#!/usr/bin/env python3 """ 逻辑正确性验证框架 v2 对比 Skill 引擎和 PyJhora 对60张名人星盘的 Yoga 检测结果 """ import json, sys, os from pathlib import Path # ==== 路径 ==== PROJECT_DIR = Path(__file__).resolve().parents[1] SKILL_DIR = str(PROJECT_DIR) RULES_PATH = os.path.join(SKILL_DIR, 'references', 'yoga_rules.json') from yoga_engine import YogaEngine # CROSS_NAME_MAP: PyJhora Yoga名 → Skill rule_id # 复用 validate_comprehensive.py 中的映射（简化版，只保留核心映射） # ---- 从 validate_comprehensive.py 内联的映射逻辑 ---- PYJHORA_VARIANTS = { # Dharidhra variants 'dharidhra_yoga_144': 'bvr_dharidhra_precise', 'dharidhra_yoga_147': 'bvr_dharidhra_precise', 'dharidhra_yoga_148': 'bvr_dharidhra_precise', 'dharidhra_yoga_149': 'bvr_dharidhra_precise', 'dharidhra_yoga_150': 'bvr_dharidhra_precise', 'dharidhra_yoga_151': 'bvr_dharidhra_precise', 'dharidhra_yoga_152': 'bvr_dharidhra_precise', # Dehasthoulya variants 'dehasthoulya_yoga_114': 'bvr_dehasthoulya_yoga', 'dehasthoulya_yoga_115': 'bvr_dehasthoulya_yoga', 'dehasthoulya_yoga_116': 'bvr_dehasthoulya_yoga', # Mathibhramana variants 'mathibhramana_yoga_291': 'bvr_mathibhramana_yoga', 'mathibhramana_yoga_292': 'bvr_mathibhramana_yoga', 'mathibhramana_yoga_293': 'bvr_mathibhramana_yoga', 'mathibhramana_yoga_294': 'bvr_mathibhramana_yoga', 'mathibhramana_yoga_variation': 'bvr_mathibhramana_yoga', # Krisanga variants 'krisanga_yoga_112': 'bvr_krisanga_yoga', 'krisanga_yoga_113': 'bvr_krisanga_yoga', # Kaalanirdesat variants 'kaalanirdesat_puthra_yoga_227': 'bvr_kaalanirdesat_puthra_yoga', 'kaalanirdesat_puthra_yoga_228': 'bvr_kaalanirdesat_puthra_yoga', 'kaalanirdesat_puthranaasa_yoga_230': 'bvr_kaalanirdesat_puthranaasa_yoga', # Matrunasa variants 'matrunasa_yoga_198': 'bvr_matrunasa_precise', 'matrunasa_yoga_199': 'bvr_matrunasa_precise', # Matrudeerghayur variants 'matrudeerghayur_yoga_196': 'bvr_matrudeerghayur_yoga', 'matrudeerghayur_yoga_197': 'bvr_matrudeerghayur_yoga', # Nishkapata variants 'nishkapata_yoga_205': 'bvr_nishkapata_precise', 'nishkapata_yoga_206': 'bvr_nishkapata_precise', # Bahu puthra variants 'bahu_puthra_yoga_220': 'bvr_bahu_puthra_precise', 'bahu_puthra_yoga_221': 'bvr_bahu_puthra_precise', # Bandhu pujya variants 'bandhu_pujya_yoga_193': 'bvr_bandhu_pujya_yoga', 'bandhu_pujya_yoga_194': 'bvr_bandhu_pujya_yoga', # Ayatna griha prapta variants 'ayatna_griha_prapta_yoga_189': 'bvr_ayatna_griha_prapta_yoga', 'ayatna_griha_prapta_yoga_190': 