def __init__(self, hidden_size, output_size, context_dim, embed, trained_aspect):
super(align_WdeRnnEncoder, self).__init__()
self.hidden_size = hidden_size
self.blstm = nn.LSTM(hidden_size, 300, bidirectional=True, batch_first=True)
self.embedded = nn.Embedding.from_pretrained(embed)
self.aspect_embed = nn.Embedding.from_pretrained(trained_aspect)
self.tanh = nn.Tanh()
self.hidden_layer = nn.Linear(hidden_size * 2, hidden_size)
self.context_input_ = nn.Linear(600, 50)
self.embedding_layers = nn.Linear(0 + hidden_size, output_size)
self.classifier_layer = nn.Linear(output_size, 2)
self.softmax = nn.Softmax(dim=2)
# self.slf_attention = attention.MultiHeadAttention(600, 3)
# self.slf_attention = attention.MultiHeadAttentionDotProduct(3, 600, 300, 300, 0.01)
# self.Position_wise = attention.PositionwiseFeedForward(600, 600, 0.01)
self.min_context = nn.Linear(300, 50)
self.attention = attention.NormalAttention(600, 50, 50)
self.gate = Gate.Gate(300, 50, 50, 300)
self.dropout = nn.Dropout(config.dropout)
def __init__(self, hidden_size, output_size, context_dim, embed):
super(AttentionEncoder, self).__init__()
self.slf_attention = attention.MultiHeadAttentionDotProduct(3, 300, 300, 300)
# self.slf_attention = attention.MultiHeadAttention(300, 300, 300, 3)
self.hidden_size = hidden_size
self.embedded = nn.Embedding.from_pretrained(embed)
self.tanh = nn.Tanh()
self.hidden_layer = nn.Linear(hidden_size, hidden_size)
self.context_input_ = nn.Linear(300, 50)
self.embedding_layers = nn.Linear(0 + hidden_size, output_size)
self.classifier_layer = nn.Linear(output_size, 2)
self.softmax = nn.Softmax(dim=2)
# self.slf_attention = attention.MultiHeadAttention(600, 3)
self.Position_wise = attention.PositionwiseFeedForward(300, 300)
self.attention = attention.NormalAttention(300, 50, 50)
self.gate = Gate.Gate(300, 50, 50, 300)
self.dropout = nn.Dropout(config.dropout)
self.position_enc = nn.Embedding.from_pretrained(
get_sinusoid_encoding_table(301, 300, padding_idx=200),
freeze=True)
def __init__(self, hidden_dim, output_size, context_dim, embed):
super(WdeRnnEncoderFix, self).__init__()
self.hidden_dim = hidden_dim
self.embed_dim = config.embed_dim
self.blstm = nn.LSTM(self.embed_dim,
self.hidden_dim,
bidirectional=True,
batch_first=True)
self.embedded = nn.Embedding.from_pretrained(embed)
# self.aspect_embed = nn.Embedding.from_pretrained(trained_aspect)
self.tanh = nn.Tanh()
self.hidden_layer = nn.Linear(hidden_dim * 2, hidden_dim)
self.context_input_ = nn.Linear(600, 50)
self.embedding_layers = nn.Linear(0 + hidden_dim, output_size)
# self.slf_attention = attention.MultiHeadAttention(600, 3)
# self.slf_attention = attention.MultiHeadAttentionDotProduct(3, 600, 300, 300, 0.01)
# self.Position_wise = attention.PositionwiseFeedForward(600, 600, 0.01)
self.attention = attention.NormalAttention(600, 50, 50)
self.gate = Gate.Gate(300, 50, 50, 300)
self.min_context = nn.Linear(300, 50)
self.aspect_mean = AspectMean(config.maxlen, if_expand=False)