def __init__(self, num_class=2, config=None):
super(RCNN, self).__init__()
main_dir = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(main_dir, "config.yml")
default_config = read_yaml(config_path)
config = (merge_dictb_to_dicta(default_config, config)
if config is not None else default_config)
# Embedding Layer
self.embeddings = nn.Embedding(config.vocab_size, config.embedding_dim)
# Bi-directional LSTM for RCNN
self.lstm = nn.LSTM(
input_size=config.embedding_dim,
hidden_size=config.hidden_dim,
num_layers=config.n_layers,
dropout=config.dropout,
bidirectional=True,
)
self.dropout = nn.Dropout(config.dropout)
# Linear layer to get "convolution output" to be passed to Pooling Layer
self.W = nn.Linear(
config.embedding_dim + 2 * config.hidden_dim,
config.hidden_dim_linear,
)
# Tanh non-linearity
self.tanh = nn.Tanh()
# Fully-Connected Layer
self.fc = nn.Linear(config.hidden_dim_linear, num_class)
def __init__(self, output_dim, hparams):
super().__init__()
main_dir = Path(os.path.dirname(os.path.realpath(__file__))).parent
config_path = os.path.join(main_dir, "config/bilstm.yml")
default_hparams = read_yaml(config_path)
hparams = merge_dictb_to_dicta(default_hparams, hparams)
self.embedding = nn.Embedding(
hparams.vocab_size, hparams.embedding_dim, padding_idx=hparams.pad_idx
)
self.rnn = nn.LSTM(
hparams.embedding_dim,
hparams.hidden_dim,
num_layers=hparams.n_layers,
bidirectional=hparams.bidirectional,
dropout=hparams.dropout,
)
self.fc = nn.Linear(hparams.hidden_dim * 2, output_dim)
self.dropout = nn.Dropout(hparams.dropout)
self.criterion = nn.CrossEntropyLoss()
def __init__(self, output_dim, config=None):
super().__init__()
main_dir = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(main_dir, "config.yml")
default_config = read_yaml(config_path)
config = (merge_dictb_to_dicta(default_config, config)
if config is not None else default_config)
self.embedding = nn.Embedding(config.vocab_size,
config.embedding_dim,
padding_idx=config.pad_idx)
self.convs = nn.ModuleList([
nn.Conv2d(
in_channels=1,
out_channels=config.n_filters,
kernel_size=(fs, config.embedding_dim),
) for fs in config.filter_sizes
])
self.fc = nn.Linear(
len(config.filter_sizes) * config.n_filters, output_dim)
self.dropout = nn.Dropout(config.dropout)
self.criterion = nn.CrossEntropyLoss()
def __init__(self, num_class=2, config=None):
super(Seq2SeqAttention, self).__init__()
main_dir = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(main_dir, "config.yml")
default_config = read_yaml(config_path)
config = (merge_dictb_to_dicta(default_config, config)
if config is not None else default_config)
# Embedding Layer
self.embeddings = nn.Embedding(config.vocab_size, config.embedding_dim)
# Encoder RNN
self.lstm = nn.LSTM(
input_size=config.embedding_dim,
hidden_size=config.hidden_dim,
num_layers=config.n_layers,
bidirectional=config.bidirectional,
)
# Dropout Layer
self.dropout = nn.Dropout(config.dropout)
# Fully-Connected Layer
self.fc = nn.Linear(config.hidden_dim * (1 + config.bidirectional) * 2,
num_class)
# Softmax non-linearity
self.softmax = nn.Softmax()
self.config = config
def __init__(self, output_dim, config=None):
super().__init__()
main_dir = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(main_dir, "config.yml")
default_config = read_yaml(config_path)
config = (
merge_dictb_to_dicta(default_config, config)
if config is not None
else default_config
)
self.embedding = nn.Embedding(
config.vocab_size, config.embedding_dim, padding_idx=config.pad_idx
)
self.rnn = nn.LSTM(
config.embedding_dim,
config.hidden_dim,
num_layers=config.n_layers,
bidirectional=config.bidirectional,
dropout=config.dropout,
)
self.fc = nn.Linear(config.hidden_dim * 2, output_dim)
self.dropout = nn.Dropout(config.dropout)
self.criterion = nn.CrossEntropyLoss()
def __init__(self, num_class=2, config=None):
super(FastText, self).__init__()
main_dir = os.path.dirname(os.path.realpath(__file__))
config_path = os.path.join(main_dir, "config.yml")
default_config = read_yaml(config_path)
config = (
merge_dictb_to_dicta(default_config, config)
if config is not None
else default_config
)
# Embedding Layer
self.embeddings = nn.Embedding(config.vocab_size, config.embedding_dim)
# Hidden Layer
self.fc1 = nn.Linear(config.embedding_dim, config.hidden_dim)
# Output Layer
self.fc2 = nn.Linear(config.hidden_dim, num_class)