'bvr_ayatna_griha_prapta_yoga', # Grihanasa variants 'grihanasa_yoga_191': 'bvr_grihanasa_yoga', 'grihanasa_yoga_192': 'bvr_grihanasa_yoga', # Dattha puthra variants 'dattha_puthra_yoga_222': 'bvr_dattha_puthra_yoga', 'dattha_puthra_yoga_223': 'bvr_dattha_puthra_yoga', # Kushtaroga variants 'kushtaroga_yoga_268': 'bvr_kushtaroga_yoga', # Sarpasaapa variants 'sarpasaapa_yoga_212': 'bvr_sarpasaapa_yoga', 'sarpasaapa_yoga_213': 'bvr_sarpasaapa_yoga', # Raja bhanga variants 'raja_bhanga_yoga_298': 'bvr_raja_bhanga_yoga', 'raja_bhanga_yoga_299': 'bvr_raja_bhanga_yoga', # Andha variants 'andha_yoga_288': 'bvr_andha_yoga', 'andha_yoga_289': 'bvr_andha_yoga', # Vahana variants 'vahana_yoga_209': 'bvr_vahana_yoga', 'vahana_yoga_210': 'bvr_vahana_yoga', # Kapata family 'kapata_yoga_202': 'bvr_kapata_yoga', 'kapata_yoga_203': 'bvr_kapata_yoga', 'kapata_yoga_204': 'bvr_kapata_yoga', 'pisacha_grastha_yoga': 'bvr_pisacha_grastha_yoga', 'kaalanirdesat_puthranaasa_yoga_229': 'bvr_kaalanirdesat_puthranaasa_yoga', 'dwadasa_sahodara_yoga': 'bvr_dwadasa_sahodara_yoga', 'sapthasankhya_sahodara_yoga': 'bvr_sapthasankhya_sahodara_yoga', 'karma_malika_yoga': 'bvr_karma_malika_precise', 'ardha_chandra_yoga': 'bvr_ardha_chandra_yoga', # Kalanidhi variants 'kalaanidhi_yoga': 'bvr_kalaanidhi_yoga', 'kalanidhi_yoga': 'bvr_kalaanidhi_yoga', # Soola/Sula 'soola_yoga': 'bvr_soola_yoga', 'sula_yoga': 'bvr_soola_yoga', # Kedara 'kedara_yoga': 'bvr_kedaara_yoga', # Kaahala/Kahala are distinct PyJHora functions: # - kaahala_yoga: L4 and Jupiter in mutual quadrants + strong L1 # - kahala_yoga: L4 and L9 in mutual kendras + strong L1 'kaahala_yoga': 'bvr_kaahala_yoga', 'kahala_yoga': 'kahala_yoga', # Direct mappings for common yogas 'vosi_yoga': 'bvr_vosi_precise', 'sunaphaa_yoga': 'bvr_sunaphaa_precise', 'amala_yoga': 'bvr_amala_precise', 'anaphaa_yoga': 'bvr_anaphaa_precise', 'sareera_soukhya_yoga': 'bvr_sareera_soukhya_precise', 'rogagrastha_yoga': 'bvr_rogagrastha_precise', 'sada_sanchara_yoga': 'bvr_sada_sanchara_precise', 'swaveeryaddhana_yoga': 'bvr_swaveeryaddhana_precise', 'anthya_vayasi_dhana_yoga': 'bvr_anthya_vayasi_dhana_yoga', 'matrumooladdhana_yoga': 'bvr_matrumooladdhana_yoga', 'yuddha_praveena_yoga': 'bvr_yuddha_praveena_yoga', 'bandhubhisthyaktha_yoga': 'bvr_bandhubhisthyaktha_precise', 'sraddhannabhuktha_yoga': 'bvr_sraddhannabhuktha_precise', 'bhratruvriddhi_yoga': 'bvr_bhratruvriddhi_precise', 'sodaranasa_yoga': 