What food is in season in Singapore?

YEARS = [str(y) for y in range(2016, 2026)]
FOOD_CHAPTERS = {2, 3, 4, 7, 8, 9, 10, 11, 16, 17, 18, 19, 20, 21, 22}
MIN_AVG_USD = 5_000_000
MIN_SMALLEST_OVER_BIGGEST = 0.2
MIN_YEARS_MATCH = 9
MIN_WINDOW_SHARE = 0.30
COMPARE = 3          # flattest foods shown for comparison (not picked)
JOINS = {"MANDARIN": ["080520", "080521", "080522", "080529"]}   # 0805.20 became 0805.21, .22 and .29 from 2018
MONTHS = "Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec".split()

def series(codes, table, years=YEARS):
    """{(year, month): total} for a food made of one or more codes."""
    out = defaultdict(float)
    for c in codes:
        for (cc, y, m), v in table.items():
            if cc == c and y in years:
                out[(y, m)] += v
    return out

def analyse(codes, value, weight, years=YEARS):
    s = series(codes, value, years)
    annual = {y: sum(s.get((y, m), 0.0) for m in range(1, 13)) for y in years}
    if min(annual.values()) <= 0:
        return None, "a year with no imports"
    avg = sum(annual.values()) / len(years)
    if avg < MIN_AVG_USD:
        return None, "under US$5m a year"
    if min(annual.values()) / max(annual.values()) < MIN_SMALLEST_OVER_BIGGEST:
        return None, "yearly total jumps (code change?)"
    share = [sum(s.get((y, m), 0.0) / annual[y] for y in years) / len(years) for m in range(1, 13)]
    peak = max(range(12), key=lambda i: share[i]) + 1
    win = [((peak - 2 + k) % 12) + 1 for k in range(3)]          # peak month and the months either side
    win_share = sum(share[m - 1] for m in win)
    busiest = {}
    for y in years:
        busiest[y] = max(range(1, 13), key=lambda m: s.get((y, m), 0.0))
    hits = sum(1 for y in years if min((busiest[y] - peak) % 12, (peak - busiest[y]) % 12) <= 1)
    # the same measure by weight, as a cross-check
    sw = series(codes, weight, years)
    wann = {y: sum(sw.get((y, m), 0.0) for m in range(1, 13)) for y in years}
    wpeak = None
    if all(v > 0 for v in wann.values()):
        wshare = [sum(sw.get((y, m), 0.0) / wann[y] for y in years) / len(years) for m in range(1, 13)]
        wpeak = max(range(12), key=lambda i: wshare[i]) + 1
    monthly = {y: [round(s.get((y, m), 0.0)) for m in range(1, 13)] for y in years}
    month_avg = [sum(s.get((y, m), 0.0) for y in years) / len(years) for m in range(1, 13)]
    yearly_win = {y: sum(s.get((y, m), 0.0) for m in win) / annual[y] for y in years}
    peak_month_record = max(((s.get((y, m), 0.0), y, m) for y in years for m in range(1, 13)))
    low_month_record = min(((s.get((y, m), 0.0), y, m) for y in years for m in range(1, 13)))
    return dict(
        avg_usd=avg, annual=annual, share=share, peak=peak, win=win, win_share=win_share, hits=hits,
        busiest=busiest, weight_peak=wpeak, month_avg=month_avg, monthly=monthly, yearly_win=yearly_win,
        record_high=list(peak_month_record), record_low=list(low_month_record),
        top_month_share=max(share), flat_ratio=max(share) * 12,
    ), "ok"

def circ(a, b):
    return min((a - b) % 12, (b - a) % 12)

# inside main(): which foods are picked
def pick(candidates, value, weight):
    steps, dropped, passed, analysed = {}, {}, {}, {}
    for key, cs in candidates:
        if int(cs[0][:2]) not in FOOD_CHAPTERS:
            continue
        steps["food chapters"] += 1
        res, why = analyse(cs, value, weight)
        if res is None:
            continue
        steps["enough imports every year"] += 1
        res["codes"] = cs
        res["description"] = desc.get(cs[0], "")
        analysed[key] = res
        if res["hits"] < MIN_YEARS_MATCH:
            continue
        steps["same busiest month in 9 of 10 years"] += 1
        if not res["weight_peak"] or circ(res["weight_peak"], res["peak"]) > 1:
            dropped["weight"].append(key)
            continue
        steps["weight agrees"] += 1
        if "n.e.c." in res["description"]:
            dropped["catchall"].append(key)
            continue
        steps["names a food"] += 1
        if res["win_share"] < MIN_WINDOW_SHARE:
            dropped["even"].append(key)
            continue
        steps["30% in busiest 3 months"] += 1
        ex, _ = analyse(cs, value, weight, [y for y in YEARS if y not in ("2020", "2021")])   # the same food without 2020 and 2021
        res["ex_covid_peak"] = ex["peak"] if ex else None
        res["ex_covid_win_share"] = ex["win_share"] if ex else None
        passed[key] = res