Distributed-Mnist

Distributed version of mnist tensorflow training

Instructions

1. Simple Usage:

• for Parameter Server 0, python distributed_mnist.py --job_name='ps' --task_index=0
• for Parameter Server 1, python distributed_mnist.py --job_name='ps' --task_index=1
• for Client 0 , python distributed-mnist.py --job_name='worker' --task_index=0
• for Client 1 , python distributed-mnist.py --job_name='worker' --task_index=1

Remember to edit flag ps_hosts/worker_hosts inside the python sript before running th program
Inside the script, flag num_workers/num_parameter_servers should be consistent with length of flag ps_hosts/worker_hosts

2. Optional Arguments:

• --ps_hosts: list of [hostname:port] for ps jobs
• --worker_hosts: list of [hostname:port] for ps jobs
• --num_workers: length of workers_hosts
• --num_parameter_servers: length of ps_hosts
• --train_steps: number of training steps to perform
• --batch_size: training batch size

Also see python distributed_mnist.py -h for more info

Dependency

• TensorFlow