def __init__(self, ernie, crf_lr=100):
super().__init__()
self.num_classes = ernie.num_classes
self.ernie = ernie # allow ernie to be config
self.crf = LinearChainCrf(
self.num_classes, crf_lr=crf_lr, with_start_stop_tag=False)
self.crf_loss = LinearChainCrfLoss(self.crf)
self.viterbi_decoder = ViterbiDecoder(
self.crf.transitions, with_start_stop_tag=False)
def __init__(self,
word_emb_dim,
hidden_size,
vocab_size,
num_labels,
emb_lr=2.0,
crf_lr=0.2,
with_start_stop_tag=True):
super(BiGruCrf, self).__init__()
self.word_emb_dim = word_emb_dim
self.vocab_size = vocab_size
self.num_labels = num_labels
self.hidden_size = hidden_size
self.emb_lr = emb_lr
self.crf_lr = crf_lr
self.init_bound = 0.1
self.word_embedding = nn.Embedding(
num_embeddings=self.vocab_size,
embedding_dim=self.word_emb_dim,
weight_attr=paddle.ParamAttr(learning_rate=self.emb_lr,
initializer=nn.initializer.Uniform(
low=-self.init_bound,
high=self.init_bound)))
self.gru = nn.GRU(
input_size=self.word_emb_dim,
hidden_size=self.hidden_size,
num_layers=2,
direction='bidirectional',
weight_ih_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(low=-self.init_bound,
high=self.init_bound),
regularizer=paddle.regularizer.L2Decay(coeff=1e-4)),
weight_hh_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(low=-self.init_bound,
high=self.init_bound),
regularizer=paddle.regularizer.L2Decay(coeff=1e-4)))
self.fc = nn.Linear(
in_features=self.hidden_size * 2,
out_features=self.num_labels + 2 \
if with_start_stop_tag else self.num_labels,
weight_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(
low=-self.init_bound, high=self.init_bound),
regularizer=paddle.regularizer.L2Decay(coeff=1e-4)))
self.crf = LinearChainCrf(self.num_labels, self.crf_lr,
with_start_stop_tag)
self.crf_loss = LinearChainCrfLoss(self.crf)
self.viterbi_decoder = ViterbiDecoder(self.crf.transitions,
with_start_stop_tag)
def __init__(self,
emb_size,
hidden_size,
word_num,
label_num,
use_w2v_emb=False):
super(BiGRUWithCRF, self).__init__()
if use_w2v_emb:
self.word_emb = TokenEmbedding(
extended_vocab_path='./data/word.dic', unknown_token='OOV')
else:
self.word_emb = nn.Embedding(word_num, emb_size)
self.gru = nn.GRU(emb_size,
hidden_size,
num_layers=2,
direction='bidirectional')
self.fc = nn.Linear(hidden_size * 2, label_num + 2) # BOS EOS
self.crf = LinearChainCrf(label_num)
self.crf_loss = LinearChainCrfLoss(self.crf)
self.viterbi_decoder = ViterbiDecoder(self.crf.transitions)
def __init__(self, skep, num_classes):
super().__init__()
self.num_classes = num_classes
self.skep = skep # allow skep to be config
gru_hidden_size = 128
self.gru = nn.GRU(self.skep.config["hidden_size"],
gru_hidden_size,
num_layers=2,
direction='bidirect')
self.fc = nn.Linear(
gru_hidden_size * 2,
self.num_classes,
weight_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(low=-0.1, high=0.1),
regularizer=paddle.regularizer.L2Decay(coeff=1e-4)))
self.crf = LinearChainCrf(self.num_classes,
crf_lr=0.2,
with_start_stop_tag=False)
self.crf_loss = LinearChainCrfLoss(self.crf)
self.viterbi_decoder = ViterbiDecoder(self.crf.transitions,
with_start_stop_tag=False)
def __init__(self,
ernie_ctm,
num_tag,
num_cls_label,
crf_lr=100,
ignore_index=0):
super(ErnieCtmWordtagModel, self).__init__()
self.num_tag = num_tag
self.num_cls_label = num_cls_label
self.ernie_ctm = ernie_ctm
self.tag_classifier = nn.Linear(self.ernie_ctm.config["hidden_size"],
self.num_tag)
self.sent_classifier = nn.Linear(self.ernie_ctm.config["hidden_size"],
self.num_cls_label)
self.crf = LinearChainCrf(
self.num_tag, crf_lr, with_start_stop_tag=False)
self.crf_loss = LinearChainCrfLoss(self.crf)
self.viterbi_decoder = ViterbiDecoder(
self.crf.transitions, with_start_stop_tag=False)
self.ignore_index = ignore_index
self.apply(self.init_weights)