TrainCfg

import com.mindspore.config.TrainCfg;

Configuration parameters used for model training on the device.

Public Member Functions

function	Supported At Cloud-side Inference	Supported At Device-side Inference
boolean init()	✕	√
boolean init(String loss_name)	✕	√
void free()	✕	√
boolean addMixPrecisionCfg(boolean dynamicLossScale, float lossScale, int thresholdIterNum)	✕	√
long getTrainCfgPtr()	✕	√

init

public boolean init()

Init train config.

• Returns

  init status.

public boolean init(String loss_name)

Init train config specified loss name.

• Parameters

• loss_name: loss_name loss name used for split inference and train part.

• Returns

  初始化状态。

free

public void free()

Free train config.

addMixPrecisionCfg

public boolean addMixPrecisionCfg(boolean dynamicLossScale, float lossScale, int thresholdIterNum)

Add mix precision config to train config.

• Parameters

• dynamicLossScale: dynamicLossScale if dynamic or fix loss scale factor.

• lossScale: loss scale factor.

• thresholdIterNum: thresholdIterNum a threshold for modifying loss scale when dynamic loss scale is enabled.

• Returns

  add status.

getTrainCfgPtr

public long getTrainCfgPtr()

Get train config pointer.

• Returns

  train config pointer.