Spaces:
Sleeping
Sleeping
| from dataclasses import dataclass, asdict, field | |
| from typing import Optional, Dict | |
| class Molecule: | |
| smiles: str | |
| cn: float | |
| cn_error: float | |
| cn_score: float = 0.0 # keep for backwards compat, will be set by fitness() | |
| bp: Optional[float] = None | |
| ysi: Optional[float] = None | |
| density: Optional[float] = None | |
| lhv: Optional[float] = None | |
| dynamic_viscosity: Optional[float] = None | |
| chemical_valid: bool = True | |
| chemical_flags: str = 'OK' | |
| ood_warning: bool = False | |
| confidence_score: float = 100.0 | |
| mixture_dcn: Optional[float] = None | |
| blend_ratio: Optional[float] = None | |
| def fitness(self, config) -> float: | |
| """ | |
| Single scalar fitness for weighted parent selection. | |
| Higher is always better (so we can use as selection weight). | |
| - maximize_cn mode: raw CN value | |
| - target_cn mode: inverted error (so lower error = higher fitness) | |
| - YSI penalty: applied as soft penalty, not hard filter | |
| """ | |
| if config.maximize_cn: | |
| base = self.cn | |
| else: | |
| # Avoid division by zero; small epsilon floor | |
| base = 1.0 / (self.cn_error + 1e-6) | |
| if config.minimize_ysi and self.ysi is not None: | |
| # Soft penalty: YSI above 50 is penalised, below is neutral/rewarded | |
| ysi_penalty = max(0.0, (self.ysi - 50) / 50) # 0 at ysi=50, 1 at ysi=100 | |
| base *= (1.0 - 0.3 * ysi_penalty) # up to 30% penalty | |
| return max(base, 1e-9) # keep strictly positive for sampling weights | |
| def to_dict(self) -> Dict: | |
| return {k: v for k, v in asdict(self).items() if v is not None} |