-
Notifications
You must be signed in to change notification settings - Fork 0
/
cluster_script.py
263 lines (246 loc) · 15 KB
/
cluster_script.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
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
"""Thin experiment runner which takes all simulation parameters from a database.
Usage:
cluster_script.py --cindex=<id> --isgpu=<bool> --schema=<schema>
cluster_script.py (-h | --help)
Arguments:
FILE An argument for passing in a file.
Options:
-h --help Show this screen.
--cindex=<cindex> Index given by the cluster to specify which job
is to be executed [default: 0]
--isgpu=<bool> Boolean to show if the gpu is to be used or not
--schema=<schema> Schema containing the job information
"""
import inspect
import logging
import os
import sys
import traceback
from datetime import datetime
import dill
import h5py
import numpy as np
from docopt import docopt
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import StratifiedShuffleSplit
from autoqild import *
from experiments.dbconnection import DBConnector
from experiments.utils import *
DIR_PATH = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
if __name__ == "__main__":
######################## DOCOPT ARGUMENTS: #################################
arguments = docopt(__doc__)
cluster_id = int(arguments["--cindex"])
is_gpu = bool(int(arguments["--isgpu"]))
schema = arguments["--schema"]
###################### POSTGRESQL PARAMETERS ###############################
config_file_path = os.path.join(DIR_PATH, EXPERIMENTS, 'config', 'autosca.json')
dbConnector = DBConnector(config_file_path=config_file_path, is_gpu=is_gpu, schema=schema)
os.environ["HIP_LAUNCH_BLOCKING"] = "1"
os.environ["CUDA_LAUNCH_BLOCKING"] = "2"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
for i in range(5000):
start = datetime.now()
dbConnector.job_description = None
if 'CCS_REQID' in os.environ.keys():
cluster_id = int(os.environ['CCS_REQID'])
print("************************************************************************************")
print(f"Getting the job for iteration {i}")
dbConnector.fetch_job_arguments(cluster_id=cluster_id)
if dbConnector.job_description is not None:
try:
seed = int(dbConnector.job_description["seed"])
job_id = int(dbConnector.job_description["job_id"])
fold_id = int(dbConnector.job_description["fold_id"])
dataset_name = dbConnector.job_description["dataset"]
n_inner_folds = int(dbConnector.job_description["inner_folds"])
dataset_params = dbConnector.job_description["dataset_params"]
learner_name = dbConnector.job_description["learner"]
fit_params = dbConnector.job_description["fit_params"]
learner_params_db = dbConnector.job_description["learner_params"]
duration = dbConnector.job_description["duration"]
hp_iters = int(dbConnector.job_description["hp_iters"])
hp_ranges = dbConnector.job_description["hp_ranges"]
learning_problem = dbConnector.job_description["learning_problem"]
experiment_schema = dbConnector.job_description["experiment_schema"]
experiment_table = dbConnector.job_description["experiment_table"]
validation_loss = dbConnector.job_description["validation_loss"]
hash_value = dbConnector.job_description["hash_value"]
LEARNING_PROBLEM = learning_problem.lower()
if validation_loss == 'None':
validation_loss = None
random_state = np.random.RandomState(seed=seed + fold_id)
BASE_DIR = os.path.join(DIR_PATH, EXPERIMENTS, LEARNING_PROBLEM)
log_path = os.path.join(BASE_DIR, LOGS_FOLDER, f"{hash_value}.log")
create_directory_safely(BASE_DIR, False)
create_directory_safely(log_path, True)
setup_logging(log_path=log_path)
setup_random_seed(random_state=random_state)
logger = logging.getLogger('Experiment')
