732642856
d2cd5a369f
- Add sanitized Jyotish benchmark suite with fictional/public smoke samples - Include benchmark scripts and markdown reports while excluding raw JSON/CSV outputs - Add open-source Jyotish project comparison research - Add complete technique coverage roadmap - Add privacy-safe PDF chart validation methodology - Update SKILL.md and CHANGELOG with v6.1.9 scope and privacy boundaries Validation: - py_compile benchmarks/jyotish/scripts/*.py passed - quality gate passed with 35 pytest tests and golden case
346 lines
15 KiB
Python
346 lines
15 KiB
Python
# NOTE: This script was sanitized for the public repository in v6.1.9.
|
||
# It assumes it is run from the repository root unless JYOTISH_BENCHMARK_ROOT
|
||
# or JYOTISH_SKILL_SCRIPT is provided. Raw output directories are generated locally
|
||
# and are intentionally not committed.
|
||
#!/usr/bin/env python3
|
||
import csv
|
||
import json
|
||
import os
|
||
import sys
|
||
from datetime import datetime
|
||
from pathlib import Path
|
||
|
||
ROOT = Path(__file__).resolve().parents[1]
|
||
DATA = ROOT / 'data/benchmark_samples.json'
|
||
OUT = ROOT / 'outputs'
|
||
LOCAL_CANON = OUT / 'canonical'
|
||
PYJHORA_OUT = OUT / 'pyjhora'
|
||
|
||
# Keep personal data out of benchmark: samples are fictional/public smoke cases only.
|
||
PLANETS = ['Sun', 'Moon', 'Mars', 'Mercury', 'Jupiter', 'Venus', 'Saturn', 'Rahu', 'Ketu']
|
||
PYJHORA_PLANET_ID = {
|
||
0: 'Sun',
|
||
1: 'Moon',
|
||
2: 'Mars',
|
||
3: 'Mercury',
|
||
4: 'Jupiter',
|
||
5: 'Venus',
|
||
6: 'Saturn',
|
||
7: 'Rahu',
|
||
8: 'Ketu',
|
||
}
|
||
DASHA_LORDS = {8: 'Ketu', 5: 'Venus', 0: 'Sun', 1: 'Moon', 2: 'Mars', 7: 'Rahu', 4: 'Jupiter', 6: 'Saturn', 3: 'Mercury'}
|
||
SIGNS = ['Aries', 'Taurus', 'Gemini', 'Cancer', 'Leo', 'Virgo', 'Libra', 'Scorpio', 'Sagittarius', 'Capricorn', 'Aquarius', 'Pisces']
|
||
NAKSHATRAS = [
|
||
'Ashwini', 'Bharani', 'Krittika', 'Rohini', 'Mrigashira', 'Ardra', 'Punarvasu', 'Pushya', 'Ashlesha',
|
||
'Magha', 'Purva Phalguni', 'Uttara Phalguni', 'Hasta', 'Chitra', 'Swati', 'Vishakha', 'Anuradha', 'Jyeshtha',
|
||
'Mula', 'Purva Ashadha', 'Uttara Ashadha', 'Shravana', 'Dhanishta', 'Shatabhisha', 'Purva Bhadrapada', 'Uttara Bhadrapada', 'Revati'
|
||
]
|
||
|
||
|
||
def patch_swisseph():
|
||
import swisseph as swe
|
||
for name in ['SIDM_KRISHNAMURTI_VP291', 'SIDM_TRUE_MULA', 'SIDM_TRUE_CITRA', 'SIDM_TRUE_REVATI']:
|
||
if not hasattr(swe, name):
|
||
setattr(swe, name, getattr(swe, 'SIDM_KRISHNAMURTI', 1))
|
||
|
||
orig_calc_ut = swe.calc_ut
|
||
def calc_ut(jd, body, flags=0, *args, **kwargs):
|
||
if 'flags' in kwargs:
|
||
flags = kwargs.pop('flags')
|
||
return orig_calc_ut(jd, body, flags)
|
||
swe.calc_ut = calc_ut
|
||
|
||
orig_houses_ex = swe.houses_ex
|
||
def houses_ex(tjdut, lat, lon, hsys=b'P', flags=0, *args, **kwargs):
|
||
if 'flags' in kwargs:
|
||
flags = kwargs.pop('flags')
|
||
if 'hsys' in kwargs:
|
||
hsys = kwargs.pop('hsys')
|
||
return orig_houses_ex(tjdut, lat, lon, hsys, flags)
|
||
swe.houses_ex = houses_ex
|
||
return swe
|
||
|
||
|
||
def sign_name(sign_idx):
|
||
return SIGNS[int(sign_idx) % 12]
|
||
|
||
|
||
def nakshatra_from_abs(abs_lon):
|
||
x = abs_lon % 360.0
|
||
unit = 360.0 / 27.0
|
||
idx = int(x // unit)
|
||
pada = int((x % unit) // (unit / 4.0)) + 1
|
||
return NAKSHATRAS[idx], pada
