File size: 8,241 Bytes
7cbb55b | 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 | """
Training script for nonlinear Poisson equation using GINO.
This script trains a Graph Neural Operator (GINO) on the nonlinear Poisson
equation with boundary conditions. The model learns to solve elliptic PDEs
with complex geometries and boundary conditions using graph-based representations.
"""
import sys
import torch
import wandb
from torch.nn.parallel import DistributedDataParallel as DDP
import torch.nn.functional as F
# Set warning filter to show each warning only once
import warnings
warnings.filterwarnings("once", category=UserWarning)
from neuralop.losses.data_losses import LpLoss, MSELoss
from neuralop.training import Trainer, setup
from neuralop.data.datasets.nonlinear_poisson import load_nonlinear_poisson_pt
from neuralop.losses.equation_losses import PoissonBoundaryLoss, PoissonEqnLoss
from neuralop.losses.meta_losses import WeightedSumLoss
from neuralop.models import get_model
from neuralop.utils import get_wandb_api_key, count_model_params
# Configuration setup
config_name = "default"
from zencfg import make_config_from_cli
import sys
sys.path.insert(0, "../")
from config.poisson_gino_config import Default
config = make_config_from_cli(Default)
config = config.to_dict()
# Distributed training setup, if enabled
device, is_logger = setup(config)
# WandB logging configuration
wandb_args = None
if config.wandb.log and is_logger:
wandb.login(key=get_wandb_api_key())
if config.wandb.name:
wandb_name = config.wandb.name
else:
wandb_name = "_".join(
f"{var}"
for var in [
config_name,
config.model.in_gno_radius,
config.model.out_gno_radius,
config.model.gno_weighting_function,
config.model.gno_weight_function_scale,
config.model.fno_n_modes,
config.model.fno_n_layers,
config.data.n_train,
config.data.n_test,
]
)
wandb_args = dict(
config=config,
name=wandb_name,
group=config.wandb.group,
project=config.wandb.project,
entity=config.wandb.entity,
)
if config.wandb.sweep:
for key in wandb.config.keys():
config.params[key] = wandb.config[key]
wandb.init(**wandb_args)
else:
wandb_init_args = None
# Make sure we only print information when needed
config.verbose = config.verbose and is_logger
# Print configuration details
if config.verbose:
print(f"##### CONFIG #####\n")
print(config)
sys.stdout.flush()
# Load the nonlinear Poisson dataset
train_loader, test_loader, data_processor = load_nonlinear_poisson_pt(
data_path=config.data.file,
query_res=config.data.query_resolution,
n_train=config.data.n_train,
n_test=config.data.n_test,
n_in=config.data.n_in,
n_out=config.data.n_out,
n_eval=config.data.n_eval,
n_bound=config.data.n_bound,
val_on_same_instance=config.data.single_instance,
train_out_res=config.data.train_out_res,
input_min_sample_points=config.data.input_min,
input_max_sample_points=config.data.input_max,
input_subsample_level=config.data.sample_random_in,
output_subsample_level=config.data.sample_random_out,
return_dict=config.data.return_queries_dict,
)
# Create test loaders dictionary
test_loaders = {"test": test_loader}
# Model initialization
model = get_model(config)
model = model.to(device)
# Distributed data parallel setup
if config.distributed.use_distributed:
model = DDP(
model, device_ids=[device.index], output_device=device.index, static_graph=True
)
# Create the optimizer
optimizer = torch.optim.Adam(
model.parameters(),
lr=config.opt.learning_rate,
weight_decay=config.opt.weight_decay,
)
if config.opt.scheduler == "ReduceLROnPlateau":
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
optimizer,
factor=config.opt.gamma,
patience=config.opt.scheduler_patience,
mode="min",
)
elif config.opt.scheduler == "CosineAnnealingLR":
scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
optimizer, T_max=config.opt.scheduler_T_max
)
elif config.opt.scheduler == "StepLR":
scheduler = torch.optim.lr_scheduler.StepLR(
optimizer, step_size=config.opt.step_size, gamma=config.opt.gamma
)
else:
raise ValueError(f"Got scheduler={config.opt.scheduler}")
# Create the losses: MSE and relative L2
mse_loss = MSELoss()
l2_loss = LpLoss(d=2, p=2)
class GINOLoss(object):
"""
Custom loss wrapper for GINO models that handle dictionary outputs.