print(lp_metric_dict[learning_problem].keys())
logger.info(f"DB config filePath {config_file_path}")
logger.info(f"Arguments {arguments}")
logger.info(f"Job Description {print_dictionary(dbConnector.job_description)}")
duration = get_duration_seconds(duration)
dataset_params['random_state'] = random_state
dataset_params['fold_id'] = fold_id
# if 'more_instances' in learner_name:
# dataset_params["samples_per_class"] = 250 * dataset_params["n_features"]
dataset_reader = get_dataset_reader(dataset_name, dataset_params)
X, y = dataset_reader.generate_dataset()
n_features = X.shape[-1]
n_classes = len(np.unique(y))
sss = StratifiedShuffleSplit(n_splits=1, test_size=0.30, random_state=0)
train_index, test_index = list(sss.split(X, y))[0]
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
optimizers_file_path = os.path.join(BASE_DIR, OPTIMIZER_FOLDER, f"{hash_value}.pkl")
create_directory_safely(optimizers_file_path, True)
learner = learners[learner_name]
learner_params = convert_learner_params(learner_params_db)
learner_params['random_state'] = random_state
logger.info(f"Time Taken till now: {seconds_to_time(duration_till_now(start))} seconds")
if learner in [MultiLayerPerceptron, MineMIEstimator, AutoTabPFNClassifier, MineMIEstimatorHPO]:
n_jobs = 1
else:
n_jobs = 10
logger.info(f"Actual Mutual Information {dataset_reader.get_bayes_mi(MCMC_MI_ESTIMATION)}")
learner_params = {**learner_params, **dict(n_features=n_features, n_classes=n_classes)}
if learner in [BayesPredictor, AutoGluonClassifier, MineMIEstimator, MajorityVoting]:
if learner == BayesPredictor:
learner_params = {'dataset_obj': dataset_reader}
estimator = learner(**learner_params)
estimator.fit(X_train, y_train)
y_true, y_pred, p_pred = estimator.get_bayes_predictor_scores()
elif learner == MajorityVoting:
estimator = learner(**learner_params)
estimator.fit(X_train, y_train)
p_pred, y_pred = get_scores(X, estimator)
y_true = np.copy(y)
elif learner == MineMIEstimator:
learner_params = {**learner_params, **dict(n_features=n_features, n_classes=n_classes)}
estimator = learner(**learner_params)
estimator.fit(X_train, y_train, **fit_params)
p_pred, y_pred = get_scores(X, estimator)
y_true = np.copy(y)
else:
estimator = get_automl_learned_estimator(optimizers_file_path, logger)
if estimator is None:
folder = os.path.join(BASE_DIR, OPTIMIZER_FOLDER, f"{hash_value}gluon")
if not os.path.isdir(folder):
os.mkdir(folder)
logger.info(f"AutoGluon learner params {print_dictionary(learner_params)}")
learner_params['output_folder'] = folder
learner_params['eval_metric'] = validation_loss
estimator = learner(**learner_params)
estimator.fit(X_train, y_train, **fit_params)
dill.dump(estimator, open(optimizers_file_path, "wb"))
logger.info("AutoML pipeline trained and saving the model")
else:
logger.info("AutoML pipeline trained and reusing it")
p_pred, y_pred = get_scores(X, estimator)
y_true = np.copy(y)
else:
# inner_cv_iterator = StratifiedKFold(n_splits=n_inner_folds, shuffle=True, random_state=random_state)
inner_cv_iterator = StratifiedShuffleSplit(n_splits=n_inner_folds, test_size=0.30,
random_state=random_state)
search_space = create_search_space(hp_ranges, logger)
logger.info(f"Search Space {search_space}")
estimator = learner(**learner_params)
bayes_search_params = dict(estimator=estimator, search_spaces=search_space, n_iter=hp_iters,
scoring=validation_loss, n_jobs=n_jobs, cv=inner_cv_iterator,
error_score=0, random_state=random_state,
optimizers_file_path=optimizers_file_path)
bayes_search = BayesSearchCV(**bayes_search_params)
search_keys = list(search_space.keys())
search_keys.sort()
logger.info(f"Search Keys {search_keys}")
