def __init__(self, config):
super(XLNetForXMLC, self).__init__(config)
self.num_labels = config.num_labels
self.transformer = XLNetModel(config)
self.sequence_summary = SequenceSummary(config)
self.logits_proj = nn.Linear(config.d_model, config.num_labels)
self.init_weights()
def __init__(self, config):
super(XlnetForMultiLable, self).__init__(config)
self.transformer = XLNetModel(config)
self.sequence_summary = SequenceSummary(config)
self.classifier = nn.Linear(config.hidden_size, config.num_labels)
self.apply(self._init_weights)
def __init__(self, config):
super().__init__(config)
self.attn_type = config.attn_type
self.same_length = config.same_length
self.transformer = XLNetModel(config)
self.lm_loss = nn.Linear(config.d_model, config.n_token, bias=True)
self.init_weights()
self.tie_weights()
def __init__(self, config, weight=None):
super(XLNetForSequenceClassification, self).__init__(config)
self.num_labels = config.num_labels
self.weight = weight
self.transformer = XLNetModel(config)
self.sequence_summary = SequenceSummary(config)
self.logits_proj = nn.Linear(config.d_model, config.num_labels)
self.init_weights()
def __init__(self, config):
super(XLNetForXMLC, self).__init__(config)
self.num_labels = config.num_labels
self.transformer = XLNetModel(config)
self.sequence_summary = SequenceSummary(config)
self.logits_proj = nn.Linear(config.d_model, config.num_labels)
self.loss_fct = HingeLoss(margin=1.0, squared=True)
self.init_weights()
def __init__(self, config):
super(XLNetCalculator, self).__init__(config)
self.start_n_top = config.start_n_top
self.end_n_top = config.end_n_top
self.transformer = XLNetModel(config)
self.start_logits = PoolerStartLogits(config)
self.end_logits = PoolerEndLogits(config)
self.answer_class = PoolerAnswerMode(config, num_mode=5)
self.init_weights()
def __init__(self, cfg):
super(DSB_XLNetModel, self).__init__()
self.cfg = cfg
cate_col_size = len(cfg.cate_cols)
cont_col_size = len(cfg.cont_cols)
self.cate_emb = nn.Embedding(cfg.total_cate_size,
cfg.emb_size,
padding_idx=0)
self.cate_proj = nn.Sequential(
nn.Linear(cfg.emb_size * cate_col_size, cfg.hidden_size // 2),
nn.LayerNorm(cfg.hidden_size // 2),
)
self.cont_emb = nn.Sequential(
nn.Linear(cont_col_size, cfg.hidden_size // 2),
nn.LayerNorm(cfg.hidden_size // 2),
)
self.config = XLNetConfig(
3, # not used
d_model=cfg.hidden_size,
n_layer=cfg.nlayers,
n_head=cfg.nheads,
d_inner=cfg.hidden_size,
#ff_activation="gelu",
#untie_r=True,
#attn_type="bi",
#initializer_range=0.02,
#layer_norm_eps=1e-12,
dropout=cfg.dropout,
#mem_len=None,
#reuse_len=None,
#bi_data=False,
#clamp_len=-1,
#same_length=False,
#summary_type="last",
#summary_use_proj=True,
#summary_activation="tanh",
summary_last_dropout=cfg.dropout,
#start_n_top=5,
#end_n_top=5,
)
self.encoder = XLNetModel(self.config)
def get_reg():
return nn.Sequential(
nn.Linear(cfg.hidden_size, cfg.hidden_size),
nn.LayerNorm(cfg.hidden_size),
nn.Dropout(cfg.dropout),
nn.ReLU(),
nn.Linear(cfg.hidden_size, cfg.target_size),
)
self.reg_layer = get_reg()
def __init__(self, config, lambd, mean_pool=False):
super().__init__(config)
self.num_labels = config.num_labels
self.transformer = XLNetModel(config)
self.sequence_summary = SequenceSummary(config)
self.logits_proj = nn.Linear(config.d_model, config.num_labels)
self.lambd = lambd
print("Gradient reversal Prarameter is: {}".format(self.lambd))
self.grl = GradientReversal(self.lambd)
self.domain_classifier = nn.Linear(config.hidden_size, 2)
self.init_weights()
