/
predictive_performance.py
executable file
·248 lines (216 loc) · 9.61 KB
/
predictive_performance.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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
import ampligraph.datasets
import ampligraph.latent_features
from ampligraph.datasets import load_wn18, load_fb15k, load_fb15k_237
from ampligraph.latent_features import TransE, DistMult, ComplEx
from ampligraph.evaluation import select_best_model_ranking, hits_at_n_score, mar_score, evaluate_performance, mrr_score
import argparse, os, json, sys
import numpy as np
import logging
logging.basicConfig(level=logging.DEBUG)
from os import path
from beautifultable import BeautifulTable
from tqdm import tqdm
import yaml
import tensorflow as tf
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
import warnings
warnings.simplefilter(action="ignore", category=Warning)
def display_scores(scores):
output_readme = {}
output_rst = {}
for obj in scores:
output_rst[obj["dataset"]] = BeautifulTable()
output_rst[obj["dataset"]].set_style(BeautifulTable.STYLE_RST)
output_rst[obj["dataset"]].column_headers = ["Model", "MR", "MRR", "H @ 1", "H @ 3", "H @ 10", "Hyperparameters"]
for obj in scores:
try:
output_rst[obj["dataset"]].append_row([obj["model"],
"{0:.2f}".format(obj["mr"]),
"{0:.2f}".format(obj["mrr"]),
"{0:.2f}".format(obj["H@1"]),
"{0:.2f}".format(obj["H@3"]),
"{0:.2f}".format(obj["H@10"]),
yaml.dump(obj["hyperparams"],default_flow_style=False)])
except:
output_rst[obj["dataset"]].append_row([obj["model"],
".??",
".??",
".??",
".??",
".??",
".??"])
for key, value in output_rst.items():
print(key)
print(value)
# clean datasets with unseen entities
def clean_data(train, valid, test, throw_valid = False):
train_ent = set(train.flatten())
valid_ent = set(valid.flatten())
test_ent = set(test.flatten())
if not throw_valid:
# filter test
train_valid_ent = set(train.flatten()) | set(valid.flatten())
ent_test_diff_train_valid = test_ent - train_valid_ent
idxs_test = []
if len(ent_test_diff_train_valid) > 0:
count_test = 0
c_if=0
for row in test:
tmp = set(row)
if len(tmp & ent_test_diff_train_valid) != 0:
idxs_test.append(count_test)
c_if+=1
count_test = count_test + 1
filtered_test = np.delete(test, idxs_test, axis=0)
logging.debug("fit validation case: shape test: {0} - filtered test: {1}: {2} triples with unseen entties removed".format(test.shape, filtered_test.shape, c_if))
return valid, filtered_test
else:
#filter valid
ent_valid_diff_train = valid_ent - train_ent
idxs_valid = []
if len(ent_valid_diff_train) > 0:
count_valid = 0
c_if=0
for row in valid:
tmp = set(row)
if len(tmp & ent_valid_diff_train) != 0:
idxs_valid.append(count_valid)
c_if+=1
count_valid = count_valid + 1
filtered_valid = np.delete(valid, idxs_valid, axis=0)
logging.debug("not fitting validation case: shape valid: {0} - filtered valid: {1}: {2} triples with unseen entties removed".format(valid.shape, filtered_valid.shape, c_if))
# filter test
ent_test_diff_train = test_ent - train_ent
idxs_test = []
if len(ent_test_diff_train) > 0:
count_test = 0
c_if=0
for row in test:
tmp = set(row)
if len(tmp & ent_test_diff_train) != 0:
idxs_test.append(count_test)
c_if+=1
count_test = count_test + 1
filtered_test = np.delete(test, idxs_test, axis=0)
logging.debug("not fitting validation case: shape test: {0} - filtered test: {1}: {2} triples with unseen entties removed".format(test.shape, filtered_test.shape, c_if))
return filtered_valid, filtered_test
def run_single_exp(config, dataset, model):
hyperparams = config["hyperparams"][dataset][model]
if hyperparams is None:
logging.info("dataset {0}...model {1} experiment is not conducted yet...".format(dataset, config["model_name_map"][model]))
return {
"hyperparams": ".??"
