"""Compute every saved-analysis result for the dataset from FARS.
Run from this folder: python -I build_dataset.py
.
Writes data/results-for-dataset.json."""
import runpy, io, contextlib, json, datetime as dt, statistics as st
with contextlib.redirect_stdout(io.StringIO()):
g = runpy.run_path('analyze.py', run_name='x')
ai, al, nights, rank, rows, tg, base = g['ai_by_hour'], g['all_by_hour'], g['nights'], g['rank'], g['rows'], g['thanksgiving'], g['baselines']
nth, last = g['nth_weekday'], g['last_weekday']
Y = range(2015, 2025)
byd = {d: a for d, a, t in nights}
alld = {d: t for d, a, t in nights}
named = {r['night']: r for r in rows}
def win(d, h0, h1, t):
s = 0
for k in range(h0, h1):
x = dt.datetime.combine(d, dt.time(0)) + dt.timedelta(hours=k)
s += t.get((x.date(), x.hour), 0)
return s
days = [dt.date(2015, 1, 1) + dt.timedelta(days=i) for i in range((dt.date(2024, 12, 30) - dt.date(2015, 1, 1)).days + 1)]
# 1. all named nights + typical-night baselines
allrows = []
for r in sorted(rows, key=lambda r: -r['avg_ai']):
ranks = [p['rank'] for p in r['per_year']]
allrows.append({'night': r['night'], 'avgAlcoholImpairedDeaths': r['avg_ai'], 'avgAllTrafficDeaths': r['avg_all'],
'yearsCounted': r['years'], 'medianRankOf365': r['median_rank'], 'bestRank': min(ranks), 'worstRank': max(ranks)})
blabel = {'Average night (all nights)': 'Typical night (all nights)', 'Average Wednesday night': 'Typical Wednesday night',
'Average Friday night': 'Typical Friday night', 'Average Saturday night': 'Typical Saturday night',
'Average Saturday night, June-Aug.': 'Typical summer Saturday night (June-Aug.)',
'Average Saturday night, Oct.-Nov.': 'Typical fall Saturday night (Oct.-Nov.)',
'Average Wednesday night, Oct.-Dec. (not Thanksgiving Eve)': 'Typical Wednesday night, Oct.-Dec. (not Thanksgiving Eve)'}
baserows = [{'night': blabel[k], 'avgAlcoholImpairedDeaths': v[0], 'avgAllTrafficDeaths': v[1], 'nightsAveraged': v[2]} for k, v in base.items()]
# 2. lead chart rows
short = [("New Year's Eve", "New Year's Eve"), ('Saturday of Memorial Day weekend', 'Memorial Day weekend Saturday'),
('Saturday of Labor Day weekend', 'Labor Day weekend Saturday'), ('Fourth of July', 'Fourth of July'),
('Cinco de Mayo', 'Cinco de Mayo'), ('Thanksgiving Eve (Wed.)', 'Thanksgiving Eve'), ('Halloween', 'Halloween'),
('Christmas Eve', 'Christmas Eve'), ("St. Patrick's Day", "St. Patrick's Day"), ('Super Bowl Sunday', 'Super Bowl Sunday')]
chartrows = [{'night': s, 'deaths': named[n]['avg_ai']} for n, s in short]
# 3. Thanksgiving Eve by year
tgrows = []
for y in Y:
e = tg(y) - dt.timedelta(days=1)
a = byd[e]
weds = sorted([v for d, v in byd.items() if d.year == y and d.weekday() == 2], reverse=True)
sats = [v for d, v in byd.items() if d.year == y and d.weekday() == 5]
nye = byd.get(dt.date(y, 12, 31))
tgrows.append({'year': y, 'date': e.isoformat(), 'alcoholImpairedDeaths': round(a, 1), 'allTrafficDeaths': int(alld[e]),
'rankOfAllNights': rank[e][0], 'nightsInYear': rank[e][1], 'rankAmongWednesdays': 1 + sum(round(v, 6) > round(a, 6) for v in weds),
'saturdayNightsWithMoreDeaths': sum(round(s, 6) > round(a, 6) for s in sats), 'saturdayNightsInYear': len(sats),
'newYearsEveDeaths': None if nye is None else round(nye, 1)})
