plot.py 文件源码

def get_normalized_scores(entries, args):
    """Generate smoothed scores based on mean, stdev of that days times. """

    times_by_date = defaultdict(dict)
    for e in entries:
        x = e.seconds
        # if x > 0:
        #     x = np.log(x)
        times_by_date[e.date][e.userid] = x

    sorted_dates = sorted(times_by_date.keys())

    # failures come with a heaver ranking penalty
    MAX_SCORE = 1.5
    FAILURE_PENALTY = -2

    def mk_score(mean, t, stdev):
        if t < 0:
            return FAILURE_PENALTY
        if stdev == 0:
            return 0

        score = (mean - t) / stdev
        return np.clip(score, -MAX_SCORE, MAX_SCORE)

    # scores are the stdev away from mean of that day
    scores = {}
    for date, user_times in times_by_date.items():
        times = [t for t in user_times.values() if t is not None]
        # make failures 1 minute worse than the worst time
        times = [t if t >= 0 else max(times) + 60 for t in times]
        q1, q3 = np.percentile(times, [25,75])
        stdev  = statistics.pstdev(times)
        o1, o3 = q1 - stdev, q3 + stdev
        times  = [t for t in times if o1 <= t <= o3]
        mean  = statistics.mean(times)
        stdev  = statistics.pstdev(times, mean)
        scores[date] = {
            userid: mk_score(mean, t, stdev)
            for userid, t in user_times.items()
            if t is not None
        }

    new_score_weight = 1 - args.smooth
    running = defaultdict(list)

    weighted_scores = defaultdict(dict)
    for date in sorted_dates:
        for user, score in scores[date].items():

            old_score = running.get(user)

            new_score = score * new_score_weight + old_score * (1 - new_score_weight) \
                        if old_score is not None else score

            weighted_scores[user][date] = running[user] = new_score

    ticker = matplotlib.ticker.MultipleLocator(base=0.25)
    formatter = matplotlib.ticker.ScalarFormatter(useOffset=False)

    return weighted_scores, ticker, formatter