}
logging.info("dataset {0}...model {1}...best hyperparameter:...{2}".format(dataset, config["model_name_map"][model], hyperparams))
es_code = "{0}_{1}".format(dataset, model)
load_func = getattr(ampligraph.datasets, config["load_function_map"][dataset])
X = load_func()
# print("Loaded...{0}...".format(dataset))
if dataset in config["DATASET_WITH_UNSEEN_ENTITIES"]:
logging.debug("{0} contains unseen entities in test dataset, we cleaned them...".format(dataset))
X["valid"], X["test"] = clean_data(X["train"], X["valid"], X["test"], throw_valid=True)
# load model
model_class = getattr(ampligraph.latent_features, config["model_name_map"][model])
model = model_class(**hyperparams)
# Fit the model on training and validation set
# The entire dataset will be used to filter out false positives statements
# created by the corruption procedure:
filter = np.concatenate((X['train'], X['valid'], X['test']))
if es_code in config["no_early_stopping"]:
logging.debug("Fit without early stopping...")
model.fit(X["train"])
else:
logging.debug("Fit with early stopping...")
model.fit(X["train"], True,
{
'x_valid':X['valid'][::10],
'criteria':'mrr',
'x_filter':filter,
'stop_interval': 2,
'burn_in':0,
'check_interval':100
})
# Run the evaluation procedure on the test set. Will create filtered rankings.
# To disable filtering: filter_triples=None
ranks = evaluate_performance(X['test'], model, filter, verbose=False, corrupt_side='s')
ranks2 = evaluate_performance(X['test'], model, filter, verbose=False, corrupt_side='o')
ranks.extend(ranks2)
# compute and print metrics:
mr = mar_score(ranks)
mrr = mrr_score(ranks)
hits_1 = hits_at_n_score(ranks, n=1)
hits_3 = hits_at_n_score(ranks, n=3)
hits_10 = hits_at_n_score(ranks, n=10)
return {
"mr": mr,
"mrr": mrr,
"H@1": hits_1,
"H@3": hits_3,
"H@10": hits_10,
"hyperparams": hyperparams
}
def run_all(config):
obj = []
for dataset in tqdm(config["hyperparams"].keys(), desc = "evaluation done: "):
for model in config["hyperparams"][dataset].keys():
result = run_single_exp(config, dataset, model)
obj.append({
**result,
"dataset": dataset,
"model": config["model_name_map"][model]
})
return obj
def run_single_dataset(config, dataset):
obj = []
for model in tqdm(config["hyperparams"][dataset].keys(), desc = "evaluation done: "):
result = run_single_exp(config, dataset, model)
obj.append({
**result,
"dataset": dataset,
"model": config["model_name_map"][model]
})
return obj
def run_single_model(config, model):
obj = []
for dataset in tqdm(config["hyperparams"].keys(), desc = "evaluation done: "):
result = run_single_exp(config, dataset, model)
obj.append({
**result,
"dataset": dataset,
"model": config["model_name_map"][model]
})
return obj
def main():
with open("config.json", "r") as fi:
config = json.load(fi)
# set GPU id to run
os.environ["CUDA_VISIBLE_DEVICES"]=config["CUDA_VISIBLE_DEVICES"]
# print("Will use gpu number...",config["CUDA_VISIBLE_DEVICES"])
parser = argparse.ArgumentParser()
parser.add_argument("-d", "--dataset", type=str)
parser.add_argument("-m", "--model", type=str)
args = parser.parse_args()
# print("Input dataset...{0}...input model...{1}...".format(args.dataset, args.model))
if args.dataset is None:
if args.model is None:
display_scores(run_all(config))
else:
if args.model.upper() not in config["model_name_map"]:
sys.exit("Input model is not valid...")
display_scores(run_single_model(config, args.model.upper()))
else:
if args.model is not None:
if args.model.upper() not in config["model_name_map"] or args.dataset.upper() not in config["load_function_map"]:
sys.exit("Input model or dataset is not valid...")
result = run_single_exp(config, args.dataset.upper(), args.model.upper())
display_scores([{
**result,
"dataset": args.dataset,
"model": config["model_name_map"][args.model.upper()]
}])
else:
if args.dataset.upper() not in config["load_function_map"]:
sys.exit("Input dataset is not supported yet...")
display_scores(run_single_dataset(config, args.dataset.upper()))
if __name__ == "__main__":
main()