Spaces:
Sleeping
Sleeping
File size: 12,183 Bytes
8e5ba9e | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 | """Dataset generation pipeline using Latin Hypercube Sampling.
Generates 100K structural analysis samples across beams, plates, and pressure
vessels using analytical closed-form solutions. LHS provides better coverage of
the design space than random sampling — a deliberate choice from Design of
Experiments (DoE) methodology.
Usage:
python -m src.data.generate_dataset --config configs/data_generation.yaml
"""
import argparse
import hashlib
import logging
import uuid
from pathlib import Path
from typing import Any
import numpy as np
import pandas as pd
import yaml
from scipy.stats.qmc import LatinHypercube
from src.data.schema import ProblemFamily, SafetyCategory
from src.data.solvers.beam import BEAM_SOLVERS
from src.data.solvers.plate import PLATE_SOLVERS
from src.data.solvers.vessel import VESSEL_SOLVERS
logger = logging.getLogger(__name__)
def _lhs_sample(
n_samples: int,
param_ranges: dict[str, dict],
seed: int,
) -> dict[str, np.ndarray]:
"""Generate Latin Hypercube Samples and map to physical parameter ranges.
Log-uniform parameters are sampled uniform in log-space then exponentiated,
which is critical because engineering quantities span orders of magnitude
(e.g., elastic modulus: 1 GPa to 400 GPa).
"""
param_names = list(param_ranges.keys())
n_dims = len(param_names)
sampler = LatinHypercube(d=n_dims, seed=seed)
unit_samples = sampler.random(n=n_samples) # shape: (n_samples, n_dims)
result: dict[str, np.ndarray] = {}
for i, name in enumerate(param_names):
spec = param_ranges[name]
lo, hi = float(spec["min"]), float(spec["max"])
col = unit_samples[:, i]
if spec.get("distribution") == "log_uniform":
log_lo, log_hi = np.log10(lo), np.log10(hi)
result[name] = 10.0 ** (log_lo + col * (log_hi - log_lo))
else: # uniform
result[name] = lo + col * (hi - lo)
return result
def _generate_beam_samples(
config: dict,
global_seed: int,
) -> list[dict[str, Any]]:
"""Generate beam samples across all 6 configurations."""
samples = []
beam_configs = config["beam"]["configs"]
n_per_config = config["beam"]["samples_per_config"]
param_ranges = config["beam"]["parameters"]
for cfg_idx, cfg in enumerate(beam_configs):
config_id = cfg["id"]
solver_cls = BEAM_SOLVERS[config_id]
solver = solver_cls()
seed = global_seed + cfg_idx
lhs_params = _lhs_sample(n_per_config, param_ranges, seed)
for i in range(n_per_config):
params = {k: float(v[i]) for k, v in lhs_params.items()}
# Select load type based on config
if "point" in config_id:
load_params = {
"point_load": params["point_load"],
"distributed_load": 0.0,
}
else:
load_params = {
"point_load": 0.0,
"distributed_load": params["distributed_load"],
}
solve_params = {
"length": params["length"],
"width": params["width"],
"height": params["height"],
"elastic_modulus": params["elastic_modulus"],
"yield_strength": params["yield_strength"],
**load_params,
}
result = solver.solve(solve_params)
b, h = params["width"], params["height"]
samples.append({
"sample_id": str(uuid.uuid4()),
"problem_family": ProblemFamily.BEAM.value,
"config_id": config_id,
"length": params["length"],
"width": params["width"],
"height": params["height"],
"inner_radius": None,
"outer_radius": None,
"thickness": None,
"elastic_modulus": params["elastic_modulus"],
"poisson_ratio": params["poisson_ratio"],
"yield_strength": params["yield_strength"],
"density": params["density"],
"point_load": load_params["point_load"],
"distributed_load": load_params["distributed_load"],
"internal_pressure": 0.0,
"pressure": 0.0,
"moment_of_inertia": b * h**3 / 12.0,
"section_modulus": b * h**2 / 6.0,
"cross_section_area": b * h,
"max_stress": result.max_stress,
"max_deflection": result.max_deflection,
"safety_factor": result.safety_factor,
"safety_category": result.safety_category.value,
})
logger.info(f"Generated {n_per_config} samples for {config_id}")
return samples
def _generate_plate_samples(
config: dict,
global_seed: int,
) -> list[dict[str, Any]]:
"""Generate plate samples across both configurations."""