|
||
|
||
|
||
def parse_chart_positions(rows):
|
||
result = {}
|
||
for key, value in rows:
|
||
sign_idx, deg = value
|
||
if key == 'L':
|
||
body = 'Ascendant'
|
||
else:
|
||
body = PYJHORA_PLANET_ID.get(key)
|
||
if not body:
|
||
continue
|
||
result[body] = {
|
||
'sign': sign_name(sign_idx),
|
||
'sign_idx': int(sign_idx),
|
||
'degree_in_sign': round(float(deg), 4),
|
||
}
|
||
return result
|
||
|
||
|
||
def tuple_to_date(t):
|
||
if not t:
|
||
return None
|
||
y, m, d, _fh = t
|
||
return f'{int(y):04d}-{int(m):02d}-{int(d):02d}'
|
||
|
||
|
||
def build_pyjhora_sample(sample):
|
||
swe = patch_swisseph()
|
||
from jhora import utils, const
|
||
from jhora.panchanga import drik
|
||
from jhora.horoscope.chart import charts
|
||
from jhora.horoscope.dhasa.graha import vimsottari
|
||
|
||
# Align benchmark口径: Lahiri + mean sidereal year. PyJHora default is TRUE_PUSHYA.
|
||
const._DEFAULT_AYANAMSA_MODE = 'LAHIRI'
|
||
drik.set_ayanamsa_mode('LAHIRI')
|
||
try:
|
||
const.dhasa_year_duration_default = const.DHASA_YEAR_DURATION.MEAN_SIDEREAL_YEAR
|
||
except Exception:
|
||
pass
|
||
|
||
b = sample['birth']
|
||
jd = utils.julian_day_number((b['year'], b['month'], b['day']), (b['hour'], b['minute'], 0))
|
||
place = drik.Place(sample['label'], b['lat'], b['lon'], b['tz'])
|
||
today = sample.get('today', '2026-06-03')
|
||
ty, tm, td = [int(x) for x in today.split('-')]
|
||
current_jd = utils.julian_day_number((ty, tm, td), (0, 0, 0))
|
||
|
||
rasi = parse_chart_positions(charts.rasi_chart(jd, place))
|
||
d9 = parse_chart_positions(charts.divisional_chart(jd, place, divisional_chart_factor=9, chart_method=1))
|
||
d10 = parse_chart_positions(charts.divisional_chart(jd, place, divisional_chart_factor=10, chart_method=1))
|
||
|
||
asc = rasi.get('Ascendant') or {}
|
||
planets = {}
|
||
for p in PLANETS:
|
||
pd = rasi.get(p) or {}
|
||
abs_lon = pd.get('sign_idx', 0) * 30.0 + float(pd.get('degree_in_sign', 0.0))
|
||
nak, pada = nakshatra_from_abs(abs_lon)
|
||
planets[p] = {
|
||
'sign': pd.get('sign'),
|
||
'degree_in_sign': pd.get('degree_in_sign'),
|
||
'nakshatra': nak,
|
||
'nakshatra_pada': pada,
|
||
}
|
||
|
||
dasha = {
|
||
'mahadasha_lord': None,
|
||
'mahadasha_start': None,
|
||
'mahadasha_end': None,
|
||
'antardasha_lord': None,
|
||
'antardasha_start': None,
|
||
'antardasha_end': None,
|
||
}
|
||
try:
|
||
ladder = vimsottari.get_running_dhasa_for_given_date(current_jd, jd, place, dhasa_level_index=2)
|
||
if ladder:
|
||
md = ladder[0]
|
||
dasha['mahadasha_lord'] = DASHA_LORDS.get(md[0][0], str(md[0][0]))
|
||
dasha['mahadasha_start'] = tuple_to_date(md[1])
|
||
dasha['mahadasha_end'] = tuple_to_date(md[2])
|
||
if len(ladder) > 1:
|
||
ad = ladder[1]
|
||
dasha['antardasha_lord'] = DASHA_LORDS.get(ad[0][-1], str(ad[0][-1]))
|
||
dasha['antardasha_start'] = tuple_to_date(ad[1])
|
||
dasha['antardasha_end'] = tuple_to_date(ad[2])
|
||
except Exception as exc:
|
||
dasha['error'] = f'{type(exc).__name__}: {exc}'
|
||
|
||
return {
|
||
'sample_id': sample['id'],
|
||
'engine': 'PyJHora_4_8_6_lahiri_patched',
|
||
'parameters': {
|
||
'zodiac': 'sidereal',
|
||
'ayanamsa': 'LAHIRI',
|
||
'd9_method': 'PyJHora divisional_chart chart_method=1',
|
||
'd10_method': 'PyJHora divisional_chart chart_method=1',
|
||
'dasha_year': 'mean sidereal year',
