normalize.py

import csv
import sys
from collections import defaultdict
from statistics import geometric_mean

STATS = {
    'geomean': geometric_mean,
    'min': min,
    'max': max,
}


def normalize():
    # Read input CSV.
    reader = csv.DictReader(sys.stdin)
    in_data = list(reader)

    # Get normalization baselines.
    baselines = {
        row['benchmark']: int(row['result'])
        for row in in_data
        if row['run'] == 'baseline'
    }

    # Write output CSV back out.
    writer = csv.DictWriter(sys.stdout, reader.fieldnames)
    writer.writeheader()
    ratios = defaultdict(list)
    for row in in_data:
        ratio = int(row['result']) / baselines[row['benchmark']]
        ratios[row['run']].append(ratio)
        row['result'] = ratio
        writer.writerow(row)

    # Print stats.
    for run, rs in ratios.items():
        for name, func in STATS.items():
            print(
                '{}({}) = {:.2f}'.format(name, run, func(rs)),
                file=sys.stderr,
            )


if __name__ == '__main__':
    normalize()