Skip to content

SyftJob.md

Latest commit

 

History

History
97 lines (72 loc) · 3.36 KB

SyftJob.md

File metadata and controls

97 lines (72 loc) · 3.36 KB

CLASS

SyftJob

Contents

  • Methods
    • start(chargeDetection:wifiDetection:)
    • onReady(execute:)
    • onError(execute:)
    • onRejected(execute:)
public class SyftJob: SyftJobProtocol

Represents a single training cycle done by the client

Methods

start(chargeDetection:wifiDetection:)

public func start(chargeDetection: Bool = true, wifiDetection: Bool = true)

Starts the job executing the following actions:

  1. Meters connection speed to PyGrid
  2. Registers into training cycle on PyGrid
  3. Retrieves cycle and client parameters.
  4. Downloads Plans, Model and Protocols.
  5. Triggers onReady handler
  • Parameters:
    • chargeDetection: Specifies whether to check if device is charging before continuing job execution. Default is true.
    • wifiDetection: Specifies whether to have wifi connection before continuing job execution. Default is true.

Parameters

Name Description
chargeDetection Specifies whether to check if device is charging before continuing job execution. Default is true.
wifiDetection Specifies whether to have wifi connection before continuing job execution. Default is true.

onReady(execute:)

public func onReady(execute: @escaping (_ plan: SyftPlan, _ clientConfig: FederatedClientConfig, _ report: ModelReport) -> Void)

Registers a closure to execute when the job is accepted into a training cycle.

  • Parameter execute: Closure that accepts the training plan (SyftPlan), training configuration (FederatedClientConfig) and reporting closure (ModelReport). All of these objects will be used during training.
  • parameter plan: SyftPlan use this to train your model and generate diffs
  • parameter clientConfig: contains training configuration such as batch size and learning rate.
  • parameter report: closure that accepts diffs as Data and sends them to PyGrid.

Parameters

Name Description
execute Closure that accepts the training plan (SyftPlan), training configuration (FederatedClientConfig) and reporting closure (ModelReport). All of these objects will be used during training.
plan SyftPlan use this to train your model and generate diffs
clientConfig contains training configuration such as batch size and learning rate.
report closure that accepts diffs as Data and sends them to PyGrid.

onError(execute:)

public func onError(execute: @escaping (_ error: Error) -> Void)

Registers a closure to execute whenever an error occurs during training cycle

  • Parameter execute: closure to execute during training cycle
  • parameter error: contains information about error that occurred

Parameters

Name Description
execute closure to execute during training cycle
error contains information about error that occurred

onRejected(execute:)

public func onRejected(execute: @escaping (_ timeout: TimeInterval?) -> Void)

Registers a closure to execute whenever an error occurs during training cycle

  • Parameter execute: closure to execute during training cycle
  • parameter timeout: how long you need to wait before trying again

Parameters

Name Description
execute closure to execute during training cycle
timeout how long you need to wait before trying again