samples = []
plate_configs = config["plate"]["configs"]
n_per_config = config["plate"]["samples_per_config"]
param_ranges = config["plate"]["parameters"]
for cfg_idx, cfg in enumerate(plate_configs):
config_id = cfg["id"]
solver_cls = PLATE_SOLVERS[config_id]
solver = solver_cls()
seed = global_seed + 100 + cfg_idx
lhs_params = _lhs_sample(n_per_config, param_ranges, seed)
for i in range(n_per_config):
params = {k: float(v[i]) for k, v in lhs_params.items()}
solve_params = {
"length_a": params["length_a"],
"length_b": params["length_b"],
"thickness": params["thickness"],
"elastic_modulus": params["elastic_modulus"],
"poisson_ratio": params["poisson_ratio"],
"yield_strength": params["yield_strength"],
"pressure": params["pressure"],
}
result = solver.solve(solve_params)
samples.append({
"sample_id": str(uuid.uuid4()),
"problem_family": ProblemFamily.PLATE.value,
"config_id": config_id,
"length": params["length_a"],
"width": params["length_b"],
"height": None,
"inner_radius": None,
"outer_radius": None,
"thickness": params["thickness"],
"elastic_modulus": params["elastic_modulus"],
"poisson_ratio": params["poisson_ratio"],
"yield_strength": params["yield_strength"],
"density": params["density"],
"point_load": 0.0,
"distributed_load": 0.0,
"internal_pressure": 0.0,
"pressure": params["pressure"],
"moment_of_inertia": None,
"section_modulus": None,
"cross_section_area": None,
"max_stress": result.max_stress,
"max_deflection": result.max_deflection,
"safety_factor": result.safety_factor,
"safety_category": result.safety_category.value,
})
logger.info(f"Generated {n_per_config} samples for {config_id}")
return samples
def _generate_vessel_samples(
config: dict,
global_seed: int,
) -> list[dict[str, Any]]:
"""Generate pressure vessel samples across both configurations."""
samples = []
vessel_configs = config["vessel"]["configs"]
n_per_config = config["vessel"]["samples_per_config"]
param_ranges = config["vessel"]["parameters"]
for cfg_idx, cfg in enumerate(vessel_configs):
config_id = cfg["id"]
solver_cls = VESSEL_SOLVERS[config_id]
solver = solver_cls()
seed = global_seed + 200 + cfg_idx
lhs_params = _lhs_sample(n_per_config, param_ranges, seed)
for i in range(n_per_config):
params = {k: float(v[i]) for k, v in lhs_params.items()}
r_i = params["inner_radius"]
r_o = r_i * params["radius_ratio"]
solve_params = {
"inner_radius": r_i,
"outer_radius": r_o,
"elastic_modulus": params["elastic_modulus"],
"poisson_ratio": params["poisson_ratio"],
"yield_strength": params["yield_strength"],
"internal_pressure": params["internal_pressure"],
}
result = solver.solve(solve_params)
samples.append({
"sample_id": str(uuid.uuid4()),
"problem_family": ProblemFamily.VESSEL.value,
"config_id": config_id,
"length": params.get("length", None) if config_id == "vessel_cylinder" else None,
"width": None,
"height": None,
"inner_radius": r_i,
"outer_radius": r_o,
"thickness": r_o - r_i,
"elastic_modulus": params["elastic_modulus"],
"poisson_ratio": params["poisson_ratio"],
"yield_strength": params["yield_strength"],
"density": params["density"],
"point_load": 0.0,
"distributed_load": 0.0,
"internal_pressure": params["internal_pressure"],
"pressure": 0.0,
"moment_of_inertia": None,
"section_modulus": None,
"cross_section_area": None,
"max_stress": result.max_stress,
"max_deflection": result.max_deflection,
"safety_factor": result.safety_factor,
"safety_category": result.safety_category.value,
})
logger.info(f"Generated {n_per_config} samples for {config_id}")
return samples
def generate_dataset(config_path: str) -> pd.DataFrame:
"""Generate the full dataset from config file."""
with open(config_path) as f:
config = yaml.safe_load(f)
seed = config["seed"]
np.random.seed(seed)
logger.info("Generating beam samples...")
beam_samples = _generate_beam_samples(config, seed)
logger.info("Generating plate samples...")
plate_samples = _generate_plate_samples(config, seed)
logger.info("Generating vessel samples...")
vessel_samples = _generate_vessel_samples(config, seed)
all_samples = beam_samples + plate_samples + vessel_samples
df = pd.DataFrame(all_samples)
logger.info(f"Total samples generated: {len(df)}")
logger.info(f"Safety category distribution:\n{df['safety_category'].value_counts()}")
return df
def split_and_save(df: pd.DataFrame, config_path: str) -> None:
"""Stratified split and save as Parquet files."""
with open(config_path) as f:
config = yaml.safe_load(f)
output_dir = Path(config["output"]["directory"])
output_dir.mkdir(parents=True, exist_ok=True)
splits = config["splits"]
train_frac = splits["train"]
val_frac = splits["validation"]
# Stratified split by config_id and safety_category
df_shuffled = df.sample(frac=1.0, random_state=config["seed"]).reset_index(drop=True)
n = len(df_shuffled)
n_train = int(n * train_frac)
n_val = int(n * val_frac)
train_df = df_shuffled.iloc[:n_train]
val_df = df_shuffled.iloc[n_train:n_train + n_val]
test_df = df_shuffled.iloc[n_train + n_val:]
train_df.to_parquet(output_dir / "train.parquet", index=False)
val_df.to_parquet(output_dir / "validation.parquet", index=False)
test_df.to_parquet(output_dir / "test.parquet", index=False)
logger.info(f"Saved splits: train={len(train_df)}, val={len(val_df)}, test={len(test_df)}")
logger.info(f"Output directory: {output_dir}")
def main() -> None:
parser = argparse.ArgumentParser(description="Generate structural mechanics dataset")
parser.add_argument("--config", default="configs/data_generation.yaml")
args = parser.parse_args()
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
df = generate_dataset(args.config)
split_and_save(df, args.config)
if __name__ == "__main__":
main()
|