|
||
'compat': 'monkeypatch swisseph keyword API + missing constants; dummy timezonefinder only for import',
|
||
'license_note': 'PyJHora is AGPL-3.0; used only as external benchmark, not vendored into skill.'
|
||
},
|
||
'ascendant': {
|
||
'sign': asc.get('sign'),
|
||
'degree_in_sign': asc.get('degree_in_sign'),
|
||
},
|
||
'planets': planets,
|
||
'varga': {'D9': d9, 'D10': d10},
|
||
'dasha': dasha,
|
||
}
|
||
|
||
|
||
def compare_scalar(rows, sample_id, section, body, field, local_value, pyjhora_value, tolerance=None, date_tolerance_days=None, boundary_sensitive=False, status_override=None):
|
||
status = status_override or 'match'
|
||
delta = ''
|
||
if not status_override:
|
||
if date_tolerance_days is not None:
|
||
try:
|
||
ld = datetime.strptime(str(local_value), '%Y-%m-%d')
|
||
pd = datetime.strptime(str(pyjhora_value), '%Y-%m-%d')
|
||
delta_val = abs((ld - pd).days)
|
||
delta = delta_val
|
||
status = 'match' if delta_val <= date_tolerance_days else 'mismatch'
|
||
except Exception:
|
||
status = 'not_comparable'
|
||
elif tolerance is not None:
|
||
try:
|
||
delta_val = abs(float(local_value) - float(pyjhora_value))
|
||
delta = round(delta_val, 6)
|
||
status = 'match' if delta_val <= tolerance else 'mismatch'
|
||
except Exception:
|
||
status = 'not_comparable'
|
||
else:
|
||
status = 'match' if local_value == pyjhora_value else 'mismatch'
|
||
if status == 'mismatch' and boundary_sensitive:
|
||
status = 'boundary_sensitive'
|
||
rows.append({
|
||
'sample_id': sample_id,
|
||
'section': section,
|
||
'body': body,
|
||
'field': field,
|
||
'local_skill': local_value,
|
||
'pyjhora': pyjhora_value,
|
||
'delta': delta,
|
||
'status': status,
|
||
})
|
||
|
||
|
||
def compare_one(sample_id, local, pyjhora):
|
||
rows = []
|
||
compare_scalar(rows, sample_id, 'ascendant', 'Ascendant', 'sign', local['ascendant'].get('sign'), pyjhora['ascendant'].get('sign'))
|
||
compare_scalar(rows, sample_id, 'ascendant', 'Ascendant', 'degree_in_sign', local['ascendant'].get('degree_in_sign'), pyjhora['ascendant'].get('degree_in_sign'), tolerance=0.15)
|
||
for p in PLANETS:
|
||
l = local['planets'].get(p, {})
|
||
y = pyjhora['planets'].get(p, {})
|
||
for field in ['sign', 'nakshatra', 'nakshatra_pada']:
|
||
compare_scalar(rows, sample_id, 'planet', p, field, l.get(field), y.get(field))
|
||
compare_scalar(rows, sample_id, 'planet', p, 'degree_in_sign', l.get('degree_in_sign'), y.get('degree_in_sign'), tolerance=0.15)
|
||
for varga_name in ['D9', 'D10']:
|
||
for body in ['Ascendant'] + PLANETS:
|
||
l = (local['varga'].get(varga_name) or {}).get(body) or {}
|
||
y = (pyjhora['varga'].get(varga_name) or {}).get(body) or {}
|
||
boundary_sensitive = False
|
||
try:
|
||
boundary_sensitive = abs(float(l.get('degree_in_sign', 99)) - float(y.get('degree_in_sign', -99))) > 20 and l.get('sign') != y.get('sign')
|
||
except Exception:
|
||
pass
|
||
compare_scalar(rows, sample_id, varga_name, body, 'sign', l.get('sign'), y.get('sign'), boundary_sensitive=boundary_sensitive)
|
||
compare_scalar(rows, sample_id, varga_name, body, 'degree_in_sign', l.get('degree_in_sign'), y.get('degree_in_sign'), tolerance=0.2, boundary_sensitive=boundary_sensitive)
|
||
# PyJHora dasha is useful as external signal, but currently has different default starting convention/seed in some cases.
