| """Build the reference database: one row per unique training sequence with its ECs (most frequent first).""" |
| import argparse |
| import json |
| from pathlib import Path |
|
|
| import pandas as pd |
|
|
|
|
| def build(train_path, output_path): |
| root = Path(__file__).parent |
| vocab = set(json.loads((root / 'ec_vocabulary.json').read_text())) |
| frame = pd.read_parquet(train_path)[['sequence', 'ec_number']].dropna() |
| bad = sorted(set(frame.ec_number) - vocab) |
| if bad: |
| raise ValueError(f'{len(bad)} training ECs not in vocabulary, e.g. {bad[:5]}') |
| counts = frame.ec_number.value_counts() |
| frame = frame.drop_duplicates() |
| frame['n'] = frame.ec_number.map(counts) |
| frame = frame.sort_values(['sequence', 'n', 'ec_number'], ascending=[True, False, True]) |
| db = frame.groupby('sequence', sort=False).ec_number.agg(list).reset_index(name='ecs') |
| Path(output_path).parent.mkdir(parents=True, exist_ok=True) |
| db.to_parquet(output_path, index=False, compression='zstd') |
| print(f'Wrote {len(db)} sequences, {frame.ec_number.nunique()} ECs to {output_path}') |
|
|
|
|
| if __name__ == '__main__': |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument('--train', required=True) |
| parser.add_argument('--output', default=str(Path(__file__).parent / 'weights/db.parquet')) |
| args = parser.parse_args() |
| build(args.train, args.output) |
|
|
