def Parameter(
*size,
initializer=uniform(bound=0.1),
):
m = nn.Parameter(torch.zeros(*size))
initializer(m)
return m
def Embedding(num_embeddings,
embedding_dim,
padding_idx=None,
initializer=uniform(bound=0.1)):
m = nn.Embedding(num_embeddings, embedding_dim, padding_idx=padding_idx)
if initializer is not None:
initializer(m.weight)
if padding_idx is not None:
m.weight.data[padding_idx, :].fill_(0.)
return m
def LSTMCell(
input_size,
hidden_size,
bias=True,
initializer=uniform(bound=0.1),
):
m = nn.LSTMCell(input_size, hidden_size, bias)
for name, param in m.named_parameters():
if 'weight' in name or 'bias' in name:
initializer(param)
return m
def Linear(
in_dim,
out_dim,
bias=True,
weight_initializer=uniform(bound=0.1),
bias_initializer=constant(value=0.),
):
m = nn.Linear(in_features=in_dim, out_features=out_dim, bias=bias)
weight_initializer(m.weight)
if bias:
bias_initializer(m.bias)
return m
def LSTM(
input_size,
hidden_size,
num_layers=1,
bias=True,
batch_first=False,
dropout=0.,
bidirectional=False,
initializer=uniform(bound=0.1),
):
"""
input_size: The number of expected features in the input `x`
hidden_size: The number of features in the hidden state `h`
num_layers: Number of recurrent layers. E.g., setting ``num_layers=2``
would mean stacking two LSTMs together to form a `stacked LSTM`,
with the second LSTM taking in outputs of the first LSTM and
computing the final results. Default: 1
bias: If ``False``, then the layer does not use bias weights `b_ih` and `b_hh`.
Default: ``True``
batch_first: If ``True``, then the input and output tensors are provided
as (batch, seq, feature). Default: ``False``
dropout: If non-zero, introduces a `Dropout` layer on the outputs of each
LSTM layer except the last layer, with dropout probability equal to
:attr:`dropout`. Default: 0
bidirectional: If ``True``, becomes a bidirectional LSTM. Default: ``False``
"""
m = nn.LSTM(input_size,
hidden_size,
num_layers=num_layers,
bias=bias,
batch_first=batch_first,
dropout=dropout,
bidirectional=bidirectional)
for p in m.parameters():
initializer(p)
return m