def main():
data_bin, model_bin = '../run/seqVerbMC/data_subsrl_1sv_1sa_argtrim.bin', './full_model/Roberta_BI/full_model_sptoken_ep121_a1.0_m1-0.1_m2-0.1.bin'
data = Data()
if os.path.exists(data_bin):
data.load(data_bin)
print("==ATTN== ", len(data.processes), " sequences.")
else:
data.load_tsv_plain(data_file)
data.save(data_bin)
data.dump_dataset_format('./dataset_seq.tsv', 'seq')
# W/O n-1 gram
M = torchpart()
M.load(model_bin)
M.serve_verb([' '], data, limit_ids=None,
topk=1), M.serve_arg([' '], data, limit_ids=None, topk=1)
sequence = input("Events split by @ (exit() to quit):")
#verbs, sequences, true_ids, v2s, limit_ids
while True:
sequence = sequence.strip().lower()
if sequence[:4] == 'exit':
exit()
sequence = sequence.split('@')
vtype, atype = M.serve_verb(sequence, data, limit_ids=None,
topk=6), M.serve_arg(sequence,
data,
limit_ids=None,
topk=6)
print(vtype, atype)
sequence = input("Events split by @ (exit() to quit):")
def main():
v_thres, l_thres = 50, 5
if len(sys.argv) > 1:
v_thres, l_thres = int(sys.argv[1]), int(sys.argv[2])
#data_file, data_bin, model_bin, test_file = '/shared/corpora-tmp/wikihow/wikiHowSubsequences.tsv', '../run/seqVerbMC/data_subsrl_1sv_1sa_argtrim.bin', './seqSSmrl_subsrl/RobertaVerbMC/tmp_fold_ep151_a1.0_m1-0.1_m2-0.1.bin', './seqSSmrl_subsrl/RobertaVerbMC/test_fold_ep151_a1.0_m1-0.1_m2-0.1.txt'
data_file, data_bin, model_bin, test_file = '/shared/corpora-tmp/wikihow/wikiHowSubsequences.tsv', '../run/seqVerbMC/data_subsrl_1sv_1sa_argtrim.bin', './seqSSmrl_subsrl/RobertaVerbMC/tmp_fold_ep151_a1.0_m1-0.1_m2-0.1.bin', '../process/recover_test_index_fold1.txt'
data = Data()
if os.path.exists(data_bin):
data.load(data_bin)
print("==ATTN== ", len(data.processes), " sequences.")
else:
data.load_tsv_plain(data_file)
data.save(data_bin)
# W/O n-1 gram
sequences = data.join_batch_sent(data.processes,
begin='<s> ',
sep=' </s> ')
seq_len = np.array([len(x) for x in data.processes])
r_verbs = {y: x for x, y in data.verb_vocab.items()}
n_verbs = len([x for x, y in data.verb_vocab.items()])
#print (n_verbs)
verbs = [r_verbs[x] for x in range(n_verbs)]
vid = np.array(data.verb_id)
true_senses = [data.v2s[verbs[x]] for x in vid]
r_args = {y: x for x, y in data.arg_vocab.items()}
n_args = len([x for x, y in data.arg_vocab.items()])
#print (n_args)
args = [r_args[x] for x in range(n_args)]
aid = np.array(data.arg_id)
true_arg_senses = [data.a2s[args[x]] for x in aid]
#print (true_senses[:3])
max_fold = 1
rs = sklearn.model_selection.ShuffleSplit(n_splits=max_fold,
test_size=0.1,
random_state=777)
avg_mrr, avg_hits1, avg_hits10 = [], [], []
avg_mrra, avg_hits1a, avg_hits10a = [], [], []
print(len(verbs), len(args))
test_index = []
for x in open(test_file):
test_index.append(int(x.strip()))
test_index = np.array(test_index)
test_seq = [sequences[x] for x in test_index]
test_vid = vid[test_index]
test_aid = aid[test_index]
M = torchpart()
M.load(model_bin)
#verbs, sequences, true_ids, v2s, limit_ids
M.profile_test_verb(verbs, test_seq, seq_len[test_index], test_vid,
data.v2s, v_thres, l_thres)
M.profile_test_verb(verbs, test_seq, seq_len[test_index], test_vid,
data.v2s, 525, 2)
M.profile_test_verb(verbs, test_seq, seq_len[test_index], test_vid,
data.v2s, 425, 2)
def main():
data_bin, model_bin = '../run/seqVerbMC/data_subsrl_1sv_1sa_argtrim.bin', './seqSSmrl_subsrl/RobertaVerbMC/tmp_fold_ep151_a1.0_m1-0.1_m2-0.1.bin'
data = Data()
if os.path.exists(data_bin):
data.load(data_bin)
print("==ATTN== ", len(data.processes), " sequences.")
else:
data.load_tsv_plain(data_file)
data.save(data_bin)
# W/O n-1 gram
ifile = None
if len(sys.argv) > 1:
ifile, ofile = sys.argv[1], sys.argv[2]
M = torchpart()
M.load(model_bin)
#verbs, sequences, true_ids, v2s, limit_ids
if ifile is None:
sequence = [
'set locations and date', 'search for tickets', 'compare airfares',
'purchase the ticket'
]
vtype, atype = M.serve_verb(sequence, data, limit_ids=None,
topk=10), M.serve_arg(sequence,
data,
limit_ids=None,
topk=10)
print(vtype, atype)
else:
with open(ofile, 'w') as fp:
for line in tqdm.tqdm(open(ifile)):
sequence = line.strip().split('\t')
vtype, atype = M.serve_verb(sequence,
data,
limit_ids=None,
topk=10), M.serve_arg(
sequence,
data,
limit_ids=None,
topk=10)
fp.write(line)
fp.write('\t@@@\tVERB: ' + json.dumps(vtype) + '\tARG: ' +
json.dumps(atype) + '\n')
### Read parameters ###
if len(sys.argv) > 1:
debugging = int(sys.argv[1][-1])
### Model Initialization ###
if not debugging:
data_bin, model_bin = '../data/wikihow_process/data_subsrl_1sv_1sa_argtrim.bin', './full_model/full_model_sptoken_ep121_a1.0_m1-0.1_m2-0.1.bin'
data = Data()
if os.path.exists(data_bin):
data.load(data_bin)
print("==ATTN== ", len(data.processes), " sequences.")
else:
data.load_tsv_plain(data_file)
data.save(data_bin)
M = torchpart()
M.load(model_bin)
M.serve_verb([' '], data, limit_ids=None,
topk=1), M.serve_arg([' '], data, limit_ids=None, topk=1)
### Function ###
def process_json(sequence):
sequence = sequence.split('@')
if not debugging:
vtype, atype = M.serve_verb(sequence,
data,
limit_ids=None,
topk=6,
return_emb=False), M.serve_arg(
sequence,