def main():
parser = argparse.ArgumentParser()
parser.add_argument('--mode', choices=['classification', 'regression'])
parser.add_argument('--model-dir')
parser.add_argument('--train-csv')
parser.add_argument('--test-csv')
parser.add_argument('--prediction-csv')
args = parser.parse_args()
automl = AutoML(args.model_dir)
if args.train_csv is not None:
automl.train(args.train_csv, args.mode)
automl.save()
elif args.test_csv is not None:
automl.load()
automl.predict(args.test_csv, args.prediction_csv)
else:
exit(1)
def validate_dataset(alias: str, mode: str, train_limit: int) -> np.float64:
Log.print(alias)
automl = AutoML("models/check_{}".format(alias))
automl.config["time_limit"] = train_limit
automl.train("data/check_{}/train.csv".format(alias), mode)
automl.config["time_limit"] = 300
automl.config["start_time"] = time.time()
_, score = automl.predict("data/check_{}/test.csv".format(alias), "predictions/check_{}.csv".format(alias))
return score
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--mode', choices=['classification', 'regression'])
parser.add_argument('--model-dir')
parser.add_argument('--train-csv')
parser.add_argument('--test-csv')
parser.add_argument('--prediction-csv')
args = parser.parse_args()
automl = AutoML(args.model_dir)
if args.train_csv is not None:
automl.train(args.train_csv, args.mode)
automl.save()
elif args.test_csv is not None:
automl.load()
automl.predict(args.test_csv, args.prediction_csv)
if "TIME_LIMIT" in os.environ:
u.log("Time limit is {}".format(os.environ["TIME_LIMIT"]))
else:
exit(1)
def validate_dataset(alias: str, mode: str, train_limit: int) -> np.float64:
start_time = time.time()
log(alias)
automl = AutoML("models/check_{}".format(alias))
automl.config["time_limit"] = train_limit
# automl.load()
automl.train("data/check_{}/train.csv".format(alias), mode)
score_train_val = None
if 'leak' not in automl.config:
config = automl.config
if config['mode']=='regression':
best_oof = np.min([i['score_oof'] for i in config["lgb_cv_models"]])
else:
best_oof = np.max([i['score_oof'] for i in config["lgb_cv_models"]])
out_log = pd.DataFrame(automl.config['log'])
out_log.to_csv(f'models/check_{alias}/log_train.csv')
print(pd.DataFrame(automl.config['log']))
end_time = time.time()
train_time = end_time - start_time
start_time = end_time
automl.config["time_limit"] = 300
_, score_test_val = automl.predict("data/check_{}/test.csv".format(alias), "predictions/check_{}.csv".format(alias))
out_log = pd.DataFrame(automl.config['log'])
out_log.to_csv(f'models/check_{alias}/log_test.csv')
print(pd.DataFrame(automl.config['log']))
end_time = time.time()
test_time = end_time - start_time
return best_oof, score_test_val, train_time, test_time
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--mode', choices=['classification', 'regression'])
parser.add_argument('--model-dir')
parser.add_argument('--train-csv')
parser.add_argument('--test-csv')
parser.add_argument('--prediction-csv')
args = parser.parse_args()
automl = AutoML(args.model_dir)
time_limit = int(os.environ['TIME_LIMIT'])
automl.config['time_limit'] = time_limit
log(f'{args.model_dir} - time_limit: {time_limit}')
def automl_train():
automl.train(args.train_csv, args.mode)
# automl.save()
out_log = pd.DataFrame(automl.config['log'])
out_log.to_csv(f'{args.model_dir}/log_train.csv')
print(pd.DataFrame(automl.config['log']))
print('='*20)
def automl_predict():
automl.load()
automl.predict(args.test_csv, args.prediction_csv)
time_left_for_task = time_limit - DELAY_TO_SIGKILL - BUFFER_BEFORE_SENDING_SIGTERM
print('time_left_for_task :', time_left_for_task)
if args.train_csv is not None:
target_proc = automl_train
elif args.test_csv is not None:
target_proc = automl_predict
else:
exit(1)
p = Process(target=target_proc,
kwargs={
})
p.start()
p.join(time_left_for_task)
pid = p.pid
if p.is_alive():
parent = psutil.Process(pid)
for child in parent.children(recursive=True):
child.kill()
parent.kill()
log("Starting Shutdown!")
