def __init__(self, classes_num, hidden_size=128, BERT_out_dropout=0.5, BiLSTM_dropout=0.1):
super(Encoder, self).__init__()
self.hidden_size = hidden_size
self.zeros = T.tensor(0.0).to(device)
self.neg_inf = T.tensor(-2.0**32).to(device)
self.BERT = BertModel.from_pretrained('../Pre_trained_BERT/',
output_hidden_states=True,
output_attentions=False)
self.dropout = nn.Dropout(BERT_out_dropout)
self.layer_fusion = capsule_fusion(D=768,
n_in=6, n_out=768,
in_dim=128, out_dim=1,
depth_encoding=True)
self.BiLSTM = BiLSTM_Encoder(D=768, hidden_size=hidden_size, dropout=BiLSTM_dropout)
self.hidden_fusion = capsule_fusion(D=2*hidden_size,
n_in=1, n_out=2*hidden_size,
in_dim=128, out_dim=1,
depth_encoding=False)
self.linear1 = Linear(2*hidden_size, 300)
self.linear2 = Linear(300, 1)
self.linear3 = Linear(300, classes_num)
def __init__(self,
classes_num,
hidden_size=128,
BERT_out_dropout=0.5,
BiLSTM_dropout=0.1):
super(Encoder, self).__init__()
self.BERT = BertModel.from_pretrained('../Pre_trained_BERT/',
output_hidden_states=True,
output_attentions=False)
self.dropout = nn.Dropout(BERT_out_dropout)
self.linear1 = Linear(768, 300)
self.linear2 = Linear(300, 1)
self.linear3 = Linear(300, classes_num)