def __init__(self,
seed,
lr,
seq_len,
num_classes,
n_hidden,
emb_arr=None,
cfg=None):
self.seq_len = seq_len
self.num_classes = num_classes
self.n_hidden = n_hidden
_, self.indd, vocab = get_word_emb_arr(VOCAB_DIR)
self.vocab_size = len(vocab)
print('vocab_size:', self.vocab_size)
self.model_name = os.path.abspath(__file__)
if emb_arr:
self.emb_arr = emb_arr
if cfg.fedprox:
super(ClientModel,
self).__init__(seed,
lr,
optimizer=PerturbedGradientDescent(
lr, cfg.fedprox_mu))
else:
super(ClientModel, self).__init__(seed, lr)
def __init__(self,
seed,
lr,
seq_len,
n_hidden,
num_layers,
keep_prob=1.0,
max_grad_norm=5,
init_scale=0.1,
cfg=None):
self.seq_len = seq_len
self.n_hidden = n_hidden
self.num_layers = num_layers
self.keep_prob = keep_prob
self.max_grad_norm = max_grad_norm
self.model_name = os.path.abspath(__file__)
# initialize vocabulary
self.vocab, self.vocab_size, self.unk_symbol, self.pad_symbol = self.load_vocab(
)
print('vocab_size: {}'.format(self.vocab_size))
self.initializer = tf.random_uniform_initializer(
-init_scale, init_scale)
if cfg.fedprox:
super(ClientModel,
self).__init__(seed,
lr,
optimizer=PerturbedGradientDescent(
lr, cfg.fedprox_mu))
else:
super(ClientModel, self).__init__(seed, lr)
def __init__(self, seed, lr, num_classes, cfg=None):
self.num_classes = num_classes
self.model_name = os.path.abspath(__file__)
if cfg.fedprox:
super(ClientModel,
self).__init__(seed,
lr,
optimizer=PerturbedGradientDescent(
lr, cfg.fedprox_mu))
else:
super(ClientModel, self).__init__(seed, lr)