secretflow.ml.nn.sl.backend.tensorflow.strategy

Classes:

PYUSLAsyncTFModel	ActorProxy(PYUSLAsyncTFModel) 的别名
PYUSLStateAsyncTFModel	ActorProxy(PYUSLStateAsyncTFModel) 的别名

secretflow.ml.nn.sl.backend.tensorflow.strategy.PYUSLAsyncTFModel

ActorProxy(PYUSLAsyncTFModel) 的别名 Methods:

__init__(*args, **kwargs)	Abstraction device object base class.
base_backward(gradient[, compress])	backward on fusenet
base_forward([stage, compress])	compute hidden embedding :param stage: Which stage of the base forward :param compress: Whether to compress cross device data.
build_dataset_from_builder(*x[, y, s_w, ...])	build tf.data.Dataset
build_dataset_from_numeric(*x[, y, s_w, ...])	build tf.data.Dataset
evaluate(*forward_data[, compress])	Returns the loss value & metrics values for the model in test mode.
export_base_model(model_path[, save_format])
export_fuse_model(model_path[, save_format])
fuse_net(*forward_data[, _num_returns, compress])	Fuses the hidden layer and calculates the reverse gradient only on the side with the label
get_base_losses()
get_base_weights()
get_basenet_output_num()
get_fuse_weights()
get_privacy_spent(step[, orders])	Get accountant of dp mechanism.
get_skip_gradient()
get_stop_training()
init_data()
init_training(callbacks[, epochs, steps, ...])
load_base_model(base_model_path, **kwargs)
load_fuse_model(fuse_model_path, **kwargs)
metrics()
on_epoch_begin(epoch)
on_epoch_end(epoch)
on_train_batch_begin([step])
on_train_batch_end([step])
on_train_begin()
on_train_end()
on_validation(val_logs)
predict(*forward_data[, compress])	Generates output predictions for the input hidden layer features.
reset_metrics()
save_base_model(base_model_path, **kwargs)
save_fuse_model(fuse_model_path, **kwargs)
set_dataset_stage(data_set[, stage, has_y, ...])
set_sample_weight(sample_weight[, stage])
set_steps_per_epoch(steps_per_epoch)
wrap_local_metrics()

secretflow.ml.nn.sl.backend.tensorflow.strategy.PYUSLStateAsyncTFModel

ActorProxy(PYUSLStateAsyncTFModel) 的别名 Methods:

__init__(*args, **kwargs)	Abstraction device object base class.
base_backward(gradient[, compress])	backward on fusenet
base_forward([stage, compress])	compute hidden embedding :param stage: Which stage of the base forward :param compress: Whether to compress cross device data.
build_dataset_from_builder(*x[, y, s_w, ...])	build tf.data.Dataset
build_dataset_from_numeric(*x[, y, s_w, ...])	build tf.data.Dataset
evaluate(*forward_data[, compress])	Returns the loss value & metrics values for the model in test mode.
export_base_model(model_path[, save_format])
export_fuse_model(model_path[, save_format])
fuse_net(*forward_data[, _num_returns, compress])	Fuses the hidden layer and calculates the reverse gradient only on the side with the label
get_base_losses()
get_base_weights()
get_basenet_output_num()
get_fuse_weights()
get_privacy_spent(step[, orders])	Get accountant of dp mechanism.
get_skip_gradient()
get_stop_training()
init_data()
init_training(callbacks[, epochs, steps, ...])
load_base_model(base_model_path, **kwargs)
load_fuse_model(fuse_model_path, **kwargs)
metrics()
on_epoch_begin(epoch)
on_epoch_end(epoch)
on_train_batch_begin([step])
on_train_batch_end([step])
on_train_begin()
on_train_end()
on_validation(val_logs)
predict(*forward_data[, compress])	Generates output predictions for the input hidden layer features.
reset_metrics()
save_base_model(base_model_path, **kwargs)
save_fuse_model(fuse_model_path, **kwargs)
set_dataset_stage(data_set[, stage, has_y, ...])
set_sample_weight(sample_weight[, stage])
set_steps_per_epoch(steps_per_epoch)
wrap_local_metrics()

secretflow.ml.nn.sl.backend.tensorflow.strategy.split_async

Async split learning strategy

Classes:

SLAsyncTFModel(builder_base, builder_fuse, ...)
PYUSLAsyncTFModel	ActorProxy(PYUSLAsyncTFModel) 的别名

class secretflow.ml.nn.sl.backend.tensorflow.strategy.split_async.SLAsyncTFModel(builder_base: Callable[[], Model], builder_fuse: Callable[[], Model], dp_strategy: DPStrategy, compressor: Compressor, base_local_steps: int, fuse_local_steps: int, bound_param: float, random_seed: Optional[int] = None, **kwargs)

基类：SLBaseTFModel

Methods:

__init__(builder_base, builder_fuse, ...[, ...])
base_backward(gradient[, compress])	backward on fusenet

__init__(builder_base: Callable[[], Model], builder_fuse: Callable[[], Model], dp_strategy: DPStrategy, compressor: Compressor, base_local_steps: int, fuse_local_steps: int, bound_param: float, random_seed: Optional[int] = None, **kwargs)

base_backward(gradient, compress: bool = False)

backward on fusenet

参数:

