| """Small CPU baseline with JSON weights. Replace with your own model implementation.""" |
| import json |
| from pathlib import Path |
| import re |
|
|
| import numpy as np |
|
|
| AMINO_ACIDS = 'ACDEFGHIKLMNPQRSTVWY' |
| SMILES_CHARS = 'BCNOFPSIcnops[]=#()+-@/\\1234567890' |
| FEATURE_NAMES = ['log_sequence_length', 'log_smiles_length'] + list(AMINO_ACIDS) + list(SMILES_CHARS) |
|
|
|
|
| def features(rows): |
| result = [] |
| for row in rows: |
| seq, smi = row['sequence'], row['smiles'] |
| seq_upper = seq.upper() |
| result.append([np.log1p(len(seq)), np.log1p(len(smi))] |
| + [seq_upper.count(a) / max(1, len(seq)) for a in AMINO_ACIDS] |
| + [smi.count(a) / max(1, len(smi)) for a in SMILES_CHARS]) |
| return np.asarray(result, dtype=float) |
|
|
|
|
| class Model: |
| def __init__(self, weights=None): |
| path = Path(weights) if weights else Path(__file__).parent / 'weights' / 'baseline.json' |
| self.weights = json.loads(path.read_text()) |
|
|
| def predict(self, rows): |
| x = features(rows) |
| w = self.weights |
| return (((x - np.asarray(w['mean'])) / np.asarray(w['scale'])) @ np.asarray(w['coef']) |
| + w['intercept']).tolist() |
|
|
|
|
| def validate_rows(rows, max_rows=256): |
| if not isinstance(rows, list) or not 1 <= len(rows) <= max_rows: |
| raise ValueError(f'Provide between 1 and {max_rows} rows.') |
| ids = set() |
| for row in rows: |
| if not isinstance(row, dict) or set(row) != {'id', 'smiles', 'sequence'}: |
| raise ValueError('Each row must contain exactly id, smiles, sequence.') |
| if not isinstance(row['id'], str) or not re.fullmatch(r'[A-Za-z0-9_-]{1,80}', row['id']): |
| raise ValueError('Invalid row ID.') |
| if row['id'] in ids: |
| raise ValueError('Duplicate row ID.') |
| ids.add(row['id']) |
| for key, limit in [('sequence', 50000), ('smiles', 20000)]: |
| if not isinstance(row[key], str) or not 1 <= len(row[key]) <= limit: |
| raise ValueError(f'Invalid {key}.') |
| return rows |
|
|
