def __init__(self, opt, emb_matrix=None):
super(PositionAwareRNN, self).__init__()
self.drop = nn.Dropout(opt['dropout'])
self.emb = nn.Embedding(opt['vocab_size'], opt['emb_dim'], padding_idx=constant.PAD_ID)
if opt['pos_dim'] > 0:
self.pos_emb = nn.Embedding(len(constant.POS_TO_ID), opt['pos_dim'],
padding_idx=constant.PAD_ID)
if opt['ner_dim'] > 0:
self.ner_emb = nn.Embedding(len(constant.NER_TO_ID), opt['ner_dim'],
padding_idx=constant.PAD_ID)
input_size = opt['emb_dim'] + opt['pos_dim'] + opt['ner_dim']
self.rnn = nn.LSTM(input_size, opt['hidden_dim'], opt['num_layers'], batch_first=True,\
dropout=opt['dropout'])
self.linear = nn.Linear(opt['hidden_dim'], opt['num_class'])
if opt['attn']:
self.attn_layer = layers.PositionAwareAttention(opt['hidden_dim'],
opt['hidden_dim'], 2*opt['pe_dim'], opt['attn_dim'])
self.pe_emb = nn.Embedding(constant.MAX_LEN * 2 + 1, opt['pe_dim'])
self.opt = opt
self.topn = self.opt.get('topn', 1e10)
self.use_cuda = opt['cuda']
self.emb_matrix = emb_matrix
self.init_weights()
def __init__(self, opt, emb_matrix=None, ent_emb_matrix=None):
super(PositionAwareRNN, self).__init__()
self.drop = nn.Dropout(opt["dropout"])
self.emb = nn.Embedding(opt["vocab_size"],
opt["emb_dim"],
padding_idx=constant.PAD_ID)
self.ent_emb = None
if ("ent_emb_dim" in opt and opt["ent_emb_dim"] > 0
and "ent_vocab_size" in opt and opt["ent_vocab_size"] > 0):
self.ent_emb = nn.Embedding(opt["ent_vocab_size"],
opt["ent_emb_dim"],
padding_idx=constant.PAD_ID)
if opt["pos_dim"] > 0:
self.pos_emb = nn.Embedding(len(constant.POS_TO_ID),
opt["pos_dim"],
padding_idx=constant.PAD_ID)
if opt["ner_dim"] > 0:
self.ner_emb = nn.Embedding(len(constant.NER_TO_ID),
opt["ner_dim"],
padding_idx=constant.PAD_ID)
input_size = (opt["emb_dim"] + opt["ent_emb_dim"] + opt["pos_dim"] +
opt["ner_dim"])
self.rnn = nn.LSTM(
input_size,
opt["hidden_dim"],
opt["num_layers"],
batch_first=True,
dropout=opt["dropout"],
)
self.linear = nn.Linear(opt["hidden_dim"], opt["num_class"])
if opt["attn"]:
self.attn_layer = layers.PositionAwareAttention(
opt["hidden_dim"], opt["hidden_dim"], 2 * opt["pe_dim"],
opt["attn_dim"])
self.pe_emb = nn.Embedding(constant.MAX_LEN * 2 + 1, opt["pe_dim"])
self.opt = opt
self.topn = self.opt.get("topn", 1e10)
self.emb_topn = self.opt.get("emb_topn", 0)
self.use_cuda = opt["cuda"]
self.emb_matrix = emb_matrix
self.ent_emb_matrix = ent_emb_matrix
self.init_weights()
def __init__(self, opt, emb_matrix=None):
super(PositionAwareRNN, self).__init__()
self.object_indices = opt['object_indices']
self.drop = nn.Dropout(opt['dropout'])
self.emb = nn.Embedding(opt['vocab_size'], opt['emb_dim'], padding_idx=constant.PAD_ID)
if opt['pos_dim'] > 0:
self.pos_emb = nn.Embedding(len(constant.POS_TO_ID), opt['pos_dim'],
padding_idx=constant.PAD_ID)
if opt['ner_dim'] > 0:
self.ner_emb = nn.Embedding(len(constant.NER_TO_ID), opt['ner_dim'],
padding_idx=constant.PAD_ID)
input_size = opt['emb_dim'] + opt['pos_dim'] + opt['ner_dim']
self.rnn = nn.LSTM(input_size, opt['hidden_dim'], opt['num_layers'], batch_first=True,\
dropout=opt['dropout'])
self.linear = nn.Linear(opt['hidden_dim'], opt['num_class'])
if opt['attn']:
self.attn_layer = layers.PositionAwareAttention(opt['hidden_dim'],
opt['hidden_dim'], 2*opt['pe_dim'], opt['attn_dim'])
self.pe_emb = nn.Embedding(constant.MAX_LEN * 2 + 1, opt['pe_dim'])
# LP Model
if opt.get('link_prediction', None) is not None:
link_prediction_cfg = opt.get('link_prediction', None)['model']
self.rel_emb = nn.Embedding(opt['num_relations'], link_prediction_cfg['rel_emb_dim'])
self.register_parameter('rel_bias', torch.nn.Parameter(torch.zeros((opt['num_relations']))))
self.object_indices = torch.from_numpy(np.array(self.object_indices))
if opt['cuda']:
self.object_indices = self.object_indices.cuda()
self.lp_model = initialize_link_prediction_model(link_prediction_cfg)
self.opt = opt
self.topn = self.opt.get('topn', 1e10)
self.use_cuda = opt['cuda']
self.emb_matrix = emb_matrix
self.init_weights()