def __init__(self, in_dim, out_dim, hidden_dim=100):
super(Model, self).__init__()
self.in_dim, self.out_dim = in_dim, out_dim
self.hidden_dim = hidden_dim
self.emb = cls(self.num_words, self.hidden_dim)
self.pool = torchmodels.Linear(
in_features=self.hidden_dim * 3,
out_features=self.hidden_dim
)
self.in_linear = torchmodels.Linear(self.in_dim, self.hidden_dim)
self.out_linear = torchmodels.Linear(self.hidden_dim, self.out_dim)
def __init__(self, *args, hidden_dim=300, num_layers=2,
nonlinear=nonlinear.AbstractNonlinear, **kwargs):
super(MLP, self).__init__(*args, **kwargs)
self.hidden_dim = hidden_dim
self.num_layers = num_layers
self.nonlinear_cls = nonlinear
self.input_layer = torchmodels.Linear(self.input_dim, self.hidden_dim)
self.hidden_layers = torchmodels.Sequential(
*[self.nonlinear_cls(self.hidden_dim, self.hidden_dim)
for _ in range(self.num_layers)]
)
self.output_layer = torchmodels.Linear(self.hidden_dim, self.output_dim)
def __init__(self, *args, hidden_dim=100, **kwargs):
super().__init__(*args, **kwargs)
self.hidden_dim = hidden_dim
self.input_layer = torchmodels.Sequential(
torchmodels.Linear(in_features=self.input_dim,
out_features=self.asv_dim), )
# speaker mask is a [num_speakers x num asv] BoolTensor
# that indicates active asvs for each speaker
self.speaker_asv_mask = nn.Parameter(self._create_speaker_asv_mask(),
requires_grad=False)
def __init__(self, *args, num_layers=1, dropout=0.0, **kwargs):
super(LSTMDecodingRNN, self).__init__(*args, **kwargs)
self.num_layers = num_layers
self.dropout = dropout
self.init_layer_c = torchmodels.Linear(
in_features=self.init_dim,
out_features=self.num_layers * self.hidden_dim
)
self.init_layer_h = torchmodels.Linear(
in_features=self.init_dim,
out_features=self.num_layers * self.hidden_dim
)
self.lstm = nn.LSTM(
input_size=self.input_dim,
hidden_size=self.hidden_dim,
num_layers=self.num_layers,
bidirectional=False,
dropout=self.dropout
)
def __init__(self,
*args,
rnn_dim=200,
decoding_rnn=AbstractDecodingRNN,
**kwargs):
super(RNNSentDecoder, self).__init__(*args, **kwargs)
self.rnn_dim = rnn_dim
self.rnn_cls = decoding_rnn
self.rnn = self.rnn_cls(input_dim=self.word_dim,
init_dim=self.hidden_dim,
hidden_dim=self.rnn_dim)
self.linear = torchmodels.Linear(in_features=self.rnn_dim,
out_features=self.word_dim)
def __init__(self, *args, **kwargs):
super(FunctionalNonlinear, self).__init__(*args, **kwargs)
self.linear = torchmodels.Linear(self.in_dim, self.out_dim)
self.func = self.get_func()