-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathrun_mdr.py
192 lines (156 loc) · 8.27 KB
/
run_mdr.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
import json
import signal
import sys
# import ray
import torch
import argparse
import pandas as pd
import numpy as np
from datetime import datetime
from types import SimpleNamespace
# from deepctr_torch.callbacks import ModelCheckpoint, EarlyStopping
# from ray import tune
# from ray.tune.suggest.hyperopt import HyperOptSearch
from hyperopt import fmin, tpe, Trials
from models.optimization import process_model_config, HyperOptCoordinator, Trainer
from utils.aux_funcs import init_seed, check_model_config, domain_map
from utils.constants import *
from utils.datasets import PyGDataset, DeepCTRDataset, create_dataset
from utils.in_out import print_exp_analysis, get_config, write_results, write_best_config, print_model_config,\
print_common_config
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--domains', type=lambda s: sorted(s.split(',')), default='Patio Lawn and Garden,Office Products') # default='Video Games,Toys and Games')
parser.add_argument('--graph-type', type=str, default='flattened', help="Possible values: flattened, disjoint, interacting, separate-shared")
parser.add_argument('--dataset', type=str, default='amazon')
parser.add_argument('--model-name', type=str, default='MAGRec')
parser.add_argument('--model-variant', type=str, default='', help='Only used for MAGRec: MemGNN, ASAP, SURGE')
parser.add_argument('--verbose', type=int, default=2)
parser.add_argument('--use-ray', dest='use_ray', action='store_true')
parser.add_argument('--no-use-hyperopt', dest='use_hyperopt', action='store_false') # Ignored if use_ray == True
parser.add_argument('--force-cpu', dest='force_cpu', action='store_true')
parser.add_argument('--test', dest='test', action='store_true')
parser.add_argument('--no-save-results', dest='save_results', action='store_false')
parser.set_defaults(use_ray=False, use_hyperopt=True, force_cpu=False, test=False, save_results=True)
args, _ = parser.parse_known_args()
if 'dummy' not in args.domains[0]:
args.domains = [domain_map(domain) for domain in args.domains]
config = get_config(args)
graph_type = config["global_args"]["graph_type"]
if args.model_name in ['MAGRec']:
exp_name = f'{args.model_name}_{args.model_variant}_{args.dataset}_{"_".join(args.domains)}_{graph_type}'
else:
exp_name = f'{args.model_name}_{args.dataset}_{"_".join(args.domains)}_{graph_type}'
config['global_args']['exp_name'] = exp_name
config['global_args']['device'] = 'cpu'
if torch.cuda.is_available() and not args.force_cpu:
config['global_args']['device'] = 'cuda'
if args.test:
config['hyperoptimization']['ho_max_evals'] = 2
config['optimization']['n_epochs'] = 2
config['global_args']['save_results'] = False
print_common_config(config)
model_space, space_size = process_model_config(config.get(MODEL_SPACE, {}))
run_config(config, args, model_space, space_size, exp_name)
def run_config(config, args, model_space, space_size, exp_name):
if args.use_ray:
raise NotImplementedError('Ray version does not currently work due to Signal Termination 15')
