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
|
#!/usr/bin/env python3
"""Audit and mechanically select the frozen ResNet P1 learning rates."""
import argparse
import glob
import hashlib
import json
import math
import os
import sys
ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.insert(0, ROOT)
from experiments.resnet_crossover_grid import p1_jobs, registry_sha256
GRIDDED = ("fa", "dfa", "pepita", "ep", "dualprop")
def sha256(path):
digest = hashlib.sha256()
with open(path, "rb") as handle:
for chunk in iter(lambda: handle.read(1024 * 1024), b""):
digest.update(chunk)
return digest.hexdigest()
def read_json(path):
with open(path, encoding="utf-8") as handle:
return json.load(handle)
def history_metrics(record, method):
if method == "ff":
layers = record["layers"]
completed = sum(len(layer["epochs"]) for layer in layers)
expected = record["architecture"]["depth"]
assert len(layers) == expected
assert all(len(layer["epochs"]) == 10 for layer in layers)
losses = [
float(epoch["loss"]) for layer in layers
for epoch in layer["epochs"]]
validation = float(record["final"]["accuracy"])
final_loss = losses[-1]
best_epoch = None
elif record["args"].get("method") in (
"pepita", "ep", "dualprop"):
epochs = record["epochs"]
assert len(epochs) == 10
validations = [
float(epoch["validation"]["accuracy"]) for epoch in epochs]
losses = [float(epoch["train_loss"]) for epoch in epochs]
validation = float(record["final"]["accuracy"])
completed = len(epochs)
best_epoch = max(range(len(validations)), key=validations.__getitem__) + 1
final_loss = float(record["final"]["loss"])
else:
epochs = record["epochs"]
assert len(epochs) == 10
validations = [float(epoch["eval_accuracy"]) for epoch in epochs]
losses = [float(epoch["train_loss"]) for epoch in epochs]
validation = float(record["final"]["accuracy"])
completed = len(epochs)
best_epoch = max(range(len(validations)), key=validations.__getitem__) + 1
final_loss = float(record["final"]["loss"])
if method == "ff":
best_validation = validation
else:
best_validation = max(validations + [validation])
return {
"best_validation_accuracy": best_validation,
"final_validation_accuracy": validation,
"final_validation_loss": final_loss,
"best_epoch": best_epoch,
"epochs_completed": completed,
"all_training_losses_finite": all(
math.isfinite(value) for value in losses),
}
def audit_job(job, expected_source):
manifest_path = job["output"] + ".manifest.json"
assert os.path.isfile(manifest_path), (
f"missing manifest for {job['experiment_name']}")
manifest = read_json(manifest_path)
for key in (
"stage", "method", "architecture", "rate", "experiment_name",
"output", "timeout_seconds", "command"):
assert manifest[key] == job[key], (
f"{job['experiment_name']}: drift in {key}")
assert manifest["source"] == expected_source
hardware = manifest["hardware_lock"]
assert hardware["physical_gpu_index"] in (5, 7)
assert hardware["physical_gpu_uuid"]
common = {
"method": job["method"],
"rate": job["rate"],
"experiment_name": job["experiment_name"],
"manifest": os.path.relpath(manifest_path, ROOT),
"status": manifest["status"],
"output_sha256": manifest["output_sha256"],
"driver_wall_seconds": manifest["driver_wall_seconds"],
"physical_gpu_index": hardware["physical_gpu_index"],
"physical_gpu_uuid": hardware["physical_gpu_uuid"],
}
if manifest["status"] != "completed":
assert manifest["status"] in {
"timeout", "nonzero_exit", "missing_output"}
return {
**common,
"finite": False,
"best_validation_accuracy": None,
"final_validation_accuracy": None,
"final_validation_loss": None,
"best_epoch": None,
"epochs_completed": None,
"all_training_losses_finite": False,
}
assert manifest["output_exists"] is True
assert manifest["output_sha256"] == sha256(job["output"])
