def __init__(self, input_size, hidden_size, memory_size, output_size):
super(Single_LSTM, self).__init__()
#Define model
self.lstm = mod.GD_LSTM(input_size, hidden_size, memory_size,
hidden_size)
self.w_out = Parameter(torch.rand(output_size, hidden_size),
requires_grad=1)
def __init__(self, input_size, hidden_size, memory_size, output_size):
super(Stacked_LSTM, self).__init__()
#Define model
#self.poly = mod.GD_polynet(input_size, hidden_size, hidden_size, hidden_size, None)
self.lstm1 = mod.GD_LSTM(input_size, hidden_size, memory_size,
hidden_size)
self.lstm2 = mod.GD_LSTM(hidden_size, hidden_size, memory_size,
hidden_size)
self.lstm3 = mod.GD_LSTM(hidden_size, hidden_size, memory_size,
hidden_size)
self.w_out1 = Parameter(torch.rand(output_size, hidden_size),
requires_grad=1)
self.w_out2 = Parameter(torch.rand(output_size, hidden_size),
requires_grad=1)
self.w_out3 = Parameter(torch.rand(output_size, hidden_size),
requires_grad=1)
def __init__(self, input_size, hidden_size, memory_size, output_size,
n_vocab):
super(Stacked_LSTM, self).__init__()
#Define model
#self.poly = mod.GD_polynet(input_size, hidden_size, hidden_size, hidden_size, None)
self.embeddings = nn.Embedding(n_vocab + 1, embedding_dim)
self.lstm1 = mod.GD_LSTM(embedding_dim, hidden_size, memory_size,
hidden_size)
self.lstm2 = mod.GD_LSTM(hidden_size, hidden_size, memory_size,
hidden_size)
self.dropout1 = nn.Dropout(0.1)
self.dropout2 = nn.Dropout(0.1)
#self.w_out1 = Parameter(torch.rand(output_size, hidden_size), requires_grad=1)
self.w_out2 = Parameter(torch.rand(output_size, hidden_size),
requires_grad=1)
for param in self.parameters():
# torch.nn.init.xavier_normal(param)
# torch.nn.init.orthogonal(param)
# torch.nn.init.sparse(param, sparsity=0.5)
torch.nn.init.kaiming_normal(param)