This loss function concatenates dictionary outputs from GINO models
before applying the base loss function.
"""
def __init__(self, base_loss):
super().__init__()
self.base_loss = base_loss
def __call__(self, out, y, **kwargs):
"""
Apply loss to GINO model outputs.
Parameters
----------
out : dict or torch.Tensor
Model output, either dictionary of field outputs or tensor
y : dict or torch.Tensor
Ground truth, either dictionary of field targets or tensor
**kwargs
Additional arguments passed to base loss
Returns
-------
torch.Tensor
Computed loss value
"""
if kwargs:
warnings.warn(
f"GINOLoss.__call__() received unexpected keyword arguments: {list(kwargs.keys())}. "
"These arguments will be ignored.",
UserWarning,
stacklevel=2
)
if isinstance(out, dict) and isinstance(y, dict):
y = torch.cat([y[field] for field in out.keys()], dim=1)
out = torch.cat([out[field] for field in out.keys()], dim=1)
return self.base_loss(out, y, **kwargs)
gino_mseloss = GINOLoss(mse_loss)
training_loss = config.opt.training_loss
if not isinstance(training_loss, (tuple, list)):
training_loss = [training_loss]
losses = []
weights = []
if "mse" in training_loss:
losses.append(gino_mseloss)
weights.append(config.opt.loss_weights.get("mse", 1.0))
if "equation" in training_loss:
equation_loss = PoissonEqnLoss(
interior_weight=config.opt.loss_weights.get("interior", 1.0),
boundary_weight=config.opt.loss_weights.get("boundary", 1.0),
diff_method=config.opt.get("pino_method", "autograd"),
)
losses.append(equation_loss)
weights.append(1)
if len(losses) == 1:
train_loss = losses[0]
else:
train_loss = WeightedSumLoss(losses=losses, weights=weights)
eval_losses = {"mse": mse_loss, "relative_l2": l2_loss}
if config.verbose:
print("\n### MODEL ###\n", model)
print("\n### OPTIMIZER ###\n", optimizer)
print("\n### SCHEDULER ###\n", scheduler)
print("\n### LOSSES ###")
print(f"\n * Train: {train_loss}")
print(f"\n * Test: {eval_losses}")
print(f"\n### Beginning Training...\n")
sys.stdout.flush()
trainer = Trainer(
model=model,
n_epochs=config.opt.n_epochs,
data_processor=data_processor,
device=device,
mixed_precision=config.opt.mixed_precision,
eval_interval=config.opt.eval_interval,
log_output=config.wandb.log_output,
use_distributed=config.distributed.use_distributed,
verbose=config.verbose,
wandb_log=config.wandb.log,
)
# Log model parameter count
if is_logger:
n_params = count_model_params(model)
if config.verbose:
print(f"\nn_params: {n_params}")
sys.stdout.flush()
if config.wandb.log:
to_log = {"n_params": n_params}
if config.n_params_baseline is not None:
to_log["n_params_baseline"] = (config.n_params_baseline,)
to_log["compression_ratio"] = (config.n_params_baseline / n_params,)
to_log["space_savings"] = 1 - (n_params / config.n_params_baseline)
wandb.log(to_log)
wandb.watch(model)
# Start training process
trainer.train(
train_loader,
test_loaders,
optimizer,
scheduler,
regularizer=None,
training_loss=train_loss,
eval_losses=eval_losses,
)
# Finalize WandB logging
if config.wandb.log and is_logger:
wandb.finish()
|