tensorflow-federated - TFF : Modify the process that includes an initialization and iterated computation

Question

我可以在这段代码中写什么:

def initialize_fn():
  ...

Answer 1

iterative_process.initialize() 返回的state 对象通常是一个 Python 容器(tuple, collections.OrderedDict 等)包含 numpy 数组。要看到这一点，请在初始化后尝试 print(state)。

也许在 initialze() 之后和在 next() 循环之前用从外部文件加载的值替换值:

def load_my_weights(filename):
  # read from external file, create numpy arrays

iterative_process = IterativeProcess(initialize_fn, next_fn)
state = iterative_process.initialize()
state['model'] = load_my_weights(filename)  # assumes 'model' is a key in state.
for round in range(num_rounds):
  state = iterative_process.next(state)