-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathrun_ethereum_experiment.py
101 lines (84 loc) · 4.58 KB
/
run_ethereum_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
import argparse
import os
import sys
import warnings
import numpy as np
import pandas as pd
import torch
import src.dataprocessing as dataproc
import src.training as train_n2f
import src.experimentutils as experutils
import src.runutils as runutils
import src.utils
parser = argparse.ArgumentParser(description="Run the training on ethereum graph")
parser.add_argument("path", type=str, help="path for saving submission_results")
parser.add_argument("--data_path", type=str, help="the number of dims to search",
default="data/preprocessed_ethereum_2018_2020.csv")
parser.add_argument("--split_path", type=str, help="Path to train, val, test split folder",
default=None)
parser.add_argument("--node2vec_path", type=str, help="Path to pretrained node2vec embeddings",
default=None)
parser.add_argument("--config_path", type=str, help="path were the configs can be read",
default="configs_eth/eth_config_auto_noise.json")
parser.add_argument("--max_dim", type=int, help="the number of dims to search", default=3)
parser.add_argument("--min_dim", type=int, help="the number of dims to search", default=1)
parser.add_argument("--seed", type=int, help="random seed", default=1234)
parser.add_argument("--use_gpu", action='store_true', help="Use gpu")
parser.add_argument("--gates_init", type=str, help="Which gates init to use", default='zeros')
parser.add_argument("--only_baselines", action='store_true', help="Only run the baselines")
parser.add_argument("--skip_baselines", action='store_true', help="Only run the baselines")
parser.add_argument("--verbosity", type=int, help="verbosity level", default=1)
parser.add_argument("--debug", action='store_true', help="Run in debug mode, only one epoch per model")
args = parser.parse_args()
chp_folder = os.path.join(args.path, "chpt")
os.makedirs(args.path, exist_ok=True)
os.makedirs(chp_folder, exist_ok=True)
dev = "cuda" if torch.cuda.is_available() and args.use_gpu else "cpu"
device = torch.device(dev)
if device.type == 'cuda' and args.verbosity > 0:
print(f"Using {torch.cuda.get_device_name(0)}")
experutils.set_seeds(args.seed)
configs = src.utils.load_configs(args.config_path)
configs['hyperpara_config'].emb_grad_noise = src.utils.GradientNoise(False, np.inf, 0.)
configs['hyperpara_config'].gates_grad_noise = src.utils.GradientNoise(False, np.inf, 0.)
configs['baseline_hyperpara_config'].grad_noise = src.utils.GradientNoise(False, np.inf, 0.)
configs['init_config'].gates_init = args.gates_init
configs['train_config'].max_steps = 1e5
configs['train_config'].tol = 1e-5
configs['train_config'].substep_tol = 1e-5
configs['init_config'].max_steps = 1e4
fg_max_iter = 100
if args.debug:
warnings.warn("RUNNING WITH DEBUG CONFIGURATIONS. MODELS WONT BE TRAINED.")
configs['train_config'].max_steps = 0
configs['init_config'].embedding_init = 'zeros'
configs['init_config'].gates_init = 'zeros'
configs['init_config'].max_steps = 0
fg_max_iter = 1
# fg_max_iter = None
emb_reg_weights = [3., 1, 0.3, 0.1]
gates_reg_weights = [3., 1., 0.3, 0.1]
q_indices = [0, 1, 2]
graph = dataproc.Graph.read_csv(args.data_path)
num_nodes = graph.num_vertices()
flow_summary = dataproc.flow_summary(graph.flow)
pd.DataFrame([flow_summary]).to_csv(os.path.join(args.path, f"ethereum_flow_info.csv"), header=True, index=False)
if args.split_path is None:
train_graph, val_graph, test_graph = graph.split_train_val_test_graphs((0.7, 0.15, 0.15))
train_graph.to_csv(os.path.join(args.path, f"preprocessed_ethereum_train.csv"))
val_graph.to_csv(os.path.join(args.path, f"preprocessed_ethereum_val.csv"))
test_graph.to_csv(os.path.join(args.path, f"preprocessed_ethereum_test.csv"))
else:
train_graph = dataproc.Graph.read_csv(os.path.join(args.split_path, f"preprocessed_ethereum_train.csv"))
val_graph = dataproc.Graph.read_csv(os.path.join(args.split_path, f"preprocessed_ethereum_val.csv"))
test_graph = dataproc.Graph.read_csv(os.path.join(args.split_path, f"preprocessed_ethereum_test.csv"))
base = train_n2f.TrainerBase(train_graph=train_graph, val_graph=val_graph, device=device,
loss_config=configs['loss_config'])
model_factory = experutils.ModelFactory(base=base)
if not args.skip_baselines:
runutils.run_eth_hp_search_nn_baselines(model_factory, emb_reg_weights, q_indices, fg_max_iter, args, configs)
else:
print("====> Skipping baselines")
if args.only_baselines:
sys.exit("Finished baselines with option 'only_baselines', exiting.")
runutils.run_eth_hp_search_gradient_models(model_factory, emb_reg_weights, gates_reg_weights, args, configs)