record = read_json(job["output"])
assert record["provenance"]["git_commit"] == expected_source["git_commit"]
assert record["provenance"]["git_tracked_dirty"] is False
assert record["args"]["depth"] == 20
assert record["args"]["seed"] == 0
assert record["args"]["eval_split"] == "validation"
assert record["evaluation_protocol"]["test_evaluations"] == 0
assert record["evaluation_protocol"]["test_used_for_selection"] is False
assert record["split"]["validation_examples"] == 5000
assert record["split"]["split_seed"] == 2027
if job["method"] in ("pepita", "ff", "ep", "dualprop"):
assert record["args"]["method"] == job["method"]
else:
expected_mode = {
"bp": "bp",
"fa": "hfa",
"dfa": "dfa",
"clean_kp": "kp",
"sdil": "kp_traffic",
}[job["method"]]
assert record["args"]["mode"] == expected_mode
metrics = history_metrics(record, job["method"])
return {
**common,
"finite": (
metrics["all_training_losses_finite"]
and math.isfinite(metrics["final_validation_loss"])),
**metrics,
}
def choose(candidates):
return max(candidates, key=lambda row: (
int(row["finite"]),
(
row["best_validation_accuracy"]
if row["best_validation_accuracy"] is not None else -math.inf),
int(
row["final_validation_loss"] is not None
and math.isfinite(row["final_validation_loss"])),
-row["rate"],
))
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--out",
default="results/resnet_crossover/p1_selector.json")
parser.add_argument("--allow-partial", action="store_true")
args = parser.parse_args()
jobs = p1_jobs()
manifests = [
job["output"] + ".manifest.json" for job in jobs
if os.path.isfile(job["output"] + ".manifest.json")]
missing = [
job["experiment_name"] for job in jobs
if not os.path.isfile(job["output"] + ".manifest.json")]
if missing and not args.allow_partial:
raise AssertionError(f"missing P1 jobs: {missing}")
sources = [read_json(path)["source"] for path in manifests]
if sources:
expected_source = sources[0]
assert all(source == expected_source for source in sources)
else:
expected_source = None
records = [
audit_job(job, expected_source) for job in jobs
if os.path.isfile(job["output"] + ".manifest.json")]
selected = {}
launch = None
if not missing:
launch_path = os.path.join(
ROOT, "results", "resnet_crossover", "p1_launch.json")
assert os.path.isfile(launch_path), "missing ResNet P1 launch lock"
launch = read_json(launch_path)
assert launch["stage"] == "p1"
assert launch["source"] == expected_source
assert launch["registry_sha256"] == registry_sha256(jobs)
assert launch["num_jobs"] == len(jobs)
assert launch["allowed_physical_gpus"] == [5, 7]
for method in dict((job["method"], None) for job in jobs):
candidates = [
record for record in records if record["method"] == method]
winner = (
choose(candidates) if method in GRIDDED else candidates[0])
selected[method] = {
"rate": winner["rate"],
"best_validation_accuracy":
winner["best_validation_accuracy"],
"experiment_name": winner["experiment_name"],
}
report = {
"gate": "pass" if not missing else "partial",
"stage": "resnet_crossover_p1",
"selection_rule":
"maximum best validation accuracy, then finite final loss, "
"then lower learning rate",
"source": expected_source,
"num_expected_records": len(jobs),
"num_audited_records": len(records),
"missing_experiments": missing,
"test_policy": "none",
"launch_lock": launch,
"records": records,
"selected": selected,
}
os.makedirs(os.path.dirname(os.path.abspath(args.out)), exist_ok=True)
with open(args.out, "w", encoding="utf-8") as handle:
json.dump(report, handle, indent=2, sort_keys=True)
handle.write("\n")
print(json.dumps({
"gate": report["gate"],
"num_audited_records": len(records),
"missing_experiments": missing,
"selected": selected,
}, indent=2, sort_keys=True))
if __name__ == "__main__":
main()
|
