def __init__(self, conf, vocab, char_vocab, tag_vocab):
super(RNN_CNN, self).__init__()
# Word embedding and initial dropout
self.w_input = word_emb(conf, vocab)
# CHARACTER EMBEDDINGS AND WORD-LEVEL BILSTM
self.use_chars = conf["use_chars"]
if self.use_chars:
self.char_RNN = char_RNN(conf, char_vocab)
in_shape = self.char_RNN.output_size + conf["w_dim"]
else:
in_shape = conf["w_dim"]
# CNN
self.w_cnn_layers = word_CNN(conf["w_cnn_layers"],
in_shape,
1,
keep_dims=True)
output_size = sum([l["filters"] for l in conf["w_cnn_layers"]])
# Output Layer
self.mid_dropout = nn.Dropout(conf["mid_dropout"])
self.n_tags = len(tag_vocab)
self.output = nn.Linear(output_size, self.n_tags)
# CRF Layer
self.use_crf = conf["use_crf"]
if self.use_crf:
self.crf = CRF(self.n_tags, batch_first=True)
# Maybe move to GPU
self.to(self.device)
def __init__(self, conf, vocab, char_vocab):
super(RNN, self).__init__()
# Word embedding and initial dropout
self.w_input = word_emb(conf, vocab)
# CHARACTER EMBEDDINGS AND WORD-LEVEL BILSTM
self.use_chars = conf["use_chars"]
if self.use_chars:
self.char_RNN = char_RNN(conf, char_vocab)
in_shape = self.char_RNN.output_size + conf["w_dim"]
else:
in_shape = conf["w_dim"]
# Main BiLSTM
self.word_RNN = word_RNN(in_shape, conf["w_rnn_out"],
conf["w_rnn_layers"])
output_size = self.word_RNN.output_size
# Output Layer
self.mid_dropout = nn.Dropout(conf["mid_dropout"])
self.output = nn.Linear(output_size, 1)
# reset parameters to initializer
# self.apply(init_all)
# Maybe move to GPU
self.to(self.device)
def __init__(self, conf, vocab, char_vocab, tag_vocab):
super(RNN, self).__init__()
# Word embedding and initial dropout
self.w_input = word_emb(conf, vocab)
# CHARACTER EMBEDDINGS AND WORD-LEVEL BILSTM
self.use_chars = conf["use_chars"]
if self.use_chars:
self.char_RNN = char_RNN(conf, char_vocab)
in_shape = self.char_RNN.output_size + conf["w_dim"]
else:
in_shape = conf["w_dim"]
# Main BiLSTM
self.word_RNN = word_RNN(in_shape, conf["w_rnn_out"],
conf["w_rnn_layers"])
output_size = self.word_RNN.output_size
# Output Layer
self.mid_dropout = nn.Dropout(conf["mid_dropout"])
self.n_tags = len(tag_vocab)
self.output = nn.Linear(output_size, self.n_tags)
# CRF Layer
self.use_crf = conf["use_crf"]
if self.use_crf:
self.crf = CRF(self.n_tags, batch_first=True)
# Initialize weights
# self.apply(init_all)
# Maybe move to GPU
self.to(self.device)
def __init__(self, conf, vocab, char_vocab):
super(RNN_CNN, self).__init__()
# Word embedding and initial dropout
self.w_input = word_emb(conf, vocab)
# CHARACTER EMBEDDINGS AND WORD-LEVEL BILSTM
self.use_chars = conf["use_chars"]
if self.use_chars:
self.char_RNN = char_RNN(conf, char_vocab)
in_shape = self.char_RNN.output_size + conf["w_dim"]
else:
in_shape = conf["w_dim"]
# CNN
self.w_cnn_layers = word_CNN(conf["w_cnn_layers"], in_shape, 1)
output_size = self.w_cnn_layers.output_size
# Output Layer
#self.mid_norm = nn.BatchNorm1d(output_size)
self.mid_dropout = nn.Dropout(conf["mid_dropout"])
self.output = nn.Linear(output_size, 1)
# Maybe move to GPU
self.to(self.device)