RobustRanking/extract_relations.py at main · dice-group/RobustRanking · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
from models.E5 import E5Ranker
import random
import os
import pickle
from optimizers import noise
import logging
import json
from transformers import AutoTokenizer
from torch.utils.data import DataLoader
import re
from indexing import DenseFlatIndexer
from parameters import RankingParser
import torch
from tqdm import tqdm
import numpy

parser = RankingParser(add_model_args=True)
parser.add_training_args()
parser.add_eval_args()

# args = argparse.Namespace(**params)
args = parser.parse_args()
print(args)
params = args.__dict__

#device = "cuda:"+str(params["gpu_id"])
device = "cuda:0"
model = E5Ranker(device,params)
tokenizer = AutoTokenizer.from_pretrained('intfloat/e5-base-v2')
model.load_state_dict(torch.load("rel2relations/epoch_15/pytorch_model.bin", weights_only=True))
model.to(device)

test_data = json.load(
        open("/data/QA_Experiments/QA_Experiments/data/qald_9_train_auggmented_linked.json", "r",
             encoding="utf-8"))

for el in test_data["questions"]:
    gold_entities = re.findall(r"Q[0-9]+", el["query"]["sparql"])
    gold_relations = re.findall(r"P[0-9]+", el["query"]["sparql"])
    el["gold_ent"]=gold_entities
    el["gold_rel"]=gold_relations
'''
test_data = json.load(
        open("/data/QA_Experiments/QA_Experiments/Contextual_Augmentation_EL/data/qald_10_augmented_linked.json", "r",
             encoding="utf-8"))

test_data = json.load(
        open("/data/QA_Experiments/QA_Experiments/data/lcquad_train_augmented", "r",
             encoding="utf-8"))
for el in test_data:
    gold_entities = re.findall(r"Q[0-9]+", el["sparql_wikidata"])
    gold_relations = re.findall(r"P[0-9]+", el["sparql_wikidata"])
    el["gold_ent"]=gold_entities
    el["gold_rel"]=gold_relations
'''
entities=pickle.load(open("/data/QA_Experiments/QA_Experiments/data/relation_dict_item.pkl","rb"))
entities.update(pickle.load(open("/data/QA_Experiments/QA_Experiments/data/relation_dict_literal.pkl","rb")))
entities={v.replace("http://www.wikidata.org/entity/",""): k for v,k in entities.items()}


def evaluate(eval_data,model,entities,tokenizer):
    model.eval()
    sample = list(entities.keys())
    sample_app=[]
    #for sp in eval_data:
    #    sample_app.extend(sp["gold_ent"])

    sample_app.extend(sample_app)
    sample.extend(sample_app)
    doc_indices = {}
    found =0
    not_found=0
    index = DenseFlatIndexer(vector_sz=768)
    # index=DenseHNSWFlatIndexer(vector_sz=y_dim)
    for i in tqdm(range(0, len(sample), 1000)):
        for el in range(i, min(i + 1000, len(sample))):
            doc_indices[el] = sample[el]
        sample_repr = ["passage: " + entities[sample[el]] if sample[el] in entities
                        else sample[el] for el in range(i, min(i + 1000, len(sample)))]
        # print(sample_repr)
        batch = tokenizer(sample_repr, max_length=512, padding=True, truncation=True, return_tensors='pt')
        batch.to(device)
        current_documents = []
        encs = model.encode_candidate(batch).tolist()
        x_dim = len(encs)
        y_dim = len(encs[0])
        vectors = numpy.zeros((x_dim, y_dim), dtype=numpy.float32)
        for i in range(0, len(encs)):
            vectors[i] = numpy.asarray(encs[i])

        index.index_data(vectors)
    index.index_id_to_db_id = doc_indices
    for i in tqdm(range(0, len(eval_data), 100)):
        sample_repr = ["query: " + eval_data[el]["augmented_seq"] if "augmented_seq" in eval_data[el]
                    else eval_data[el]["NNQT_question"] for el in range(i, min(i + 100, len(eval_data)))]

        batch = tokenizer(sample_repr, max_length=512, padding=True, truncation=True, return_tensors='pt')

        batch.to(device)
        current_documents = []
        encs = model.encode_context(batch).tolist()
        cands_per_doc = index.search(encs, 5)
        for q_ind in range(len(cands_per_doc)):
            question = eval_data[q_ind + i]
            if "found_rel" in question:
                ent = list(question["found_rel"].values())
            else:
                ent = []
            ent_bert = cands_per_doc[q_ind]
            for e in ent_bert:
                if not e in ent:
                    ent.append(e)
            ent_dict = []
            for e in ent:
                ent_dict.append({"uri": e, "label": entities[e]})
            question["augmented_rel"] = ent_dict
            for e in eval_data[q_ind + i]["gold_rel"]:
                if e in ent:
                    found += 1
                else:
                    not_found += 1
    results = found / (not_found + found)
    print(results)
    return results

evaluate(test_data["questions"],model,entities,tokenizer)
#evaluate(test_data,model,entities,tokenizer)
json.dump(test_data, open("/data/QA_Experiments/QA_Experiments/data/qald_9_train_auggmented_linked.json","w",encoding="utf-8"), indent=4,ensure_ascii=False)