def setUp(self):
# compose a dummy keras Sequential model
model = tf.keras.Sequential()
model.add(tf.keras.Input(shape=(16, )))
model.add(tf.keras.layers.Dense(8))
model.compile(loss="categorical_crossentropy")
with tempfile.TemporaryDirectory() as tmp_path:
model.save(tmp_path)
keras_model_bytes = _zip_dir_as_bytes(tmp_path)
self.keras_model_param = NNModelParam()
self.keras_model_param.saved_model_bytes = keras_model_bytes
self.keras_model_meta = NNModelMeta()
nn_param = HomoNNParam(config_type="keras",
early_stop="diff",
metrics="Accuracy",
optimizer="SGD",
loss="categorical_crossentropy")
nn_param.check()
self.keras_model_meta.params.CopyFrom(nn_param.generate_pb())
# a dummy pytorch version 0 model
nn_param = HomoNNParam(config_type="pytorch",
early_stop="diff",
metrics="Accuracy",
optimizer={
"optimizer": "Adam",
"lr": 0.05
},
loss="CrossEntropyLoss")
nn_param.check()
nn_define = [{
"layer": "Linear",
"name": "line1",
"type": "normal",
"config": [18, 5]
}, {
"layer": "Relu",
"type": "activate",
"name": "relu"
}, {
"layer": "Linear",
"name": "line2",
"type": "normal",
"config": [5, 4]
}]
self.pytorch_model_param = NNModelParam()
pytorch_nn_model = build_pytorch(nn_define, nn_param.optimizer,
nn_param.loss, nn_param.metrics)
self.pytorch_model_param.saved_model_bytes = pytorch_nn_model.export_model(
)
self.pytorch_model_meta = NNModelMeta()
self.pytorch_model_meta.params.CopyFrom(nn_param.generate_pb())
# a dummy pytorch lightning model
nn_param = HomoNNParam(config_type="pytorch",
early_stop="diff",
metrics="Accuracy",
optimizer={
"optimizer": "Adam",
"lr": 0.05
},
loss="NLLLoss")
nn_param.check()
nn_define = [{
"layer": "Conv2d",
"in_channels": 1,
"out_channels": 10,
"kernel_size": [5, 5]
}, {
"layer": "MaxPool2d",
"kernel_size": 2
}, {
"layer": "ReLU"
}, {
"layer": "Conv2d",
"in_channels": 10,
"out_channels": 20,
"kernel_size": [5, 5]
}, {
"layer": "Dropout2d"
}, {
"layer": "MaxPool2d",
"kernel_size": 2
}, {
"layer": "ReLU"
}, {
"layer": "Flatten"
}, {
"layer": "Linear",
"in_features": 320,
"out_features": 50
}, {
"layer": "ReLU"
}, {
"layer": "Linear",
"in_features": 50,
"out_features": 10
}, {
"layer": "LogSoftmax"
}]
pl_module = FedLightModule(
None,
layers_config=nn_define,
optimizer_config=nn_param.optimizer,
loss_config={"loss": nn_param.loss},
)
pl_trainer = pl.Trainer()
pl_trainer.model = pl_module
self.pl_model_param = NNModelParam(api_version=2)
with tempfile.TemporaryDirectory() as d:
filepath = os.path.join(d, "model.ckpt")
pl_trainer.save_checkpoint(filepath)
with open(filepath, "rb") as f:
self.pl_model_param.saved_model_bytes = f.read()
self.pl_model_meta = NNModelMeta()
self.pl_model_meta.params.CopyFrom(nn_param.generate_pb())
def __init__(self, trans_var):
super().__init__()
self.model_param = HomoNNParam()
self.aggregate_iteration_num = 0
self.transfer_variable = trans_var
def __init__(self):
super().__init__()
self.model_param = HomoNNParam()
self.role = None
def __init__(self, trans_var):
super().__init__()
self.model_param = HomoNNParam()
self.transfer_variable = trans_var
self._api_version = 0