|
||
# Keep fields in matrix, with generous date tolerance; differences are classified below in report.
|
||
for field in ['mahadasha_lord', 'antardasha_lord']:
|
||
compare_scalar(rows, sample_id, 'dasha', 'Vimshottari_current', field, local['dasha'].get(field), pyjhora['dasha'].get(field))
|
||
for field in ['mahadasha_start', 'mahadasha_end', 'antardasha_start', 'antardasha_end']:
|
||
compare_scalar(rows, sample_id, 'dasha', 'Vimshottari_current', field, local['dasha'].get(field), pyjhora['dasha'].get(field), date_tolerance_days=7)
|
||
return rows
|
||
|
||
|
||
def write_report(samples, rows):
|
||
total = len(rows)
|
||
counts = {}
|
||
by_section = {}
|
||
for r in rows:
|
||
counts[r['status']] = counts.get(r['status'], 0) + 1
|
||
stat = by_section.setdefault(r['section'], {'total': 0})
|
||
stat['total'] += 1
|
||
stat[r['status']] = stat.get(r['status'], 0) + 1
|
||
matches = counts.get('match', 0)
|
||
mismatches = [r for r in rows if r['status'] == 'mismatch']
|
||
boundary = [r for r in rows if r['status'] == 'boundary_sensitive']
|
||
non_dasha_rows = [r for r in rows if r['section'] != 'dasha']
|
||
non_dasha_match = sum(1 for r in non_dasha_rows if r['status'] == 'match')
|
||
non_dasha_ok = sum(1 for r in non_dasha_rows if r['status'] in ('match', 'boundary_sensitive'))
|
||
|
||
lines = []
|
||
lines.append('# Jyotish benchmark 第三轮：PyJHora 对比报告')
|
||
lines.append('')
|
||
lines.append('生成时间：2026-06-03')
|
||
lines.append('')
|
||
lines.append('## 1. 本轮范围')
|
||
lines.append('')
|
||
lines.append('- 外部引擎：PyJHora 4.8.6。')
|
||
lines.append('- 用途：第二个独立 Jyotish 开源项目对标，重点验证 D1、D9、D10，并初探 Vimshottari。')
|
||
lines.append('- 样本：10个公开/虚构 smoke case，不包含用户个人资料。')
|
||
lines.append('- 口径：强制 Lahiri；PyJHora 默认 TRUE_PUSHYA，因此本轮显式切换到 LAHIRI。')
|
||
lines.append('- 兼容处理：PyJHora 4.8.6 与本机 pyswisseph API 存在关键字参数/常量兼容问题，本脚本只在 benchmark 进程内 monkeypatch，不改 PyJHora 源码，不把 AGPL 代码并入 skill。')
|
||
lines.append('')
|
||
lines.append('## 2. 总体结果')
|
||
lines.append('')
|
||
lines.append(f'- 字段总数：{total}')
|
||
lines.append(f'- 匹配：{matches}')
|
||
lines.append(f'- 不匹配：{len(mismatches)}')
|
||
lines.append(f'- 边界敏感：{len(boundary)}')
|
||
lines.append(f'- 总严格匹配率：{matches / total:.2%}' if total else '- 总严格匹配率：N/A')
|
||
lines.append(f'- 非 Dasha 字段严格匹配率：{non_dasha_match / len(non_dasha_rows):.2%}' if non_dasha_rows else '- 非 Dasha 字段严格匹配率：N/A')
|
||
lines.append(f'- 非 Dasha 字段边界归因后可接受率：{non_dasha_ok / len(non_dasha_rows):.2%}' if non_dasha_rows else '- 非 Dasha 字段边界归因后可接受率：N/A')
|
||
lines.append('')
|
||
lines.append('## 3. 分区统计')
|
||
lines.append('')
|
||
lines.append('| Section | Total | Match | Mismatch | Boundary sensitive | Not comparable |')
