def __init__(self, config):
self.vocab_size = int(config['vocabulary_size'])
self.emb_size = int(config['embedding_dim'])
self.k_max_num = int(config['k_max_num'])
self.hidden_size = int(config['hidden_size'])
self.left_name, self.seq_len1 = config['left_slots'][0]
self.right_name, self.seq_len2 = config['right_slots'][0]
self.task_mode = config['training_mode']
self.emb_layer = layers.EmbeddingEnhancedLayer(self.vocab_size,
self.emb_size,
zero_pad=True,
scale=False)
self.fw_cell = tf.nn.rnn_cell.LSTMCell(num_units=self.hidden_size,
state_is_tuple=True)
self.bw_cell = tf.nn.rnn_cell.LSTMCell(num_units=self.hidden_size,
state_is_tuple=True)
if 'match_mask' in config and config['match_mask'] != 0:
self.match_mask = True
else:
self.match_mask = False
if self.task_mode == "pointwise":
self.n_class = int(config['n_class'])
self.fc2_layer = layers.FCLayer(self.k_max_num, self.n_class)
elif self.task_mode == "pairwise":
self.fc2_layer = layers.FCLayer(self.k_max_num, 1)
else:
logging.error("training mode not supported")
def __init__(self, config):
self.vocab_size = int(config['vocabulary_size'])
self.emb_size = int(config['embedding_dim'])
self.kernel_size = int(config['num_filters'])
self.win_size1 = int(config['window_size_left'])
self.win_size2 = int(config['window_size_right'])
self.dpool_size1 = int(config['dpool_size_left'])
self.dpool_size2 = int(config['dpool_size_right'])
self.hidden_size = int(config['hidden_size'])
self.left_name, self.seq_len1 = config['left_slots'][0]
self.right_name, self.seq_len2 = config['right_slots'][0]
self.emb_layer = layers.EmbeddingEnhancedLayer(self.vocab_size,
self.emb_size,
zero_pad=True,
scale=False)
self.cnn_layer = layers.CNNDynamicPoolingLayer(
self.seq_len1, self.seq_len2, self.win_size1, self.win_size2,
self.dpool_size1, self.dpool_size2, self.kernel_size)
self.relu_layer = layers.ReluLayer()
self.tanh_layer = layers.TanhLayer()
if 'match_mask' in config and config['match_mask'] != 0:
self.match_mask = True
else:
self.match_mask = False
self.fc1_layer = layers.FCLayer(
self.kernel_size * self.dpool_size1 * self.dpool_size2,
self.hidden_size)
self.n_class = int(config['n_class'])
self.fc2_layer = layers.FCLayer(self.hidden_size, self.n_class)
def __init__(self, config):
self.vocab_size = int(config['vocabulary_size'])
self.emb_size = int(config['embedding_dim'])
self.rnn_hidden_size = int(config['rnn_hidden_size'])
self.hidden_size = int(config['hidden_size'])
self.left_name, self.seq_len1 = config['left_slots'][0]
self.right_name, self.seq_len2 = config['right_slots'][0]
self.task_mode = config['training_mode']
self.emb_layer = layers.EmbeddingEnhancedLayer(self.vocab_size,
self.emb_size,
zero_pad=True,
scale=False)
self.rnn = layers.LSTMLayer(self.rnn_hidden_size)
self.extract = layers.ExtractLastLayer()
if self.task_mode == "pointwise":
self.n_class = int(config['n_class'])
self.fc1_layer = layers.FCLayer(self.rnn_hidden_size * 2,
self.hidden_size)
self.fc2_layer = layers.FCLayer(self.hidden_size, self.n_class)
elif self.task_mode == "pairwise":
self.fc1_layer = layers.FCLayer(self.rnn_hidden_size * 1,
self.hidden_size)
self.cos_layer = layers.CosineLayer()
else:
logging.error("training mode not supported")
def __init__(self, config):
self.vocab_size = int(config['vocabulary_size'])
self.emb_size = int(config['embedding_dim'])
self.kernel_size = int(config['num_filters'])
self.win_size1 = int(config['window_size_left'])
self.win_size2 = int(config['window_size_right'])
self.dpool_size1 = int(config['dpool_size_left'])
self.dpool_size2 = int(config['dpool_size_right'])
self.hidden_size = int(config['hidden_size'])
self.left_name, self.seq_len1 = config['left_slots'][0]
self.right_name, self.seq_len2 = config['right_slots'][0]
self.task_mode = config['training_mode']
self.emb_layer = layers.EmbeddingEnhancedLayer(self.vocab_size,
self.emb_size,
zero_pad=True,
scale=False) #加强版本的词向量
self.cnn_layer = layers.CNNDynamicPoolingLayer(
self.seq_len1, self.seq_len2, self.win_size1, self.win_size2,
self.dpool_size1, self.dpool_size2, self.kernel_size)
self.relu_layer = layers.ReluLayer()
self.tanh_layer = layers.TanhLayer()
if 'match_mask' in config and config['match_mask'] != 0:
self.match_mask = True #是否考虑填充区域
else:
self.match_mask = False
self.fc1_layer = layers.FCLayer(
self.kernel_size * self.dpool_size1 * self.dpool_size2,
self.hidden_size)
if self.task_mode == "pointwise":
self.n_class = int(config['n_class'])
self.fc2_layer = layers.FCLayer(self.hidden_size, self.n_class)
elif self.task_mode == "pairwise":
self.fc2_layer = layers.FCLayer(self.hidden_size, 1)
else:
logging.error("training mode not supported")
def __init__(self, config):
self.vocab_size = int(config['vocabulary_size'])
self.emb_size = int(config['embedding_dim'])
self.rnn_hidden_size = int(config['rnn_hidden_size'])
self.hidden_size = int(config['hidden_size'])
self.left_name, self.seq_len1 = config['left_slots'][0]
self.right_name, self.seq_len2 = config['right_slots'][0]
self.emb_layer = layers.EmbeddingEnhancedLayer(self.vocab_size,
self.emb_size,
zero_pad=True,
scale=False)
self.rnn = layers.LSTMLayer(self.rnn_hidden_size)
self.extract = layers.ExtractLastLayer()
self.n_class = int(config['n_class'])
self.fc1_layer = layers.FCLayer(self.rnn_hidden_size * 2,
self.hidden_size)
self.fc2_layer = layers.FCLayer(self.hidden_size, self.n_class)