Startup Option

Startup Options¶

Client¶

You need import parser from iflearner.business.homo.argument firstly and then call parser.parse_args, because we provided some common arguments that need to be parsered. If you want to add addtional arguments for yourself, you can call parser.add_argument repeatedly to add them before parser.parse_args has been called. After parsered arguments, you can create your trainer instance base on previous implemented class, and put it with args to train_client.Controller. In the end, you just need call controller.run to run your client.

The list of client default options is as follows:

option	type	describe	default
name	str	name of client	client
epochs	int	number of total epochs to run	10
server	str	the address of connecting aggerating server	localhost:50001
enable-ll	int	enable local training (1 、 0)	0
peers	str	enabled SMPC if the argument had specified. all clients' addresses and use semicolon separate all addresses. First one is your own address.
cert	str	path of server SSL cert. use secure channel to connect to server if not none

Server¶

The list of server options is as follows:

option	type	describe	default
num	int	the number of all clients	0
epochs	int	the total epoch
addr	str	The aggregation server itself listens to the address (used for client connections)	"0.0.0.0:50001"
http_addr	str	Federation training status listening address (for viewing federation training status)	"0.0.0.0:50002"
strategy	str	the aggregation starategy (FedAvg、Scaffold、FedOpt、qFedAvg、FedNova)	FedAvg
strategy_params	dict	specify the params of strategy	{}