Skip to content

Distributed training model(LR, FM) demo using ps-lite. FTRL and SGD Optimization Algorithm.

Notifications You must be signed in to change notification settings

peterzhang2029/xflow_demo

Repository files navigation

1. Introduction

Distributed LR With Parameter Server

2. Install

2.1 build ps-lite

        cd ps-lite
        make -j

        # Add the library file to the PATH.
        # Assumue current dir is $ROOT_DIR.
        export $LIBRARY_PATH=$ROOT_DIR/ps-lite/deps/lib:$LIBRARY_PATH
        export $LD_LIBRARY_PATH=$ROOT_DIR/ps-lite/deps/lib:$LD_LIBRARY_PATH
        export $CPATH=$ROOT_DIR/ps-lite/deps/include:$CPATH
        export $PATH=$ROOT_DIR/ps-lite/deps/bin:$PATH
        export $LIBRARY_PATH=$ROOT_DIR/ps-lite/build:$LIBRARY_PATH
        export $LD_LIBRARY_PATH=$ROOT_DIR/ps-lite/build:$LD_LIBRARY_PATH

2.2 build xflow_demo

        # Back to current dir.
        cd ..
        mkdir build
        cd build 
        cmake ..
        make -j

3.Run

3.1 Local

sh run_ps_local.sh

3.2 Distributed

# on scheduler node 
sh run_ps_dist_scheduler.sh

# on server node
sh run_ps_dist_server.sh

# on worker node
sh run_ps_dist_worker.sh

4. Acknowledge and Reference

  • Referring the design from xflow.

5. AD

欢迎关注公众号:机器学习爱好者

或者扫码:

avatar

学习更多的机器学习源码相关内容

About

Distributed training model(LR, FM) demo using ps-lite. FTRL and SGD Optimization Algorithm.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published