def find_matches()

in baselines/how_find_matches.py [0:0]

• 29 lines of code
• 10 McCabe index (conditional complexity)

def find_matches(args):
    """Given results files, create csv predictions file"""
    databases = [args.database]
    if args.normalization_set:
        databases.append(args.normalization_set)

    # Load list files
    queries = load_list_file(args.query_list)
    references = load_list_file(args.db_list)
    # Load scenario parameters
    params = io_helpers.load_params(args.scenario)
    exp_name = args.scenario.rsplit("/", 1)[1][:-len(args.scenario.rsplit(".", 1)[1])-1]
    eval_folder = HOW_ROOT / params['demo_eval']['exp_folder'] / exp_name / "eval"
    # Load results for both databases
    results = {}
    for database in databases:
        results[database] = load_results(eval_folder.glob(f"{database}.results*.pkl"))
    if args.normalization_set:
        assert (results[args.normalization_set]['query_ids'] == results[args.database]['query_ids']).all()

    # Normalize references scores by train scores
    ranks = results[args.database]['ranks']
    scores = results[args.database]['scores']
    if args.normalization_set:
        norm_reduction, norm_rank, norm_factor = "min", 9, 2
        if norm_reduction == "min":
            norms = results[args.normalization_set]['scores'][:,norm_rank]
        elif norm_reduction == "mean":
            norms = results[args.normalization_set]['scores'][:,:norm_rank+1].mean(axis=1)
        scores -= norm_factor * norms[:,None]
    # Take top predictions
    top_idxs = np.argsort(-scores.flatten())[:MAX_RESULTS]
    idx_query, idx_rank = np.unravel_index(top_idxs, scores.shape)
    idx_db, score = ranks[idx_query,idx_rank], scores[idx_query,idx_rank]
    predictions = [(queries[q], references[db], s) for q, db, s in zip(idx_query, idx_db, score)]
    if args.preds_filepath:
        store_predictions(predictions, args.preds_filepath)