nvmidl.apps.fed_learn package

Subpackages

Submodules

nvmidl.apps.fed_learn.fed_optimizer module

Differentially private optimizers for TensorFlow.

DPAdagradGaussianOptimizer: alias of nvmidl.apps.fed_learn.fed_optimizer.make_gaussian_optimizer_class.<locals>.DPGaussianOptimizerClass

DPAdamGaussianOptimizer: alias of nvmidl.apps.fed_learn.fed_optimizer.make_gaussian_optimizer_class.<locals>.DPGaussianOptimizerClass

DPGradientDescentGaussianOptimizer: alias of nvmidl.apps.fed_learn.fed_optimizer.make_gaussian_optimizer_class.<locals>.DPGaussianOptimizerClass

DPMomentumOptimizer: alias of nvmidl.apps.fed_learn.fed_optimizer.make_gaussian_optimizer_class.<locals>.DPGaussianOptimizerClass

class DPOptimizer(**args)

Bases: ai4med.components.optimizers.optimizer.Optimizer

DP optimizer for large batch

accumulating num_microbatches, adding noise and update the model every num_microbatches steps.

get_optimizer(lr)

This is the required method that an optimizer component must implement.

Parameters: lr – the learning rate
Returns: optimizer function

make_gaussian_optimizer_class(cls): Constructs a DP optimizer with Gaussian averaging of updates.

make_optimizer_class(cls): Constructs a DP optimizer class from an existing one.

nvmidl.apps.fed_learn.fl_conf module

class CrossSiteValidationConfiger(mmar_root: str, wf_config_file_name=None, env_config_file_name=None, log_config_file_name=None, kv_list=None)

Bases: nvmidl.utils.train_conf.EvalConfiger

The cross validation configurer is similar to eval_configer with these notable changes

It doesn’t initialize inferer or evaluator.
A model loader can be passed to get an evaluator.
It uses config_cross_site_validation.json.

finalize_config(config_ctx: dlmed.utils.wfconf.ConfigContext)

get_evaluator()

process_config_element(config_ctx: dlmed.utils.wfconf.ConfigContext, node: dlmed.utils.json_scanner.Node)

class FLClientConfiger(mmar_root: str, wf_config_file_name=None, client_config_file_name=None, env_config_file_name=None, log_config_file_name=None, kv_list=None)

Bases: nvmidl.utils.train_conf.TrainConfiger

finalize_config(config_ctx: dlmed.utils.wfconf.ConfigContext)

process_first_pass(node: dlmed.utils.json_scanner.Node)

process_second_pass(node: dlmed.utils.json_scanner.Node)

start_config(config_ctx: dlmed.utils.wfconf.ConfigContext)

class FLServerConfiger(mmar_root: str, wf_config_file_name=None, server_config_file_name=None, env_config_file_name=None, log_config_file_name=None, kv_list=None)