File size: 6,218 Bytes
ef2ae28 | 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 | """Train the ConvLSTM classifier with optional distributed data parallelism."""
import json
import os
import random
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.firecubenet import FireCubeNet
class WildfireDataset(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 wildfire data format")
expected_shape = (int(expected["sequence_days"]), int(expected["channels"]),
int(expected["patch_height"]), int(expected["patch_width"]))
if self.data["inputs"].ndim != 5 or self.data["inputs"].shape[1:] != expected_shape:
raise ValueError(f"inputs must have shape [B,{','.join(map(str, expected_shape))}]")
count = len(self.data["inputs"])
if self.data["labels"].shape != (count, 1):
raise ValueError("labels must have shape [B,1]")
if self.data["coords"].shape != (count, 2) or self.data["timestamps_unix_s"].shape != (count,):
raise ValueError("coords/timestamps shape mismatch")
if not np.isfinite(self.data["inputs"]).all() or not np.isfinite(self.data["labels"]).all():
raise ValueError("inputs and labels must be finite")
if not np.isin(self.data["labels"], (0, 1)).all():
raise ValueError("labels must be binary")
cover_sum = self.data["inputs"][:, :, 15:25].sum(axis=2)
if not np.allclose(cover_sum, 1.0, atol=1e-5):
raise ValueError("land-cover fractions must sum to one")
def __len__(self):
return len(self.data["labels"])
def __getitem__(self, index):
return (torch.from_numpy(self.data["inputs"][index]).float(),
torch.from_numpy(self.data["labels"][index]).float())
def device_from_config(config, local_rank=0):
requested = config["runtime"]["device"]
if requested == "auto":
return torch.device("cuda", local_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"])
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(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 = WildfireDataset(ROOT / config["data"]["root"] / "train.npz", config)
sampler = DistributedSampler(dataset, shuffle=True, seed=seed) if distributed else None
loader = DataLoader(dataset, batch_size=int(config["train"]["batch_size"]),
shuffle=sampler is None, sampler=sampler,
num_workers=int(config["train"]["num_workers"]))
channel_mean = dataset.data["inputs"].mean(axis=(0, 1, 3, 4)).astype(np.float32)
channel_std = dataset.data["inputs"].std(axis=(0, 1, 3, 4)).clip(1e-6).astype(np.float32)
model = FireCubeNet(**config["model"]).to(device)
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"]))
criterion = torch.nn.BCEWithLogitsLoss()
mean = torch.from_numpy(channel_mean).to(device).view(1, 1, -1, 1, 1)
std = torch.from_numpy(channel_std).to(device).view(1, 1, -1, 1, 1)
history = []
for epoch in range(int(config["train"]["epochs"])):
if sampler is not None:
sampler.set_epoch(epoch)
total, samples = 0.0, 0
wrapped.train()
for inputs, labels in loader:
inputs, labels = inputs.to(device), labels.to(device)
logits = wrapped((inputs - mean) / std)
loss = criterion(logits, labels)
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()) * len(inputs)
samples += len(inputs)
loss_sum = torch.tensor([total, samples], dtype=torch.float64, device=device)
if distributed:
torch.distributed.all_reduce(loss_sum)
if rank == 0:
history.append({"epoch": epoch + 1, "bce_with_logits": float(loss_sum[0] / loss_sum[1])})
if rank == 0:
checkpoint_path = ROOT / config["paths"]["checkpoint"]
metrics_path = ROOT / config["paths"]["training_metrics"]
checkpoint_path.parent.mkdir(parents=True, exist_ok=True)
metrics_path.parent.mkdir(parents=True, exist_ok=True)
bare_model = wrapped.module if distributed else wrapped
torch.save({
"model_state_dict": bare_model.state_dict(),
"optimizer_state_dict": optimizer.state_dict(),
"model_config": config["model"], "epoch": int(config["train"]["epochs"]),
"channel_mean": channel_mean, "channel_std": channel_std,
"format_version": config["data"]["format_version"], "seed": seed,
}, checkpoint_path)
metrics_path.write_text(json.dumps({"history": history}, indent=2) + "\n")
print(f"checkpoint={checkpoint_path.relative_to(ROOT)} final_loss={history[-1]['bce_with_logits']:.6f}")
if distributed:
torch.distributed.destroy_process_group()
if __name__ == "__main__":
main()
|