File size: 10,521 Bytes
7f7a890 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 | """
graph_hpo_sequence.py
=====================
Set up and optionally execute a graph-aware HPO -> joint insect+mammal training
-> threshold comparison -> mammal generalization pipeline.
The "graph-based" part is implemented as dependency-aware Optuna TPE:
- multivariate TPE
- grouped TPE
- median pruning
Artifacts are written under artifacts/graph_hpo/ so methodology and metrics are
preserved separately from the default training outputs.
"""
from __future__ import annotations
import argparse
import json
import shlex
import subprocess
from pathlib import Path
BASE = Path(__file__).parent.parent
ART = BASE / "artifacts"
GRAPH_ART = ART / "graph_hpo"
GENERALIZATION_ART = ART / "generalization"
def python_bin() -> str:
venv = BASE / ".venv" / "bin" / "python3"
return str(venv if venv.exists() else Path("python3"))
def build_paths() -> dict[str, Path]:
return {
"methodology_json": GRAPH_ART / "methodology.json",
"methodology_md": GRAPH_ART / "methodology.md",
"summary_json": GRAPH_ART / "sequence_summary.json",
"hpo_json": GRAPH_ART / "hpo_results.json",
"train_ckpt": GRAPH_ART / "graph_hpo_best.pth",
"train_thr": GRAPH_ART / "graph_hpo_best_thresholds.json",
"train_log": GRAPH_ART / "graph_hpo_best_log.json",
"generalization_json": GRAPH_ART / "generalization_results.json",
"generalization_csv": GRAPH_ART / "generalization_results.csv",
}
def methodology_payload(paths: dict[str, Path], args: argparse.Namespace) -> dict:
py = python_bin()
return {
"pipeline_name": "graph_hpo_joint_insect_mammal",
"objective": {
"type": "weighted_joint_micro_fmax",
"formula": "alpha * insect_val_micro_fmax + (1-alpha) * mammal_val_micro_fmax",
"alpha": args.alpha,
},
"graph_hpo": {
"sampler": "Optuna TPESampler",
"multivariate": True,
"group": True,
"pruner": "MedianPruner",
"startup_trials": args.startup_trials,
"warmup_steps": args.warmup_steps,
},
"feature_levels": {
"esm_only": 320,
"esm_seq": 331,
"esm_all": 360,
},
"datasets": {
"insect_base": "Important Files/merged_full_struct.parquet",
"insect_supp": "Important Files/merged_full_struct_with_features.parquet",
"mammal": "artifacts/generalization/mammal_full_v1.parquet",
"splits": "artifacts/splits/splits_n250000_seed42.npz",
"mlb": "Important Files/mlb_public_v1.pkl",
},
"stages": [
{
"name": "hpo",
"purpose": "Graph-aware joint insect+mammal hyperparameter tuning",
"output": str(paths["hpo_json"]),
},
{
"name": "train",
"purpose": "Train exact best feature slice with mammal data included",
"output": str(paths["train_ckpt"]),
},
{
"name": "thresholds",
"purpose": "Compare current thresholds vs precision+IC vs novelty-gated thresholds",
"output": str(ART / "threshold_comparison_results.json"),
},
{
"name": "generalization",
"purpose": "Evaluate mammal generalization with trained checkpoint and thresholds",
"output": str(paths["generalization_json"]),
},
{
"name": "summary",
"purpose": "Aggregate methodology and metrics into a single saved record",
"output": str(paths["summary_json"]),
},
],
"notes": [
"Threshold comparison currently saves metrics to artifacts/threshold_comparison_results.json.",
"Final training uses train_v3_fixed.py with explicit feature_level to match HPO output.",
"Mammal data are merged into training unless --skip-mammal is explicitly used.",
],
"stage_commands": {
"hpo": [
py, "scripts/hpo.py",
"--mammal", "artifacts/generalization/mammal_full_v1.parquet",
"--n_trials", str(args.n_trials),
"--epochs", str(args.hpo_epochs),
"--patience", str(args.hpo_patience),
"--alpha", str(args.alpha),
"--startup_trials", str(args.startup_trials),
"--warmup_steps", str(args.warmup_steps),
"--multivariate_tpe", "--group_tpe",
"--out", str(paths["hpo_json"]),
],
"train_template": [
py, "scripts/train_v3_fixed.py",
"--hidden", "<best.hidden>",
"--blocks", "<best.n_blocks>",
"--dropout", "<best.dropout>",
"--lr", "<best.lr>",
"--weight_decay", "<best.weight_decay>",
"--batch", "<best.batch>",
"--label_smooth", "<best.label_smooth>",
"--feature_level", "<best.feat_level>",
"--feature_label", "graph_hpo_best",
"--checkpoint_out", str(paths["train_ckpt"]),
"--threshold_out", str(paths["train_thr"]),
"--log_out", str(paths["train_log"]),
],
