def __init__(self, local_model, logistic_params):
super(HeteroDNNLRGuest, self).__init__(logistic_params)
self.data_shape = local_model.get_encode_dim()
self.index_tracking_list = []
self.local_model = local_model
self.local_model_update_proxy = PlainLocalModelUpdateProxy()
self.local_model_update_proxy.set_model(local_model)
class HeteroDNNLRGuest(HeteroLRGuest):
def __init__(self, local_model, logistic_params):
super(HeteroDNNLRGuest, self).__init__(logistic_params)
self.data_shape = local_model.get_encode_dim()
self.index_tracking_list = []
self.local_model = local_model
self.local_model_update_proxy = PlainLocalModelUpdateProxy()
self.local_model_update_proxy.set_model(local_model)
def set_local_model_update_proxy(
self, local_model_proxy: BaseLocalModelUpdateProxy):
self.local_model_update_proxy = local_model_proxy
self.local_model_update_proxy.set_model(self.local_model)
def transform(self, instance_table):
"""
Extract features from instances
Parameters
----------
:param instance_table: dtable consists of a collection of (index, instance) pairs
:return: instance_table: dtable consists of a collection of (index, instance) pairs,
that each instance holds newly extracted features.
"""
LOGGER.info("@ extract representative features from guest raw input")
# delegate to local_model_proxy for performing the feature extraction task
dtable, self.index_tracking_list = self.local_model_update_proxy.transform(
instance_table)
return dtable
def update_local_model(self, fore_gradient_table, instance_table, coef,
**training_info):
"""
Update local model (i.e., the parameters of local model) based on specified fore_gradient_table, instance_table,
and coef.
Parameters
----------
:param fore_gradient_table: dtable consists of a collection of (index, gradient) pairs
:param instance_table: dtable consists of a collection of (index, instance) pairs
:param coef: the coefficients of the logistic regression model
:param training_info: a dictionary holding information on states of training process
"""
LOGGER.info("@ update guest local model")
# delegate to local_model_proxy for performing the local model update task
training_info["index_tracking_list"] = self.index_tracking_list
training_info["is_host"] = False
self.local_model_update_proxy.update_local_model(
fore_gradient_table, instance_table, coef, **training_info)
def save_model(self, model_table_name, model_namespace):
LOGGER.info("@ save guest model to name/ns" + ", " +
str(model_table_name) + ", " + str(model_namespace))
lr_model_name = model_table_name + "_lr_model"
local_model_name = model_table_name + "_local_model"
super(HeteroDNNLRGuest, self).save_model(lr_model_name,
model_namespace)
save_model_parameters(self.local_model.get_model_parameters(),
local_model_name, model_namespace)
def load_model(self, model_table_name, model_namespace):
LOGGER.info("@ load guest model from name/ns" + ", " +
str(model_table_name) + ", " + str(model_namespace))
lr_model_name = model_table_name + "_lr_model"
local_model_name = model_table_name + "_local_model"
super(HeteroDNNLRGuest, self).load_model(lr_model_name,
model_namespace)
model_parameters = load_model_parameters(local_model_name,
model_namespace)
self.local_model.restore_model(model_parameters)