program_exp = re.compile(r"main\.py").search
contacts = send_signal_to_our_processes(sig=SIGTERM, filter=program_exp)
log("Sending SIG=%d to %s" % (SIGTERM, str(contacts)))
time.sleep(DELAY_TO_SIGKILL)
contacts = send_signal_to_our_processes(sig=SIGKILL, filter=program_exp)
log("Sending SIG=%d to %s" % (SIGKILL, str(contacts)))
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--mode', choices=['classification', 'regression'])
parser.add_argument('--model-dir')
parser.add_argument('--train-csv')
parser.add_argument('--test-csv')
parser.add_argument('--prediction-csv')
args = parser.parse_args()
automl = AutoML(args.model_dir)
log('####### cur time = ' +
str(datetime.datetime.now().strftime("%Y/%m/%d %H:%M:%S")))
# path_pred
automl.config['path_pred'] = args.model_dir
if args.train_csv is not None:
log('automl train...')
automl.train(args.train_csv, args.mode)
automl.save()
elif args.test_csv is not None:
log('automl predict...')
automl.load()
automl.predict(args.test_csv, args.prediction_csv)
else:
exit(1)
# end
log('####### cur time = ' +
str(datetime.datetime.now().strftime("%Y/%m/%d %H:%M:%S")))
def main():
# python main.py --mode regression --train-csv '../sdsj2018_automl_check_datasets/check_1_r/train.csv' --model-dir '../model/'
# python main.py --test-csv '../sdsj2018_automl_check_datasets/check_1_r/test.csv' --prediction-csv '../sdsj2018_automl_check_datasets/check_1_r/prediction.csv' --model-dir '../model/'
parser = argparse.ArgumentParser()
# parser.add_argument('--train-csv'
# , default="../../sdsj2018_automl_check_datasets/check_3_r/train.csv"
# )
# parser.add_argument('--test-csv'
# , default='../../sdsj2018_automl_check_datasets/check_3_r/test.csv'
# )
# parser.add_argument('--prediction-csv'
# , default='../../sdsj2018_automl_check_datasets/check_3_r/prediction.csv'
# )
# parser.add_argument('--mode', choices=['classification', 'regression']
# , default="classification"
# )
# parser.add_argument('--model-dir', default="../../model/")
parser.add_argument('--mode', choices=['classification', 'regression'])
parser.add_argument('--model-dir')
parser.add_argument('--train-csv')
parser.add_argument('--test-csv')
parser.add_argument('--prediction-csv')
parser.add_argument('--verbose', default=2)
args = parser.parse_args()
verbose = int(args.verbose)
if not args.train_csv is None:
params_autoML = \
{ 'memory': {'max_size_mb': 2*1024,
'max_size_train_samples': 100000
},
'field_target_name': 'target',
'pipeline' : {
'rename\nid_columns': {'node':'rename_id_columns',
'parents':None, #'args':{}
},
'check\ncolumns\nexists_start': {'node':'check_columns_exists',
'parents':'rename\nid_columns',
'args':{'key_stage':'preprocess_begin', 'drop_columns_test':True}
},
'drop_columns': {'node':'drop_columns',
'parents':'check\ncolumns\nexists_start', # 'args':{}
},
'fillna': {'node':'fillna',
'parents':'drop_columns', #
'args':{'args':{"number_": {'agg': 'median'},
"datetime_": {'agg': 'median'},
"string_": {'value': ''},
}},
},
'to_int8': {'node':'to_int8',
'parents':'fillna', # 'args':{}
},
'non_negative\ntarget_detect': {'node':'non_negative_target_detect',
'parents':'to_int8', # 'args':{}