callback = log_callback(search_keys)
try:
bayes_search.fit(X_train, y_train, groups=None, callback=callback, **fit_params)
except Exception as error:
log_exception_error(logger, error)
logger.error("Cannot fit the Bayes SearchCV ")
logger.info("Fitting the model with best parameters")
best_loss, learner_params = update_params_at_k(bayes_search, search_keys, learner_params, k=0)
logger.info(f"Setting the best parameters {print_dictionary(learner_params)}")
estimator = learner(**learner_params)
estimator.fit(X_train, y_train, **fit_params)
p_pred, y_pred = get_scores(X, estimator)
y_true = np.copy(y)
if issubclass(learner, MIEstimatorBase):
estimated_mi = estimator.estimate_mi(X, y)
result_file = os.path.join(BASE_DIR, RESULT_FOLDER, f"{hash_value}.h5")
logger.info(f"Result file {result_file}")
create_directory_safely(result_file, True)
f = h5py.File(result_file, 'w')
f.create_dataset('scores', data=p_pred)
f.create_dataset('predictions', data=y_pred)
f.create_dataset('ground_truth', data=y_true)
f.create_dataset('confusion_matrix', data=confusion_matrix(y_true, y_pred))
f.close()
results = {'job_id': str(job_id), 'cluster_id': str(cluster_id)}
for metric_name, evaluation_metric in lp_metric_dict[learning_problem].items():
if LOG_LOSS_MI_ESTIMATION in metric_name or PC_SOFTMAX_MI_ESTIMATION in metric_name:
calibrator_technique = None
for key in calibrators.keys():
if key in metric_name:
calibrator_technique = key
if calibrator_technique is not None:
calibrator = calibrators[calibrator_technique]
c_params = calibrator_params[calibrator_technique]
calibrator = calibrator(**c_params)
try:
p_pred_cal = probability_calibration(X_train=X_train, y_train=y_train, X_test=X_test,
classifier=estimator, calibrator=calibrator)
metric_loss = evaluation_metric(y_test, p_pred_cal)
except Exception as error:
log_exception_error(logger, error)
logger.error("Error while calibrating the probabilities setting mi using non"
"calibrated probabilities")
metric_loss = evaluation_metric(y_true, p_pred)
else:
metric_loss = evaluation_metric(y_true, p_pred)
elif metric_name in [MCMC_LOG_LOSS, MCMC_MI_ESTIMATION, MCMC_PC_SOFTMAX, MCMC_SOFTMAX]:
metric_loss = dataset_reader.get_bayes_mi(metric_name)
elif metric_name == ESTIMATED_MUTUAL_INFORMATION:
metric_loss = estimated_mi
else:
if metric_name == F_SCORE:
if n_classes > 2:
metric_loss = evaluation_metric(y_true, y_pred, average='macro')
else:
metric_loss = evaluation_metric(y_true, y_pred)
else:
metric_loss = evaluation_metric(y_true, y_pred)
if np.isnan(metric_loss) or np.isinf(metric_loss):
results[metric_name] = "\'Infinity\'"
else:
if np.around(metric_loss, 4) == 0.0:
results[metric_name] = f"{metric_loss}"
else:
results[metric_name] = f"{np.around(metric_loss, 4)}"
logger.info(f"Out of sample error {metric_name} : {metric_loss}")
print(f"Out of sample error {metric_name} : {metric_loss}")
evaluation_time = dbConnector.mark_running_job_finished(job_id, start)
dbConnector.insert_results(experiment_schema=experiment_schema, experiment_table=experiment_table,
results=results)
logger.info(f"Job finished")
print(f"Job finished")
except Exception as e:
if hasattr(e, 'message'):
message = e.message
else:
message = e
logger.error(traceback.format_exc())
message = "exception{}".format(str(message))
dbConnector.append_error_string_in_running_job(job_id=job_id, error_message=message)
except:
logger.error(traceback.format_exc())
message = "exception{}".format(sys.exc_info()[0].__name__)
dbConnector.append_error_string_in_running_job(job_id=job_id, error_message=message)
finally:
if "224" in str(cluster_id):
f = open("{}/.hash_value".format(os.environ['HOME']), "w+")
f.write(hash_value + "\n")
f.close()