File size: 2,440 Bytes
61bff85 | 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 | # src/config.py
from pathlib import Path
from dataclasses import dataclass
@dataclass
class Config:
SEED: int = 42
ROOT_DIR: Path = Path(__file__).resolve().parent.parent
def __post_init__(self):
self.DATA_DIR = self.ROOT_DIR / "data" / "processed"
self.RAW_DIR = self.ROOT_DIR / "data" / "raw"
self.OUTPUT_DIR = self.ROOT_DIR / "output"
self.CASF_DIR = self.ROOT_DIR / "data" / "external" / "CASF-2016"
self.CASF13_DIR = self.ROOT_DIR / "data" / "external" / "CASF-2013"
self.DATA_DIR.mkdir(parents=True, exist_ok=True)
self.OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
# LP-PDBBind CSV
LPPDB_CSV: str = "LP_PDBBind.csv" # filename inside RAW_DIR
# ESM — 35M everywhere (train = deploy = benchmark, honest)
ESM_MODEL: str = "facebook/esm2_t12_35M_UR50D"
ESM_LAYERS: tuple = (8, 10, 11) # 0-indexed, proportional to 20/26/30 in 150M
ESM_DIM: int = 480 # hidden dim of 35M
# Long-sequence chunking
MAX_SEQ_LEN: int = 1022
HALF_SEQ_LEN: int = 511
# Ligand
ECFP_BITS: int = 1024
ECFP_RADIUS: int = 2
# Interaction projection
INTERACT_DIM: int = 128
# GBM training
# N_TREES=3000 with early_stop=150 is the sweet spot:
# - LR=0.02 + 3000 trees → actual stopping usually ~800-1200 trees
# - 5 seeds × 4 models × 5 folds = 100 runs × ~6 min = ~10 hrs total
# - Reduce SEEDS to (42, 123, 456) for ~6 hrs if needed
N_FOLDS: int = 5
SEEDS: tuple = (42, 123, 456) # 3 seeds ~6 hrs; add 789,1337 if time allows
LR: float = 0.02
N_TREES: int = 3000
EARLY_STOP: int = 150
# TTA
TTA_SCREEN: int = 5
TTA_ACCURATE: int = 20
# Applicability domain
AD_KNN_K: int = 5
AD_PERCENTILE: float = 95.0
AD_MAX_MONO: float = 0.40 # max single-AA fraction before flagging
config = Config()
if __name__ == "__main__":
print(f"ROOT: {config.ROOT_DIR}")
print(f"DATA: {config.DATA_DIR} exists={config.DATA_DIR.exists()}")
print(f"OUTPUT: {config.OUTPUT_DIR} exists={config.OUTPUT_DIR.exists()}")
print(f"CASF: {config.CASF_DIR} exists={config.CASF_DIR.exists()}")
print(f"CASF13: {config.CASF13_DIR} exists={config.CASF13_DIR.exists()}")
lp = config.RAW_DIR / config.LPPDB_CSV
print(f"LP-PDBBind CSV: {lp} exists={lp.exists()}")
|