},
'subsample0': {'node':'subsample',
'parents':'non_negative\ntarget_detect', # 'args':{}
},
'transform\ndatetime': {'node':'transform_datetime',
'parents':'subsample0', # 'args':{}
},
'transform\ncategorical': {'node':'transform_categorical',
'parents':'transform\ndatetime', # 'args':{}
},
# 'scale': {'node':'scale',
# 'parents':'transform\ncategorical', # 'args':{}
# },
# # # 'feature_generation': {'node':'feature_generation',
# # # 'parents':'scale', # 'args':{}
# # # },
'subsample1': {'node':'subsample',
'parents':'transform\ncategorical', #'scale', # 'args':{}
},
'columns_float64\nto_float32': {'node':'columns_float64_to_32',
'parents':'subsample1', # 'args':{}
},
'check\ncolumns\nexists_end': {'node':'check_columns_exists',
'parents':'columns_float64\nto_float32',
'args':{'key_stage':'preprocess_end', 'drop_columns_test':True}
},
# # 'split_X_y': {'node':'split_X_y',
# # 'parents':'check\ncolumns\nexists_end', # 'args':{}
# # },
# 'bayes': {'node':'model',
# 'parents':'check\ncolumns\nexists_end',
# 'args':{'models': ['bayes']}
# },
'lightgbm\nend': {'node':'model',
'parents':'columns_float64\nto_float32', #'bayes',
'args':{'models': ['lightgbm']}
},
# '': {'node':'',
# 'parents':'', # 'args':{}
# },
# '': {'node':'',
# 'parents':'', # 'args':{}
# },
}
}
automl = AutoML(args.model_dir, params=params_autoML, verbose=verbose)
# automl.pipeline_draw(view=True)
automl.train(args.train_csv, args.mode)
automl.save()
elif args.test_csv is not None:
automl = AutoML(args.model_dir, params={}, verbose=verbose)
automl.load()
_, score = automl.predict(args.test_csv, args.prediction_csv)
if verbose: print('score', score)
else:
exit(1)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--mode', choices=['classification', 'regression'])
parser.add_argument('--model-dir')
parser.add_argument('--train-csv')
parser.add_argument('--test-csv')
parser.add_argument('--prediction-csv')
parser.add_argument('--test-target-csv')
args = parser.parse_args()
if args.model_dir is None:
tests = {
1: 'regression',
2: 'regression',
3: 'regression',
4: 'classification',
5: 'classification',
6: 'classification',
7: 'classification',
8: 'classification',
}
for i in [3]: # tests.keys():
folder = r'..\check_' + str(i) + '_' + tests[i][0] + '\\'
argv = [
'--train-csv',
folder + 'train.csv',
'--test-csv',
folder + 'test.csv',
'--prediction-csv',
folder + 'prediction.csv',
'--test-target-csv',
folder + 'test-target.csv',
'--model-dir',
'.',
# '--nrows', '5000' if i in [3, 4, 5, 6, 7] else '500' if i in [8] else '-1',
'--mode',
tests[i]
]
args = parser.parse_args(argv)
logf('processing', folder)
automl = AutoML(args.model_dir)
if args.train_csv is not None:
automl.train(args.train_csv, args.mode)
automl.save()
log_trail('-', '\n')
if args.test_csv is not None:
automl.load()
automl.predict(args.test_csv, args.prediction_csv)
log_trail('-', '\n')
else:
automl = AutoML(args.model_dir)
if args.train_csv is not None:
automl.train(args.train_csv, args.mode)
automl.save()
log_trail('-', '\n')
if args.test_csv is not None:
automl.load()
automl.predict(args.test_csv, args.prediction_csv)
log_trail('-', '\n')
log_trail('=', '\n\n')