-
Notifications
You must be signed in to change notification settings - Fork 0
/
search.py
260 lines (200 loc) · 8.44 KB
/
search.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
import os
import redis
import os.path as osp
import argparse
import numpy as np
from tqdm import tqdm
from math import log
from collections import defaultdict
import torch
import networkx as nx
import scipy.stats as stats
from multiprocessing import Pool
from codescholar.sast.visualizer import render_sast
from codescholar.sast.sast_utils import sast_to_prog
from codescholar.representation import config
from codescholar.search import search_config
from codescholar.search.grow import grow
from codescholar.search.elastic_search import grep_programs
from codescholar.search.init_search import init_search_q, init_search_m, init_search_mq
from codescholar.utils.search_utils import (
ping_elasticsearch,
ping_elasticindex,
wl_hash,
save_idiom,
_print_mine_logs,
_write_mine_logs,
read_graph,
)
from codescholar.utils.search_utils import load_embeddings_batched_redis
from codescholar.utils.graph_utils import nx_to_program_graph
from codescholar.utils.cluster_utils import cluster_programs
from codescholar.utils.perf import perftimer
from codescholar.constants import DATA_DIR
redis_client = redis.StrictRedis(host="localhost", port=6379, db=0)
######### IDIOM STORE ############
def _save_idiom_generation(args, idiommine_gen) -> bool:
"""save the current generation of idioms to disk.
and return if the search should continue.
"""
idiom_clusters = idiommine_gen.items()
idiom_clusters = list(sorted(idiom_clusters, key=lambda x: len(x[1]), reverse=True))
cluster_id, total_nhoods, total_idioms = 1, 0, 0
for _, idioms in idiom_clusters:
idioms = list(sorted(idioms, key=lambda x: x[1], reverse=True))
for idiom, nhoods, holes in idioms:
size_id, nhood_count = len(idiom), int(nhoods)
if args.mode == "mq":
if nx.number_connected_components(nx.to_undirected(idiom)) != 1:
continue
if nhood_count < args.min_nhoods:
continue
file = "idiom_{}_{}_{}_{}".format(size_id, cluster_id, nhood_count, holes)
path = f"{args.idiom_g_dir}{file}.png"
sast = nx_to_program_graph(idiom)
# NOTE @manishs: when growing graphs in all directions
# the root can get misplaced. Find root = node with no incoming edges!
root = [n for n in sast.all_nodes() if sast.incoming_neighbors(n) == []][0]
sast.root_id = root.id
if args.render:
render_sast(sast, path, spans=True, relpos=True)
path = f"{args.idiom_p_dir}{file}.py"
prog = sast_to_prog(sast).replace("#", "_")
if args.mode == "mq":
prog = "\n".join(
[line for line in prog.split("\n") if line.strip() != "_"]
)
save_idiom(path, prog)
# update counts
total_nhoods += nhood_count
total_idioms += 1
cluster_id += 1
# metrics
reusability = total_nhoods / total_idioms if total_idioms > 0 else 0
lreusability = log(reusability + 1 if reusability <= 0 else reusability)
diversity = len(idiom_clusters)
ldiversity = log(diversity + 1 if diversity <= 0 else diversity)
if args.stop_at_equilibrium and ldiversity >= lreusability:
return False
else:
return True
######### MAIN ############
# @perftimer
def search(args, prog_indices, beam_sets):
mine_summary = defaultdict(lambda: defaultdict(int))
size = 1
if not beam_sets:
print("Oops, BEAM SETS ARE EMPTY!")
