/
experiment.py
executable file
·212 lines (165 loc) · 7.3 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
212
import os
import pdb
import copy
import json
import logging
import argparse
import pickle as pkl
from datetime import datetime
import numpy as np
import pandas as pd
import multiprocessing as mp
from joblib import Parallel, delayed
from utils import sample_random_nodes
from synthgen import build_graph
from lib.cits.citf import CITFactory
from lib.cits.ci_eval import EvalCI
logger = logging.getLogger(__file__)
rmasks = {
'1a' : '000',
'1b' : '100',
'1c' : '110',
'2a' : '000',
'2b' : '100',
'2c' : '110',
'2d' : '001',
'2e' : '101',
'2f' : '111',
}
def init_results(target, algo_names):
results = {target: []}
for name in algo_names:
results[name + '_type_i'] = []
results[name + '_type_ii'] = []
results[name + '_aupc'] = []
results[name + '_times'] = []
return results
def fill_results(results, algo_names, target, target_vals, rdfs, ):
for i in range(len(target_vals)):
target_value = target_vals[i]
results[target].append(target_value)
for name in algo_names:
results[name + '_type_i'].append(rdfs[i].loc[name]['type_i'])
results[name + '_type_ii'].append(rdfs[i].loc[name]['type_ii'])
results[name + '_aupc'].append(rdfs[i].loc[name]['aupc'])
avg_time = (rdfs[i].loc[name]['times_null'] + rdfs[i].loc[name]['times_alt']) / 2
avg_time = avg_time / 60
results[name + '_times'].append(avg_time)
return results
def dump_results(results, out_file):
def dump_metric(suffix):
columns = list(results.columns)
suffix_tail = suffix.split('_')[-1]
t_cols = list(filter(lambda x: x.split('_')[-1]==suffix_tail in x, columns[1:]))
t_df = results[[columns[0]] + t_cols]
t_out_file = out_file.split('.')[0] + '_%s.csv' % suffix
t_df.to_csv(t_out_file, index=False)
dump_metric('exec_times')
dump_metric('type_i')
dump_metric('type_ii')
dump_metric('aupc')
def get_target_parent(g_config, target):
return g_config["alt_params"] if target in g_config["alt_params"] else g_config["params"]
def run_target_task(i, config, alpha, log_file, load_cache=False):
'''
Run experiment for i'th target value in config.
Return evaluation result as a dataframe
'''
# parse config
g_config = copy.deepcopy(config["graph"])
is_conditional = config["case"][0] == '2'
test_type = 'conditional' if is_conditional else 'marginal'
g_config["params"]["test_type"] = test_type
# init algo objs
algos = []
for name in config["algos"]:
algos.append((name, CITFactory.get_cit(name, config["seed"])))
# set target value for this taks
target_parent = get_target_parent(g_config, config["target"])
target_vals = copy.deepcopy(target_parent[config["target"]])
target_value = target_vals[i] # i'th target value
target_parent[config["target"]] = target_value
# init evaluator
evaluator = EvalCI(a=alpha, names=config["algos"])
evaluator.initialize()
approximate = config["approx"] if "approx" in config else False
def itest(null=False):
gc = copy.deepcopy(g_config)
params = gc["null_params"] if null else gc["alt_params"]
gc["params"].update(params)
g = build_graph(gc, config["case"], config["seed"] + trial)
#print("graph built")
samples = []
if config["target"] == 'sample_size':
_, samples = sample_random_nodes(g, target_value)
for algo in algos:
name, cit = algo
start = datetime.now()
p_val = cit.run_test(g=g, cond=is_conditional, rmask=rmasks[config["case"]], samples=list(samples), approx=approximate)
elapsed = (datetime.now() - start).seconds
evaluator.add_result(name, p_val, null=null, time=elapsed)
log_file = '{prefix}_{index}.csv'.format(prefix=log_file.split('.')[0], index=i)
if load_cache:
evaluator.load_p_vals(log_file)
else:
# run trials
for trial in range(config["num_trials"]):
itest(null=True)
itest(null=False)
print('%s:: trial %d task %d done' % (datetime.now(), trial, i))
evaluator.log_p_vals(log_file)
rdf = evaluator.gen_result()
# print(rdf)
return (i, rdf)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("-config", type=str, default='conf/exp_1a_dep.json', help="config", required=True)
parser.add_argument("-s", type=int, default=-1, help="seed", required=False)
parser.add_argument("-a", type=float, default=0.05, help="significance level: alpha", required=False)
parser.add_argument("-d", type=int, default=0, help="debug mode (0/1)", required=False)
parser.add_argument("-nn", type=int, default=-1, help="number of nodes", required=False)
parser.add_argument("-nt", type=int, default=-1, help="number of trials", required=False)
parser.add_argument("-jobs", type=int, default=-1, help="number of parallel jobs", required=False)
parser.add_argument("--cache", action='store_true', help="load p-vals from cache?", required=False)
parser.add_argument("--nop", action='store_true', help="don't run parallel?", required=False)
args = parser.parse_args()
# logging.basicConfig(filename='logs/experiment.log', mode='a', format='%(asctime)-15s :: %(message)s')
logger.propagate = False
logger.setLevel(logging.DEBUG if args.d else logging.INFO)
fh = logging.FileHandler('logs/experiment.log', mode='w')
fh.setFormatter(logging.Formatter('%(asctime)s :: %(levelname)s :: %(message)s'))
fh.setLevel(logging.INFO)
logger.addHandler(fh)
# parse config, override if necessary
config = json.loads(open(args.config, 'r').read())
config["seed"] = args.s if args.s != -1 else config["seed"]
config["num_trials"] = args.nt if args.nt != -1 else config["num_trials"]
if "num_nodes" in config["graph"]:
config["graph"]["num_nodes"] = args.nn if args.nn != -1 else config["graph"]["num_nodes"]
target = config["target"]
exp_name = args.config.split('/')[-1].split('.')[0]
config['exp_name'] = exp_name
# init results dict
results = init_results(target, config["algos"])
# get target values
target_parent = get_target_parent(config["graph"], target)
target_vals = copy.deepcopy(target_parent[target])
# run target task
log_file = 'logs/{prefix}.csv'.format(prefix=exp_name)
logger.info('{name} started'.format(name=exp_name))
if args.nop:
rdfs = [run_target_task(i, config, args.a, log_file, args.cache) for i in range(len(target_vals))]
else:
num_jobs = args.jobs if args.jobs != -1 else min(mp.cpu_count(), len(target_vals))
rdfs = Parallel(n_jobs=num_jobs)(delayed(run_target_task)(i, config, args.a, log_file, args.cache) for i in range(len(target_vals)))
# prepare results
rdfs = [res[1] for res in sorted(rdfs, key=lambda x: x[0])]
results = fill_results(results, config["algos"], target, target_vals, rdfs)
results = pd.DataFrame(data=results)
print(results)
# store results
out_file = 'out/' + args.config.split('/')[1].split('.')[0] + '.csv'
dump_results(results, out_file)
logger.info('{name} finished'.format(name=exp_name))
if __name__ == '__main__':
main()