-
Notifications
You must be signed in to change notification settings - Fork 3
/
Inference.py
173 lines (151 loc) · 6.82 KB
/
Inference.py
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
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
import json
import os.path
import sys
import time
import traceback
from clocq import config
from clocq.CLOCQAlgorithm import CLOCQAlgorithm
from clocq.Evaluation import Evaluation
from clocq.knowledge_base.KnowledgeBase import KnowledgeBase
from clocq.knowledge_base.KnowledgeBaseHDT import KnowledgeBaseHDT
from clocq.knowledge_base.KnowledgeBaseSPARQL import KnowledgeBaseSPARQL
from clocq.StringLibrary import StringLibrary
from clocq.WikidataSearchCache import WikidataSearchCache
def _run_method(method, method_name, parameter_tuples, data_split, is_clocq=True, store_jsonl=False):
"""Runs the given method on the benchmarks with the specified parameter settings."""
print("Running: ", method_name, "Testing: ", data_split, "KB: ", kb_name)
# initialize results
eval_path = method_name + ".res"
detailed_eval_path = method_name + ".json"
if os.path.isfile(detailed_eval_path):
with open(detailed_eval_path, "r") as fp:
results = json.load(fp)
else:
results = dict()
# result path for disambiguations and search spaces
result_path = method_name + ".jsonl"
# go through parameter settings
for parameters in parameter_tuples:
# go through benchmarks
for (benchmark_file, benchmark_name) in config.BENCHMARKS:
# load data
with open(benchmark_file, "r") as fp:
benchmark = json.load(fp)
benchmark = benchmark[data_split]
# initialize scores
answer_presence = list()
neighborhood_sizes_facts = list()
neighborhood_sizes_items = list()
all_answer_connecting_facts = list()
kb_item_tuples = list()
timings = list()
# iterate through benchmark
for i, instance in enumerate(benchmark):
question = instance["question"]
answers = instance["answers"]
# retrieve search space
question_start = time.time()
result = method.get_seach_space(question, parameters)
timing = time.time() - question_start
answer_connecting_facts = list()
question_context_tuples = list()
# iterate through contexts and accumulate result
kb_item_tuple = result["kb_item_tuple"]
search_space = result["search_space"]
result, answer_connecting_facts = evaluation.evaluate(search_space, answers)
# store search space and disambiguations to disk
if store_jsonl:
instance["kb_item_tuple"] = kb_item_tuple
instance["search_space"] = search_space
with open(result_path, "a") as fp:
fp.write(json.dumps(instance))
fp.write("\n")
# remember results
kb_item_tuples.append(kb_item_tuple)
answer_presence.append(result.hit)
neighborhood_sizes_facts.append(result.neighbordhood_size_facts)
neighborhood_sizes_items.append(result.neighbordhood_size_items)
all_answer_connecting_facts.append(answer_connecting_facts)
timings.append(timing)
# print results
method.print_results((question, method_name, i + 1, sum(answer_presence) / (i + 1)))
# create result
avg_neighborhood_sizes_facts = (
f"{round(sum(neighborhood_sizes_facts)/len(neighborhood_sizes_facts), -2)/1000}k"
)
avg_neighborhood_sizes_items = (
f"{round(sum(neighborhood_sizes_items)/len(neighborhood_sizes_items), -2)/1000}k"
)
avg_answer_presence = round(sum(answer_presence) / len(answer_presence), 3)
result = {
"method_name": method_name,
"NER_method": str(config.NER),
"parameters": parameters,
"data_split": data_split,
"instances": len(benchmark),
# 'neighborhood_sizes_facts': neighborhood_sizes_facts,
# 'neighborhood_sizes_items': neighborhood_sizes_items,
"avg_neighborhood_sizes_facts": avg_neighborhood_sizes_facts,
"avg_neighborhood_sizes_items": avg_neighborhood_sizes_items,
# "kb_item_tuples": kb_item_tuples,
# 'timings': timings,
"avg_time_consumed": round(sum(timings) / len(timings), 2),
"answer_presence": answer_presence,
"avg_answer_presence": avg_answer_presence
# 'answer_connecting_facts': all_answer_connecting_facts,
}
# append results
if not results.get(benchmark_name):
results[benchmark_name] = list()
results[benchmark_name].append(result)
# store result
with open(eval_path, "a") as fp:
fp.write(
f"Benchmark: {benchmark_name}, Answer presence: {avg_answer_presence}, Search space size: {avg_neighborhood_sizes_items}, Parameters: {parameters}\n"
)
with open(detailed_eval_path, "w") as fp:
fp.write(json.dumps(results, indent=4))
# store caches
if is_clocq:
method.store_caches()
if __name__ == "__main__":
if len(sys.argv) < 2:
sys.exit(0)
params = sys.argv[1:]
# define whether test or dev is run
if "--test" in params:
data_split = "test"
elif "--dev" in params:
data_split = "dev"
else:
data_split = "test"
# required modules
string_lib = StringLibrary(config.PATH_TO_STOPWORDS, config.TAGME_TOKEN, config.PATH_TO_TAGME_NER_CACHE)
evaluation = Evaluation(string_lib)
wikidata_search_cache = WikidataSearchCache(config.PATH_TO_WIKI_SEARCH_CACHE)
# load kb (always needed for evaluation)
if "--hdt" in params:
kb_name = "hdt"
kb = KnowledgeBaseHDT(config.PATH_TO_HDT_FILE, config.PATH_TO_KB_DICTS, string_lib)
elif "--sparql" in params:
kb_name = "sparql"
kb = KnowledgeBaseSPARQL(config.PATH_TO_KB_DICTS, string_lib)
elif "--dummy" in params:
kb_name = "dummy"
kb = KnowledgeBase(config.PATH_TO_KB_LIST, config.PATH_TO_KB_DICTS, max_items=10)
else:
kb_name = "clocq"
kb = KnowledgeBase(config.PATH_TO_KB_LIST, config.PATH_TO_KB_DICTS)
method_name = "results/clocq_" + data_split + "_" + kb_name
clocq = CLOCQAlgorithm(
kb,
string_lib,
method_name,
config.NER,
config.PATH_TO_STOPWORDS,
config.PATH_TO_WIKI2VEC_MODEL,
config.PATH_TO_WIKIPEDIA_MAPPINGS,
config.PATH_TO_NORM_CACHE,
wikidata_search_cache=wikidata_search_cache,
)
_run_method(clocq, method_name, config.CLOCQ_PARAMS, data_split, is_clocq=True)