'aet_word_voca': aet_word_voca,
'aet_word_embeddings': aet_word_embeddings,
'dr': args.dropout_rate,
'args': args
}
if ModelClass == MulRelRanker:
config['df'] = args.df
config['n_loops'] = args.n_loops
config['n_rels'] = args.n_rels
config['mulrel_type'] = args.mulrel_type
else:
raise Exception('unknown model class')
pprint(config)
ranker = EDRanker(config=config)
dev_datasets = [('aida-A', conll.testA), ('aida-B', conll.testB),
('msnbc', conll.msnbc), ('aquaint', conll.aquaint),
('ace2004', conll.ace2004), ('clueweb', conll.clueweb),
('wikipedia', conll.wikipedia)]
if args.mode == 'train':
print('training...')
config = {'lr': args.learning_rate, 'n_epochs': args.n_epochs}
pprint(config)
ranker.train(conll.train, dev_datasets, config)
elif args.mode == 'eval':
org_dev_datasets = dev_datasets # + [('aida-train', conll.train)]
dev_datasets = []
if args.multi_instance or args.semisup:
config['n_negs'] = args.n_negs
if ModelClass == MulRelRanker:
config['inference'] = args.inference
config['df'] = args.df
config['n_loops'] = args.n_loops
config['ent_top_n'] = args.ent_top_n
config['n_rels'] = args.n_rels
config['mulrel_type'] = args.mulrel_type
else:
raise Exception('unknown model class')
# pprint(config)
ranker = EDRanker(config=config)
if args.mode == 'prerank':
conll = D.CoNLLDataset(datadir, person_path, conll_path)
if args.filelist is not None:
if args.multi_instance:
conll.train = {}
with open(args.filelist, 'r') as flist:
for fname in flist:
fname = fname.strip()
print('load file from', fname)
conll_path = fname
cands_path = conll_path + '.csv'
data = D.CoNLLDataset.load_file(conll_path, cands_path, person_path)
print('#docs', len(data))
'entity_embeddings': entity_embeddings,
'snd_word_voca': snd_word_voca,
'snd_word_embeddings': snd_word_embeddings,
'dr': args.dropout_rate,
'args': args}
if ModelClass == MulRelRanker:
config['df'] = args.df
config['n_loops'] = args.n_loops
config['n_rels'] = args.n_rels
config['mulrel_type'] = args.mulrel_type
else:
raise Exception('unknown model class')
pprint(config)
ranker = EDRanker(config=config)
dev_datasets = [('test', conll.test),
('dev', conll.dev),
]
if args.mode == 'train':
print('training...')
config = {'lr': args.learning_rate, 'n_epochs': args.n_epochs}
pprint(config)
ranker.train(conll.train, dev_datasets, config)
elif args.mode == 'eval':
org_dev_datasets = dev_datasets # + [('aida-train', conll.train)]
dev_datasets = []
for dname, data in org_dev_datasets:
'snd_word_voca': snd_word_voca,
'snd_word_embeddings': snd_word_embeddings,
'dr': args.dropout_rate,
'args': args
}
if ModelClass == MulRelRanker:
config['df'] = args.df
config['n_loops'] = args.n_loops
config['n_rels'] = args.n_rels
config['mulrel_type'] = args.mulrel_type
else:
raise Exception('unknown model class')
pprint(config)
ranker = EDRanker(config=config)
dev_datasets = [
('aida-A', conll.testA),
('aida-B', conll.testB),
('msnbc', conll.msnbc),
('aquaint', conll.aquaint),
('ace2004', conll.ace2004),
('clueweb', conll.clueweb),
('wikipedia', conll.wikipedia),
('twitter-microposts', conll.twitter_microposts),
('twitter-mena', conll.twitter_mena),
('twitter-brian', conll.twitter_brian),
('twitter-train', conll.twitter_train),
('twitter-val', conll.twitter_val),
('twitter-test', conll.twitter_test),
voca_emb_dir + '/glove/word_embeddings.npy')
print('snd word voca size', snd_word_voca.size())
entity_voca, entity_embeddings = utils.load_voca_embs(
voca_emb_dir + 'dict.entity', voca_emb_dir + 'entity_embeddings.npy')
config = {
'hid_dims': args.hid_dims,
'emb_dims': entity_embeddings.shape[1],
'freeze_embs': True,
'tok_top_n': args.tok_top_n,
'margin': args.margin,
'word_voca': word_voca,
'entity_voca': entity_voca,
'word_embeddings': word_embeddings,
'entity_embeddings': entity_embeddings,
'snd_word_voca': snd_word_voca,
'snd_word_embeddings': snd_word_embeddings,
'dr': args.dropout_rate,
'args': args
}
if ModelClass == MulRelRanker:
config['df'] = args.df
config['n_loops'] = args.n_loops
config['n_rels'] = args.n_rels
config['mulrel_type'] = args.mulrel_type
else:
raise Exception('unknown model class')
ranker = EDRanker(config=config)
app.run(host='0.0.0.0', port=5555)