File size: 6,256 Bytes
a00d152 | 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 | """Fit the paper's six station-wise model configurations."""
import json
import os
import sys
from pathlib import Path
import numpy as np
import torch
import yaml
from torch.nn.parallel import DistributedDataParallel
from torch.utils.data import DataLoader, Dataset, DistributedSampler
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT))
from model.globalsurgeml import GlobalSurgeML
CONFIGURATIONS = {
"LR-RS": ("rs_daily", "linear"),
"LR-RS-lag": ("rs_lagged", "linear"),
"RF-RS-lag": ("rs_lagged", "random_forest"),
"LR-AR": ("ar_daily", "linear"),
"LR-AR-lag": ("ar_lagged", "linear"),
"RF-AR-lag": ("ar_lagged", "random_forest"),
}
class SurgeDataset(Dataset):
def __init__(self, path, config):
self.data = np.load(path)
expected = config["data"]
if str(self.data["format_version"]) != expected["format_version"]:
raise ValueError("incompatible storm-surge data format")
dimensions = {"rs_daily": expected["rs_daily_features"], "rs_lagged": expected["rs_lagged_features"],
"ar_daily": expected["ar_daily_features"], "ar_lagged": expected["ar_lagged_features"]}
for key, width in dimensions.items():
if self.data[key].shape[1:] != (int(width),):
raise ValueError(f"{key} must have shape [N,{width}]")
if self.data["targets_m"].shape[1:] != (1,):
raise ValueError("targets_m must have shape [N,1]")
def __len__(self):
return len(self.data["targets_m"])
def __getitem__(self, index):
return {key: torch.from_numpy(self.data[key][index]).float() for key in
("rs_daily", "rs_lagged", "ar_daily", "ar_lagged", "targets_m")}
def device_from_config(config, rank=0):
requested = config["runtime"]["device"]
if requested == "auto":
return torch.device("cuda", rank) if torch.cuda.is_available() else torch.device("cpu")
return torch.device(requested)
def main():
config = yaml.safe_load((ROOT / "conf/config.yaml").read_text())
seed = int(config["seed"])
np.random.seed(seed)
torch.manual_seed(seed)
distributed = int(os.environ.get("WORLD_SIZE", "1")) > 1
local_rank = int(os.environ.get("LOCAL_RANK", "0"))
if distributed:
torch.distributed.init_process_group("nccl" if torch.cuda.is_available() else "gloo")
rank = torch.distributed.get_rank() if distributed else 0
device = device_from_config(config, local_rank)
if device.type == "cuda":
torch.cuda.set_device(device)
dataset = SurgeDataset(ROOT / config["data"]["root"] / "train.npz", config)
sampler = DistributedSampler(dataset, shuffle=True) if distributed else None
loader = DataLoader(dataset, batch_size=int(config["train"]["batch_size"]), sampler=sampler,
shuffle=sampler is None, num_workers=int(config["train"]["num_workers"]))
full = dataset.data
means, scales, states, history = {}, {}, {}, []
models = {}
for model_index, (name, (feature_key, method)) in enumerate(CONFIGURATIONS.items()):
array = full[feature_key].astype(np.float32)
means[name] = array.mean(0).astype(np.float32)
scales[name] = array.std(0).clip(1e-6).astype(np.float32)
standardized = (array - means[name]) / scales[name]
model = GlobalSurgeML(array.shape[1], method, config["model"], seed + model_index).to(device)
if method == "linear":
model.regressor.select_features(standardized, full["targets_m"][:, 0],
float(config["model"]["p_value_threshold"]))
wrapped = DistributedDataParallel(model, device_ids=[local_rank] if device.type == "cuda" else None) if distributed else model
optimizer = torch.optim.Adam(wrapped.parameters(), lr=float(config["train"]["learning_rate"]),
weight_decay=float(config["train"]["weight_decay"]))
for epoch in range(int(config["train"]["epochs"])):
if sampler:
sampler.set_epoch(epoch)
total, steps = 0.0, 0
for batch in loader:
features = (batch[feature_key].to(device) - torch.from_numpy(means[name]).to(device)) / torch.from_numpy(scales[name]).to(device)
target = batch["targets_m"].to(device)
prediction = wrapped(features)
loss = torch.nn.functional.mse_loss(prediction, target)
optimizer.zero_grad(set_to_none=True)
loss.backward()
torch.nn.utils.clip_grad_norm_(wrapped.parameters(), float(config["train"]["gradient_clip_norm"]))
optimizer.step()
total += float(loss.detach())
steps += 1
history.append({"model": name, "epoch": epoch + 1, "mse_m2": total / max(steps, 1)})
model = wrapped.module if distributed else wrapped
else:
model.regressor.fit(standardized, full["targets_m"][:, 0])
prediction = model(torch.from_numpy(standardized).to(device))
history.append({"model": name, "epoch": 1,
"mse_m2": float(torch.nn.functional.mse_loss(prediction.cpu(), torch.from_numpy(full["targets_m"])).item())})
models[name] = model
states[name] = model.state_dict()
if rank == 0:
checkpoint = ROOT / config["paths"]["checkpoint"]
metrics = ROOT / config["paths"]["training_metrics"]
checkpoint.parent.mkdir(parents=True, exist_ok=True)
metrics.parent.mkdir(parents=True, exist_ok=True)
torch.save({"model": states, "model_config": config["model"], "configurations": CONFIGURATIONS,
"feature_means": means, "feature_scales": scales,
"format_version": config["data"]["format_version"], "target_unit": "m"}, checkpoint)
metrics.write_text(json.dumps({"history": history}, indent=2) + "\n")
print(f"checkpoint={checkpoint.relative_to(ROOT)} models={len(states)}")
if distributed:
torch.distributed.destroy_process_group()
if __name__ == "__main__":
main()
|