• gradient – gradient of fusenet hidden layer

• compress – Whether to decompress gradient.

secretflow.ml.nn.sl.backend.tensorflow.strategy.split_async.PYUSLAsyncTFModel

ActorProxy(PYUSLAsyncTFModel) 的别名 Methods:

__init__(*args, **kwargs)	Abstraction device object base class.
base_backward(gradient[, compress])	backward on fusenet
base_forward([stage, compress])	compute hidden embedding :param stage: Which stage of the base forward :param compress: Whether to compress cross device data.
build_dataset_from_builder(*x[, y, s_w, ...])	build tf.data.Dataset
build_dataset_from_numeric(*x[, y, s_w, ...])	build tf.data.Dataset
evaluate(*forward_data[, compress])	Returns the loss value & metrics values for the model in test mode.
export_base_model(model_path[, save_format])
export_fuse_model(model_path[, save_format])
fuse_net(*forward_data[, _num_returns, compress])	Fuses the hidden layer and calculates the reverse gradient only on the side with the label
get_base_losses()
get_base_weights()
get_basenet_output_num()
get_fuse_weights()
get_privacy_spent(step[, orders])	Get accountant of dp mechanism.
get_skip_gradient()
get_stop_training()
init_data()
init_training(callbacks[, epochs, steps, ...])
load_base_model(base_model_path, **kwargs)
load_fuse_model(fuse_model_path, **kwargs)
metrics()
on_epoch_begin(epoch)
on_epoch_end(epoch)
on_train_batch_begin([step])
on_train_batch_end([step])
on_train_begin()
on_train_end()
on_validation(val_logs)
predict(*forward_data[, compress])	Generates output predictions for the input hidden layer features.
reset_metrics()
save_base_model(base_model_path, **kwargs)
save_fuse_model(fuse_model_path, **kwargs)
set_dataset_stage(data_set[, stage, has_y, ...])
set_sample_weight(sample_weight[, stage])
set_steps_per_epoch(steps_per_epoch)
wrap_local_metrics()

secretflow.ml.nn.sl.backend.tensorflow.strategy.split_state_async

Stateful async split learning strategy Reference:

[1] Chen, X., Li, J., & Chakrabarti, C. Communication and computation reduction for split learning using asynchronous training[C]. arXiv preprint arXiv:2107.09786, 2021.(https://arxiv.org/abs/2107.09786)

Classes:

SLStateAsyncTFModel(builder_base, ...[, ...])
PYUSLStateAsyncTFModel	ActorProxy(PYUSLStateAsyncTFModel) 的别名

class secretflow.ml.nn.sl.backend.tensorflow.strategy.split_state_async.SLStateAsyncTFModel(builder_base: Callable[[], Model], builder_fuse: Callable[[], Model], dp_strategy: DPStrategy, compressor: Compressor, loss_thres: float = 0, split_steps: int = 1, max_fuse_local_steps: int = 1, random_seed: Optional[int] = None, **kwargs)

基类：SLBaseTFModel

Methods:

__init__(builder_base, builder_fuse, ...[, ...])
get_skip_gradient()

__init__(builder_base: Callable[[], Model], builder_fuse: Callable[[], Model], dp_strategy: DPStrategy, compressor: Compressor, loss_thres: float = 0, split_steps: int = 1, max_fuse_local_steps: int = 1, random_seed: Optional[int] = None, **kwargs)

get_skip_gradient()

secretflow.ml.nn.sl.backend.tensorflow.strategy.split_state_async.PYUSLStateAsyncTFModel

ActorProxy(PYUSLStateAsyncTFModel) 的别名 Methods:

__init__(*args, **kwargs)	Abstraction device object base class.
base_backward(gradient[, compress])	backward on fusenet
base_forward([stage, compress])	compute hidden embedding :param stage: Which stage of the base forward :param compress: Whether to compress cross device data.
build_dataset_from_builder(*x[, y, s_w, ...])	build tf.data.Dataset
build_dataset_from_numeric(*x[, y, s_w, ...])	build tf.data.Dataset
evaluate(*forward_data[, compress])	Returns the loss value & metrics values for the model in test mode.
export_base_model(model_path[, save_format])
export_fuse_model(model_path[, save_format])
fuse_net(*forward_data[, _num_returns, compress])	Fuses the hidden layer and calculates the reverse gradient only on the side with the label
get_base_losses()
get_base_weights()
get_basenet_output_num()
get_fuse_weights()
get_privacy_spent(step[, orders])	Get accountant of dp mechanism.
get_skip_gradient()
get_stop_training()
init_data()
init_training(callbacks[, epochs, steps, ...])
load_base_model(base_model_path, **kwargs)
load_fuse_model(fuse_model_path, **kwargs)
metrics()
on_epoch_begin(epoch)
on_epoch_end(epoch)
on_train_batch_begin([step])
on_train_batch_end([step])
on_train_begin()
on_train_end()
on_validation(val_logs)
predict(*forward_data[, compress])	Generates output predictions for the input hidden layer features.
reset_metrics()
save_base_model(base_model_path, **kwargs)
save_fuse_model(fuse_model_path, **kwargs)
set_dataset_stage(data_set[, stage, has_y, ...])
set_sample_weight(sample_weight[, stage])
set_steps_per_epoch(steps_per_epoch)
wrap_local_metrics()