed machine learning framework scaling to industrial-level problems. It is a joint project by CMU SML-Lab, Baidu IDL, and Google.
- Compiler: gcc >= 4.7.2 (prefer >=4.9) or llvm >= 3.4
- OS: tested on Ubuntu 12.10, 13.10, 14.04, RHEL 4U3, Mac OS X 10.9
- Dependent libraries: zeromq, gflags, glogs, gtest, protobuf, zlib, snappy, eigen3. You can use install.sh to build them from sources.
git clone https://github.com/mli/parameter_server
cd parameter_server
git clone https://github.com/mli/third_party
cd third_party && ./install.sh
cd .. && make -j8
Options for building:
- depended libraries are available in somewhere
export THIRD=/usr/local && make -j8
- statical linking
export STATIC=1 && make -j8
- Use
data/rcv1_binary.shto download an example dataset - Run on the local machine:
script/local.sh - Start by
mpirun:script/mpi_root.sh config/rcv1_mpi.conf - Start by a cluster resource manager: In progress.