return mine_summary
num_gpus = torch.cuda.device_count()
# create a pool of workers for each GPU
pools = [Pool(args.n_workers) for _ in range(num_gpus)]
continue_search = True
while continue_search and len(beam_sets) != 0:
results = []
for i, beam_set in enumerate(beam_sets):
gpu = i % num_gpus
# use apply_async to submit the grow task to the pool
result = pools[gpu].apply_async(grow, (args, prog_indices, beam_set, gpu))
results.append(result)
# idioms for generation i
idiommine_gen = defaultdict(list)
new_beam_sets = []
# beam search over this generation
pbar = tqdm(total=len(beam_sets), desc=f"[search {size}]")
for result in results:
new_beams = result.get() # Wait for the result and get it
pbar.update(1)
# candidates from only top-scoring beams in the beam set
for new_beam in new_beams[:1]:
score, holes, neigh, _, _, graph_idx = new_beam
graph = read_graph(args, graph_idx)
neigh_g = graph.subgraph(neigh).copy()
neigh_g.remove_edges_from(nx.selfloop_edges(neigh_g))
for v in neigh_g.nodes:
neigh_g.nodes[v]["anchor"] = 1 if v == neigh[0] else 0
neigh_g_hash = wl_hash(neigh_g)
idiommine_gen[neigh_g_hash].append((neigh_g, score, holes))
mine_summary[len(neigh_g)][neigh_g_hash] += 1
if len(new_beams) > 0:
new_beam_sets.append(new_beams)
pbar.close()
# save generation
beam_sets = new_beam_sets
size += 1
# _print_mine_logs(mine_summary)
if size >= args.min_idiom_size and size <= args.max_idiom_size:
continue_search = _save_idiom_generation(args, idiommine_gen)
# Terminate all pools
for pool in pools:
pool.terminate()
pool.join()
return mine_summary
def main(args):
if (args.mode == "q" or args.mode == "mq") and args.seed is None:
parser.error("query modes require --seed to begin search.")
if not ping_elasticsearch():
raise ConnectionError(
"Elasticsearch not running on localhost:9200! Please start Elasticsearch and try again."
)
if not ping_elasticindex():
raise ValueError(
"Elasticsearch index `python_files` not found! Please run `elastic_search.py` to create the index."
)
# sample and constrain the search space
if args.mode == "mq":
prog_indices = set()
for i, seed in enumerate(args.seed.split(";")):
if i == 0:
prog_indices = set(grep_programs(args, seed))
else:
prog_indices = prog_indices & set(grep_programs(args, seed))
prog_indices = list(prog_indices)[: args.prog_samples]
else:
prog_indices = grep_programs(args, args.seed)[: args.prog_samples]
# load all embeddings of prog_indices to redis
# TODO: do this offline for *all* progs?
load_embeddings_batched_redis(args, prog_indices)
# identify seed programs by clustering
if args.mode in ["q", "mq"]:
seed_indices = cluster_programs(args, prog_indices, n_clusters=10)
# STEP 1: initialize search space
if args.mode == "q":
beam_sets = init_search_q(args, seed_indices, seed=args.seed)
elif args.mode == "mq":
beam_sets = init_search_mq(args, seed_indices, seeds=args.seed.split(";"))
elif args.mode == "m":
prog_indices = grep_programs(args, args.seed)[: args.prog_samples]
beam_sets = init_search_m(args, prog_indices)
else:
raise ValueError(f"Invalid search mode {args.mode}!")
# STEP 2: search for idioms; saves idioms gradually
mine_summary = search(args, prog_indices, beam_sets)
_write_mine_logs(mine_summary, f"{args.result_dir}/mine_summary.log")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
config.init_optimizer_configs(parser)
config.init_encoder_configs(parser)
search_config.init_search_configs(parser)
args = parser.parse_args()
# data config
args.prog_dir = f"{DATA_DIR}/{args.dataset}/source/"
args.source_dir = f"{DATA_DIR}/{args.dataset}/graphs/"
args.emb_dir = f"{DATA_DIR}/{args.dataset}/emb/"
args.result_dir = (
f"./results/{args.seed}/"
if (args.mode == "q" or args.mode == "mq")
else "./results/"
)
args.idiom_g_dir = f"{args.result_dir}/idioms/graphs/"
args.idiom_p_dir = f"{args.result_dir}/idioms/progs/"
# model config
args.test = True
args.model_path = f"../representation/ckpt/model.pt"
args.batch_size = 512
if args.render and not osp.exists(args.idiom_g_dir):
os.makedirs(args.idiom_g_dir)
if not osp.exists(args.idiom_p_dir):
os.makedirs(args.idiom_p_dir)
torch.multiprocessing.set_start_method("spawn")
main(args)