|
||
lines.append('|---|---:|---:|---:|---:|---:|')
|
||
for section, stat in sorted(by_section.items()):
|
||
lines.append(f"| {section} | {stat.get('total',0)} | {stat.get('match',0)} | {stat.get('mismatch',0)} | {stat.get('boundary_sensitive',0)} | {stat.get('not_comparable',0)} |")
|
||
lines.append('')
|
||
if mismatches:
|
||
lines.append('## 4. 不匹配字段')
|
||
lines.append('')
|
||
lines.append('| Sample | Section | Body | Field | Local skill | PyJHora | Delta |')
|
||
lines.append('|---|---|---|---|---|---|---:|')
|
||
for r in mismatches[:160]:
|
||
lines.append(f"| {r['sample_id']} | {r['section']} | {r['body']} | {r['field']} | {r['local_skill']} | {r['pyjhora']} | {r['delta']} |")
|
||
lines.append('')
|
||
if boundary:
|
||
lines.append('## 4b. 边界敏感字段')
|
||
lines.append('')
|
||
lines.append('| Sample | Section | Body | Field | Local skill | PyJHora | Delta |')
|
||
lines.append('|---|---|---|---|---|---|---:|')
|
||
for r in boundary[:80]:
|
||
lines.append(f"| {r['sample_id']} | {r['section']} | {r['body']} | {r['field']} | {r['local_skill']} | {r['pyjhora']} | {r['delta']} |")
|
||
lines.append('')
|
||
lines.append('## 5. 判断')
|
||
lines.append('')
|
||
lines.append('- PyJHora 作为第二开源引擎已经接入成功。')
|
||
lines.append('- D1/D9/D10若高匹配，说明当前 skill 的分盘算法不仅与 Swiss direct 自算一致，也能通过独立 Jyotish 项目的实测。')
|
||
lines.append('- Dasha 部分若存在系统性差异，优先视为 PyJHora seed_star / dasha year / 起运规则口径差异，不能马上判定本 skill 错；需要 JHora 或 Drik Panchang 再仲裁。')
|
||
lines.append('- PyJHora 是 AGPL-3.0，适合做外部 benchmark，不适合把其源码或派生实现并入当前 skill。')
|
||
return '\n'.join(lines)
|
||
|
||
|
||
def main():
|
||
PYJHORA_OUT.mkdir(parents=True, exist_ok=True)
|
||
samples = json.loads(DATA.read_text())
|
||
all_rows = []
|
||
for sample in samples:
|
||
pyjhora = build_pyjhora_sample(sample)
|
||
(PYJHORA_OUT / f"{sample['id']}.pyjhora.json").write_text(json.dumps(pyjhora, ensure_ascii=False, indent=2))
|
||
local = json.loads((LOCAL_CANON / f"{sample['id']}.canonical.json").read_text())
|
||
all_rows.extend(compare_one(sample['id'], local, pyjhora))
|
||
|
||
matrix = OUT / 'pyjhora_comparison_matrix.csv'
|
||
with matrix.open('w', newline='') as f:
|
||
writer = csv.DictWriter(f, fieldnames=['sample_id', 'section', 'body', 'field', 'local_skill', 'pyjhora', 'delta', 'status'])
|
||
writer.writeheader()
|
||
writer.writerows(all_rows)
|
||
|
||
report = OUT / 'jyotish_benchmark_round3_pyjhora_compare.md'
|
||
report.write_text(write_report(samples, all_rows))
|
||
print(json.dumps({'report': str(report), 'matrix': str(matrix), 'samples': len(samples), 'fields': len(all_rows)}, ensure_ascii=False, indent=2))
|
||
|
||
|
||
if __name__ == '__main__':
|
||
main()
|