/
experiment.py
211 lines (175 loc) · 9.22 KB
/
experiment.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
# import required packages
import numpy as np
import json
import copy
import pickle
from multiprocessing import Pool
from itertools import product
from telegram_bot import TelegramNotifier
from multi_class_classification import MultiClassClassification
from multi_label_classification import MultiLabelClassification
from clustering import Clustering
from link_prediction import LinkPrediction
class Experiment:
# experiment types
CLUSTERING = 'clustering'
CLASSIFICATION = 'classification'
MULTI_LABEL_CLASSIFICATION = 'multi_label_classification'
LINK_PREDICTION = 'link_prediction'
def __init__(self, method_name='Verse-PPR', dataset_name='Test-Data', performance_function='both', node_labels={},
embeddings_file_path='', node_embedings=None, embedding_dimensionality=128, repetitions=10,
experiment_params={}, experiment_type='clustering', results_file_path=None, random_seeds=None,
pickle_path=None, telegram_config=None, node2id_filepath=None):
"""
Initialize experiment with given configuration parameters
:param method_name:
:param dataset_name:
:param performance_function:
:param node_labels:
:param embeddings_file_path:
:param node_embedings:
:param embedding_dimensionality:
:param repetitions:
:param experiment_params:
:param experiment_type:
:param results_file_path:
:param random_seeds:
:param telegram_config:
"""
self.method_name = method_name
self.dataset_name = dataset_name
self.performance_function = performance_function
self.embeddings_file_path = embeddings_file_path
self.node_embeddings = node_embedings
self.embedding_dimensionality = embedding_dimensionality
self.node_labels = node_labels
self.repetitions = repetitions
self.experiment_params = experiment_params
self.experiment_type = experiment_type
self.results_file_path = results_file_path
self.random_seed = random_seeds
self.random_seeds = random_seeds
self.executed_runs = []
self.pickle_path = pickle_path
self.experiments = []
self.telegram_config = telegram_config
self.telegram_notifier = None
self.node2id_filepath = node2id_filepath
assert len(self.random_seed) == self.repetitions, 'random seed array length and number of ' \
'repetitions are not equal'
self.generate_cross_product_params()
if self.node_embeddings is None:
self.read_node_embeddings_from_binary_file()
self.experiment_results = {
'method': self.method_name,
'dataset': self.dataset_name,
'embedding_file': self.embeddings_file_path,
'repetitions': self.repetitions,
'parameterizations': []
}
def generate_results_file(self):
with open(self.results_file_path, 'w') as results_file:
results_file.write(json.dumps(self.experiment_results, ensure_ascii=False, indent=4))
def generate_pickle_file(self):
with open(self.pickle_path, 'wb') as pickle_file:
pickle.dump(self, pickle_file)
def generate_cross_product_params(self):
cross_product_experiment_params = []
for values in product(*self.experiment_params.values()):
cross_product_experiment_params.append(dict(zip(self.experiment_params.keys(), values)))
self.experiment_params = cross_product_experiment_params
def read_node_embeddings_from_binary_file(self):
"""
Read given binary file and convert it to numpy (num_of_nodes, embedding_dimensions) shaped embeddings matrix
:return:
"""
embeddings_file = open(self.embeddings_file_path, "r")
embeddings_file_content = np.fromfile(embeddings_file, dtype=np.float32)
num_of_nodes = int(np.shape(embeddings_file_content)[0] / self.embedding_dimensionality)
self.node_embeddings = embeddings_file_content.reshape((num_of_nodes, self.embedding_dimensionality))
def build_telegram_bot(self):
self.telegram_config["experiment"] = {}
self.telegram_config["experiment"]["experiment_type"] = self.experiment_type
self.telegram_config["experiment"]["method_name"] = self.method_name
self.telegram_config["experiment"]["performance_function"] = self.performance_function
self.telegram_config["experiment"]["dataset_name"] = self.dataset_name
self.telegram_notifier = TelegramNotifier(self.telegram_config)
def init_run(self, run_params):
if self.experiment_type == self.CLASSIFICATION:
return MultiClassClassification(method_name=self.method_name, dataset_name=self.dataset_name,
performance_function=self.performance_function,
embeddings=self.node_embeddings, **run_params,
node_labels=self.node_labels,
node2id_filepath=self.node2id_filepath)
elif self.experiment_type == self.CLUSTERING:
return Clustering(method_name=self.method_name, dataset_name=self.dataset_name,
embeddings=self.node_embeddings, **run_params, node_labels=self.node_labels,
performance_function=self.performance_function,
node2id_filepath=self.node2id_filepath)
elif self.experiment_type == self.MULTI_LABEL_CLASSIFICATION:
return MultiLabelClassification(method_name=self.method_name, dataset_name=self.dataset_name,
node_labels=self.node_labels, **run_params,
performance_function=self.performance_function,
embeddings=self.node_embeddings,
node2id_filepath=self.node2id_filepath)
elif self.experiment_type == self.LINK_PREDICTION:
return LinkPrediction(method_name=self.method_name, dataset_name=self.dataset_name,
node_embeddings=self.node_embeddings, **run_params,
performance_function=self.performance_function,
node2id_filepath=self.node2id_filepath)
def run(self):
print('Start {} experiment on {} data set with {} embeddings\nRepeated {} times and evaluated through {}'
'performance function(s)'.format(self.experiment_type, self.dataset_name, self.method_name,
self.repetitions, self.performance_function))
if self.telegram_config:
self.build_telegram_bot()
for index, run_params in enumerate(self.experiment_params):
self.experiment_results['parameterizations'].append({
'params': run_params,
'runs': []
})
if self.telegram_notifier is not None:
try:
self.telegram_notifier.start_experiment(run_params)
except:
print("Failed sending message")
experiment = self.init_run(run_params)
self.executed_runs.append(experiment)
self.experiments = [copy.deepcopy(experiment) for rep in range(self.repetitions)]
pool = Pool(self.repetitions)
results = pool.map(self.perform_single_run,
[(index, rep, run_params, self.experiments[rep]) for rep in range(self.repetitions)])
self.experiment_results['parameterizations'][index]['runs'].extend(results)
if self.telegram_notifier is not None:
try:
self.telegram_notifier.finish_experiment(run_params)
except:
print("Failed sending message")
print('Finished {} experiment on {} data set with {} embeddings'
.format(self.experiment_type, self.dataset_name, self.method_name))
if self.results_file_path is not None:
self.generate_results_file()
print('Saved results in file {}'.format(self.results_file_path))
if self.pickle_path is not None:
self.generate_pickle_file()
print('Saved experiment as pickle-model in file {}'.format(self.pickle_path))
return self.experiment_results
def perform_single_run(self, single_run_params):
index, rep, run_params, experiment = single_run_params
random_seed = self.random_seeds[rep]
experiment.preprocess_data(random_seed=random_seed)
experiment.train(random_seed=random_seed)
experiment.predict()
evaluation = experiment.evaluate()
run_results = {
'run': rep + 1,
'random_seed': self.random_seeds[rep],
'experiment': id(experiment),
'evaluation': evaluation
}
if self.telegram_notifier is not None:
try:
self.telegram_notifier.finished_run(run_results["evaluation"])
except:
print("Failed sending message")
return run_results