def __init__(self, cls_num_labels=2, tok_num_labels=2, tok2id=None):
super(TaggerFromDebiaser, self).__init__()
global ARGS
global CUDA
if ARGS.pointer_generator:
self.debias_model = seq2seq_model.PointerSeq2Seq(
vocab_size=len(tok2id),
hidden_size=ARGS.hidden_size,
emb_dim=768,
dropout=0.2,
tok2id=tok2id)
else:
self.debias_model = seq2seq_model.Seq2Seq(
vocab_size=len(tok2id),
hidden_size=ARGS.hidden_size,
emb_dim=768,
dropout=0.2,
tok2id=tok2id)
assert ARGS.debias_checkpoint
print('LOADING DEBIASER FROM ' + ARGS.debias_checkpoint)
self.debias_model.load_state_dict(torch.load(ARGS.debias_checkpoint))
print('...DONE')
self.cls_classifier = nn.Sequential(
nn.Linear(ARGS.hidden_size, ARGS.hidden_size), nn.Dropout(0.1),
nn.ReLU(), nn.Linear(ARGS.hidden_size, cls_num_labels),
nn.Dropout(0.1))
self.tok_classifier = nn.Sequential(
nn.Linear(ARGS.hidden_size, ARGS.hidden_size), nn.Dropout(0.1),
nn.ReLU(), nn.Linear(ARGS.hidden_size, tok_num_labels),
nn.Dropout(0.1))
test=True,
add_del_tok=ARGS.add_del_tok)
# # # # # # # # ## # # # ## # # MODEL # # # # # # # # ## # # # ## # #
if ARGS.pointer_generator:
debias_model = seq2seq_model.PointerSeq2Seq(
vocab_size=len(tok2id),
hidden_size=ARGS.hidden_size,
emb_dim=768,
dropout=0.2,
tok2id=tok2id) # 768 = bert hidden size
else:
debias_model = seq2seq_model.Seq2Seq(vocab_size=len(tok2id),
hidden_size=ARGS.hidden_size,
emb_dim=768,
dropout=0.2,
tok2id=tok2id)
if ARGS.extra_features_top:
tagging_model = tagging_model.BertForMultitaskWithFeaturesOnTop.from_pretrained(
ARGS.bert_model,
cls_num_labels=ARGS.num_categories,
tok_num_labels=ARGS.num_tok_labels,
cache_dir=ARGS.working_dir + '/cache',
tok2id=tok2id)
elif ARGS.extra_features_bottom:
tagging_model = tagging_model.BertForMultitaskWithFeaturesOnBottom.from_pretrained(
ARGS.bert_model,
cls_num_labels=ARGS.num_categories,
tok_num_labels=ARGS.num_tok_labels,