def __init__(self, hetero_nn_param):
super(HeteroNNKerasGuestModel, self).__init__()
self.bottom_model = None
self.interactive_model = None
self.top_model = None
self.bottom_nn_define = None
self.top_nn_define = None
self.interactive_layer_define = None
self.config_type = None
self.optimizer = None
self.loss = None
self.metrics = None
self.hetero_nn_param = None
self.transfer_variable = None
self.model_builder = None
self.bottom_model_input_shape = 0
self.top_model_input_shape = None
self.batch_size = None
self.is_empty = False
self.set_nn_meta(hetero_nn_param)
self.model_builder = nn_model.get_nn_builder(
config_type=self.config_type)
self.data_converter = KerasSequenceDataConverter()
self.selector = SelectorFactory.get_selector(
hetero_nn_param.selector_param.method,
hetero_nn_param.selector_param.selective_size,
beta=hetero_nn_param.selector_param.beta,
random_rate=hetero_nn_param.selector_param.random_state,
min_prob=hetero_nn_param.selector_param.min_prob)
def __init__(self, hetero_nn_param):
super(HeteroNNKerasGuestModel, self).__init__()
self.bottom_model = None
self.interactive_model = None
self.top_model = None
self.bottom_nn_define = None
self.top_nn_define = None
self.interactive_layer_define = None
self.config_type = None
self.optimizer = None
self.loss = None
self.metrics = None
self.hetero_nn_param = None
self.transfer_variable = None
self.model_builder = None
self.bottom_model_input_shape = 0
self.top_model_input_shape = None
self.is_empty = False
self.set_nn_meta(hetero_nn_param)
self.model_builder = nn_model.get_nn_builder(
config_type=self.config_type)
self.data_converter = KerasSequenceDataConverter()
def _init_model(self, param):
super(HomoNNClient, self)._init_model(param)
self.batch_size = param.batch_size
self.aggregate_every_n_epoch = 1
self.nn_define = param.nn_define
self.config_type = param.config_type
self.optimizer = param.optimizer
self.loss = param.loss
self.metrics = param.metrics
self.data_converter = nn_model.get_data_converter(self.config_type)
self.model_builder = nn_model.get_nn_builder(config_type=self.config_type)
def _init_model(self, param: HomoNNParam):
super()._init_model(param=param)
self.batch_size = param.batch_size
self.aggregate_every_n_epoch = param.aggregate_every_n_epoch
self.nn_define = param.nn_define
self.config_type = param.config_type
self.optimizer = param.optimizer
self.loss = param.loss
self.metrics = param.metrics
self.encode_label = param.encode_label
self.data_converter = nn_model.get_data_converter(self.config_type)
self.model_builder = nn_model.get_nn_builder(config_type=self.config_type)
def initialize_nn(self, input_shape):
"""
initializing nn weights
"""
loss = "keep_predict_loss"
self.nn_builder = get_nn_builder(config_type=self.config_type)
self.nn: NNModel = self.nn_builder(loss=loss, nn_define=self.nn_define, optimizer=self.optimizer, metrics=None,
input_shape=input_shape)
LOGGER.debug('printing nn layers structure')
for layer in self.nn._model.layers:
LOGGER.debug('input shape {}, output shape {}'.format(layer.input_shape, layer.output_shape))
def client_set_params(self, param):
self.nn_model = None
self._summary = dict(loss_history=[], is_converged=False)
self._header = []
self._label_align_mapping = None
self.param = param
self.enable_secure_aggregate = param.secure_aggregate
self.max_aggregate_iteration_num = param.max_iter
self.batch_size = param.batch_size
self.aggregate_every_n_epoch = param.aggregate_every_n_epoch
self.nn_define = param.nn_define
self.config_type = param.config_type
self.optimizer = param.optimizer
self.loss = param.loss
self.metrics = param.metrics
self.encode_label = param.encode_label
self.data_converter = nn_model.get_data_converter(self.config_type)
self.model_builder = nn_model.get_nn_builder(config_type=self.config_type)