# ray.init(num_gpus=2)
#
# resources_per_trial = {'gpu': config['hyperoptimization'].get('gpu_per_trial', 0),
# 'cpu': config['hyperoptimization']['cpu_per_trial']}
# if resources_per_trial['gpu'] == 0 and config['global_args']['device'] == 'cuda':
# config['global_args']['device'] = 'cpu'
# raise Warning('GPU resources per trial are set to 0 even though there are GPUs available!')
#
# hyperopt_search = HyperOptSearch(space=model_space,
# n_initial_points=1,
# metric=config['evaluation']['val_metric'],
# mode=config['evaluation']['val_mode'],
# random_state_seed=RNG_SEED)
#
# analysis = tune.run(
# tune.with_parameters(objective_fn, config=config),
# local_dir='./ray_results',
# name=exp_name,
# search_alg=hyperopt_search,
# num_samples=min(space_size, config['hyperoptimization']['ho_max_evals']),
# metric=config['evaluation']['val_metric'],
# mode=config['evaluation']['val_mode'],
# verbose=0,
# resources_per_trial=resources_per_trial,
# max_failures=0)
#
# print_exp_analysis(analysis)
elif args.use_hyperopt:
ho_coordinator = HyperOptCoordinator(config, objective_fn)
trials = Trials()
best_trial = fmin(ho_coordinator.objective,
model_space,
algo=tpe.suggest,
max_evals=min(space_size, config['hyperoptimization']['ho_max_evals']),
trials=trials,
rstate=np.random.RandomState(RNG_SEED))
# if config['global_args']['save_results']:
# write_best_config('./', exp_name, {'trial_params': best_trial, 'fixed_params': config[DEFAULT_MODEL_CONF],
# })
else:
config['global_args']['verbose'] = 1
objective_fn(None, config=config)
print('Run config finished!')
def objective_fn(args, **params):
init_seed()
working_dir = os.getcwd()
config = params['config']
model_config = config[DEFAULT_MODEL_CONF].copy()
if args is not None:
model_config.update(args)
global_cfg = SimpleNamespace(**config['global_args'])
data_cfg = SimpleNamespace(**config['dataset'])
optim_cfg = SimpleNamespace(**config['optimization'])
eval_cfg = SimpleNamespace(**config['evaluation'])
model_cfg = SimpleNamespace(**model_config)
global_cfg.working_dir = working_dir
global_cfg.start_date = datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%SZ")
check_model_config(global_cfg.graph_type, model_cfg, global_cfg.model_name)
print_model_config(model_cfg)
results = mdr_pipeline(global_cfg, data_cfg, model_cfg, optim_cfg, eval_cfg)
file_name = f'{global_cfg.model_name}_{global_cfg.start_date}.json'
if global_cfg.save_results:
write_results(results, working_dir, global_cfg.exp_name, file_name, model_cfg)
return results, file_name
def mdr_pipeline(global_cfg, data_cfg, model_cfg, optim_cfg, eval_cfg):
# Create graph dataset
deepctr_models = [name for name, m in sys.modules['deepctr_torch.models'].__dict__.items() if isinstance(m, type)]
if global_cfg.model_name in deepctr_models:
dataset_class = DeepCTRDataset
elif global_cfg.model_name in GRAPH_MODELS:
dataset_class = PyGDataset
dataset = create_dataset(domains=global_cfg.domains, dataset=global_cfg.dataset, min_wsize=data_cfg.min_win_size,
max_wsize=data_cfg.max_win_size, ctr_ratio_range=data_cfg.ctr_ratio_range,
split_type=data_cfg.split_type, rebuild=data_cfg.rebuild, kcore=data_cfg.kcore,
graph_type=global_cfg.graph_type, verbose=global_cfg.verbose,
dataset_class=dataset_class, batch_size=optim_cfg.batch_size, use_seq=data_cfg.use_seq)
# ASSERT number of edges is the same in edge_attr and edge_index
# for i in tqdm(range(len(dataset))):
# g = dataset.get(i)
# assert g.edge_attr.shape[0] == g.edge_index.shape[1]
if global_cfg.verbose == 1:
dataset.print_split_info()
trainer = Trainer(global_cfg, model_cfg, optim_cfg, eval_cfg, dataset)
results = trainer.fit_predict(dataset)
print('These are the results')
# Uncomment for pretty print of the results
# print(json.dumps(results, indent=4))
print(results)
return results
def handler(signum, frame):
# print('Signal handler called with signal', signum)
# print('frame:', frame)
pass
catchable_sigs = [signal.SIGWINCH, signal.SIGTERM]
for sig in catchable_sigs:
signal.signal(sig, handler)
if __name__ == '__main__':
pd.options.mode.chained_assignment = None
main()