"threshold_comparison": [
py, "scripts/threshold_comparison.py",
],
"generalization_template": [
py, "scripts/eval_generalization.py",
"--checkpoint", str(paths["train_ckpt"]),
"--thresholds", str(paths["train_thr"]),
"--mlb", "Important Files/mlb_public_v1.pkl",
"--taxon_parquet", "artifacts/generalization/mammal_embeddings_v3.parquet",
"--taxon_name", "mammals_graph_hpo",
"--obo", "go-basic.obo",
"--out", str(paths["generalization_json"]),
],
},
}
def methodology_markdown(payload: dict) -> str:
lines = [
"# Graph HPO Training Sequence",
"",
"## Objective",
f"- `{payload['objective']['formula']}`",
f"- `alpha = {payload['objective']['alpha']}`",
"",
"## Graph-Aware HPO",
f"- Sampler: `{payload['graph_hpo']['sampler']}`",
f"- Multivariate: `{payload['graph_hpo']['multivariate']}`",
f"- Grouped: `{payload['graph_hpo']['group']}`",
f"- Pruner: `{payload['graph_hpo']['pruner']}`",
"",
"## Feature Levels",
]
for name, dim in payload["feature_levels"].items():
lines.append(f"- `{name}` -> `{dim}` dims")
lines += ["", "## Stages"]
for stage in payload["stages"]:
lines.append(f"- `{stage['name']}`: {stage['purpose']}")
lines.append(f" Output: `{stage['output']}`")
lines += ["", "## Notes"]
for note in payload["notes"]:
lines.append(f"- {note}")
lines += ["", "## Stage Commands"]
for name, cmd in payload["stage_commands"].items():
lines.append(f"- `{name}`")
lines.append(f" `{shlex.join(cmd)}`")
lines.append("")
return "\n".join(lines)
def build_commands(paths: dict[str, Path], args: argparse.Namespace) -> dict[str, list[str]]:
py = python_bin()
hpo_cmd = [
py,
str(BASE / "scripts" / "hpo.py"),
"--mammal", "artifacts/generalization/mammal_full_v1.parquet",
"--n_trials", str(args.n_trials),
"--epochs", str(args.hpo_epochs),
"--patience", str(args.hpo_patience),
"--alpha", str(args.alpha),
"--startup_trials", str(args.startup_trials),
"--warmup_steps", str(args.warmup_steps),
"--multivariate_tpe",
"--group_tpe",
"--out", str(paths["hpo_json"]),
]
return {
"hpo": hpo_cmd,
}
def run(cmd: list[str]) -> None:
print("$", shlex.join(cmd))
subprocess.run(cmd, cwd=BASE, check=True)
def load_json(path: Path):
if path.exists():
with open(path) as f:
return json.load(f)
return None
def write_summary(paths: dict[str, Path], payload: dict) -> None:
hpo = load_json(paths["hpo_json"])
threshold_metrics = load_json(ART / "threshold_comparison_results.json")
generalization = load_json(paths["generalization_json"]) or load_json(GENERALIZATION_ART / "generalization_results.json")
summary = {
"methodology": payload,
"artifacts": {k: str(v) for k, v in paths.items()},
"existing_metrics": {
"hpo": hpo,
"threshold_comparison": threshold_metrics,
"generalization": generalization,
},
}
with open(paths["summary_json"], "w") as f:
json.dump(summary, f, indent=2)
def main() -> None:
p = argparse.ArgumentParser(description="Set up / run graph-aware HPO training pipeline")
p.add_argument("--n_trials", type=int, default=40)
p.add_argument("--hpo_epochs", type=int, default=20)
p.add_argument("--hpo_patience", type=int, default=6)
p.add_argument("--alpha", type=float, default=0.6)
p.add_argument("--startup_trials", type=int, default=5)
p.add_argument("--warmup_steps", type=int, default=5)
p.add_argument("--run_hpo", action="store_true")
p.add_argument("--smoke", action="store_true",
help="Run a minimal 1-trial / 1-epoch HPO smoke test")
args = p.parse_args()
GRAPH_ART.mkdir(parents=True, exist_ok=True)
paths = build_paths()
payload = methodology_payload(paths, args)
with open(paths["methodology_json"], "w") as f:
json.dump(payload, f, indent=2)
with open(paths["methodology_md"], "w") as f:
f.write(methodology_markdown(payload))
cmds = build_commands(paths, args)
if args.smoke:
smoke_cmd = cmds["hpo"].copy()
for flag, value in [("--n_trials", "1"), ("--epochs", "1"), ("--patience", "1")]:
idx = smoke_cmd.index(flag)
smoke_cmd[idx + 1] = value
smoke_idx = smoke_cmd.index("--out")
smoke_cmd[smoke_idx + 1] = str(GRAPH_ART / "hpo_smoke.json")
run(smoke_cmd)
elif args.run_hpo:
run(cmds["hpo"])
write_summary(paths, payload)
print(f"Methodology JSON: {paths['methodology_json']}")
print(f"Methodology MD: {paths['methodology_md']}")
print(f"Summary JSON: {paths['summary_json']}")
if __name__ == "__main__":
main()
|