# 4. head to head: in how many years was each named night deadlier than Thanksgiving Eve
e = {p['year']: p['ai'] for p in named['Thanksgiving Eve (Wed.)']['per_year']}
h2h = []
for r in rows:
if r['night'] == 'Thanksgiving Eve (Wed.)':
continue
yrs = [p for p in r['per_year'] if p['year'] in e]
h2h.append({'night': r['night'], 'yearsDeadlierThanThanksgivingEve': sum(p['ai'] > e[p['year']] for p in yrs), 'yearsCompared': len(yrs)})
h2h.sort(key=lambda x: -x['yearsDeadlierThanThanksgivingEve'] / x['yearsCompared'])
# 5. holiday periods per 24 hours
def period(start, hours, t):
return sum(t.get(((start + dt.timedelta(hours=k)).date(), (start + dt.timedelta(hours=k)).hour), 0) for k in range(hours))
at6 = lambda d: dt.datetime.combine(d, dt.time(18))
P = {'Thanksgiving (Wed. 6 p.m. to Mon. 6 a.m.)': lambda y: at6(tg(y) - dt.timedelta(days=1)),
'Memorial Day (Fri. 6 p.m. to Tue. 6 a.m.)': lambda y: at6(last(y, 5, 0) - dt.timedelta(days=3)),
'Labor Day (Fri. 6 p.m. to Tue. 6 a.m.)': lambda y: at6(nth(y, 9, 0, 1) - dt.timedelta(days=3))}
prow = []
for k, f in P.items():
a = [period(f(y), 108, ai) for y in Y]
t = [period(f(y), 108, al) for y in Y]
prow.append({'period': k, 'avgAlcoholImpairedDeathsPerPeriod': round(st.mean(a), 1), 'alcoholImpairedDeathsPer24Hours': round(st.mean(a) / 4.5, 1),
'allTrafficDeathsPer24Hours': round(st.mean(t) / 4.5, 1)})
avg24 = sum(win(d, 0, 24, ai) for d in days) / len(days)
avg24all = sum(win(d, 0, 24, al) for d in days) / len(days)
prow.append({'period': 'Typical 24 hours (all days)', 'avgAlcoholImpairedDeathsPerPeriod': None,
'alcoholImpairedDeathsPer24Hours': round(avg24, 1), 'allTrafficDeathsPer24Hours': round(avg24all, 1)})
tgperiod = []
for y in Y:
s = at6(tg(y) - dt.timedelta(days=1))
tgperiod.append({'year': y, 'alcoholImpairedDeaths': round(period(s, 108, ai), 1), 'allTrafficDeaths': int(period(s, 108, al))})
# 6. time-window sensitivity
wrows = []
for label, h0, h1 in [('Evening to morning, 6 p.m. to 5:59 a.m. (main)', 18, 30), ('Late night, 9 p.m. to 2:59 a.m.', 21, 27),
('Calendar day, midnight to 11:59 p.m.', 0, 24)]:
vals = {d: win(d, h0, h1, ai) for d in days}
ev = [vals[tg(y) - dt.timedelta(days=1)] for y in Y]
rk = []
for y in Y:
ys = sorted([v for d, v in vals.items() if d.year == y], reverse=True)
rk.append(1 + sum(round(v, 6) > round(vals[tg(y) - dt.timedelta(days=1)], 6) for v in ys))
wrows.append({'window': label, 'thanksgivingEveAvg': round(st.mean(ev), 1),
'typicalSaturdayAvg': round(st.mean(v for d, v in vals.items() if d.weekday() == 5), 1),
'typicalWednesdayAvg': round(st.mean(v for d, v in vals.items() if d.weekday() == 2), 1),
'newYearsEveAvg': round(st.mean(vals[dt.date(y, 12, 31)] for y in range(2015, 2024)), 1),
'thanksgivingEveBestRankAnyYear': min(rk), 'thanksgivingEveMedianRank': st.median(rk)})
# 7. annual totals
trows = [{'year': y, 'alcoholImpairedDeaths': round(g['total_ai_by_year'][y]), 'allTrafficDeaths': int(g['total_all_by_year'][y])} for y in Y]
json.dump({'allrows': allrows, 'baserows': baserows, 'chartrows': chartrows, 'tgrows': tgrows, 'h2h': h2h, 'prow': prow,
'tgperiod': tgperiod, 'wrows': wrows, 'trows': trows, 'unknownHour': round(g['unknown_hour_ai'], 1)},
open('data/results-for-dataset.json', 'w'), indent=1)
print(json.dumps({'prow': prow, 'wrows': wrows, 'tgperiod': tgperiod}, indent=0))