'bvr_sodaranasa_yoga', 'dehapushti_yoga': 'bvr_dehapushti_yoga', 'mridanga_yoga': 'bvr_mridanga_yoga', 'bheri_yoga': 'bvr_bheri_yoga', 'pushkala_yoga': 'bvr_pushkala_yoga', 'parakrama_yoga': 'bvr_parakrama_yoga', 'lakshmi_yoga': 'bvr_lakshmi_yoga', 'adhi_yoga': 'bvr_adhi_yoga', 'brahma_yoga': 'bvr_brahma_yoga', 'bhandhana_yoga': 'bvr_bhandhana_yoga', 'harihara_brahma_yoga': 'bvr_harihara_brahma_yoga', 'hara_yoga': 'bvr_hara_yoga', 'hari_yoga': 'bvr_hari_yoga', 'madhya_vayasi_dhana_yoga': 'bvr_madhya_vayasi_dhana_yoga', 'balya_dhana_yoga': 'bvr_balya_dhana_yoga', 'go_yoga': 'bvr_go_yoga', 'sara_yoga': 'bvr_sara_yoga', 'ishu_yoga': 'bvr_ishu_yoga', 'kshayaroga_yoga': 'bvr_kshayaroga_yoga', 'kalatramooladdhana_yoga': 'bvr_kalatramooladdhana_yoga', 'vishnu_yoga': 'bvr_vishnu_yoga', 'vichitra_saudha_prakara_yoga': 'bvr_vichitra_saudha_prakara_yoga', 'jananatpurvam_pitru_marana_yoga': 'bvr_jananatpurvam_pitru_marana_yoga', 'bhratrumooladdhanaprapti_yoga': 'bvr_bhratrumooladdhanaprapti_yoga', 'thrikaala_gnana_yoga': 'bvr_thrikaala_gnana_yoga', 'yuddhatpaschaddrudha_yoga': 'bvr_yuddhatpaschaddrudha_yoga', 'yuddhatpoorvadridhachitta_yoga': 'bvr_yuddhatpoorvadridhachitta_yoga', 'dhurmarana_yoga': 'bvr_dhurmarana_yoga', 'theevrabuddhi_yoga': 'bvr_theevrabuddhi_yoga', 'vaatharoga_yoga': 'bvr_vaatharoga_yoga', 'amaranantha_dhana_yoga': 'bvr_amaranantha_dhana_yoga', 'pittharoga_yoga': 'bvr_pittharoga_yoga', 'sarpasaapa_yoga': 'bvr_sarpasaapa_yoga', 'parannabhojana_yoga': 'bvr_parannabhojana_yoga', 'veenaa_yoga': 'bvr_veenaa_yoga', 'duradhara_yoga': 'bvr_duradhara_yoga', 'annadana_yoga': 'bvr_annadana_yoga', 'yukthi_samanwithavagmi_yoga_154': 'bvr_yukthi_samanwithavagmi_yoga', 'nipuna_yoga': 'bvr_nipuna_yoga', 'surya_budha_yoga': 'bvr_nipuna_yoga', 'parvata_yoga': 'bvr_parvata_yoga', 'hamsa_yoga': 'bvr_hamsa_yoga', 'ruchaka_yoga': 'bvr_ruchaka_yoga', 'bhadra_yoga': 'bvr_bhadra_yoga', 'chatussagara_yoga': 'bvr_chatussagara_yoga', 'sarpa_yoga': 'bvr_sarpa_yoga', 'kemadruma_yoga': 'bvr_kemadruma_yoga', 'sankha_yoga': 'bvr_sankha_yoga', 'sumukha_yoga': 'bvr_sumukha_yoga', 'ubhayachara_yoga': 'bvr_ubhayachara_yoga', 'sasa_yoga': 'bvr_sasa_yoga', 'matsya_yoga': 'bvr_matsya_yoga', 'koorma_yoga': 'bvr_koorma_yoga', 'naukaa_yoga': 'bvr_naukaa_yoga', 'vihaga_yoga': 'bvr_vihaga_yoga', 'yuga_yoga': 'bvr_yuga_yoga', 'gola_yoga': 'bvr_gola_yoga', # Fix previously unmapped 'matru_sneha_yoga': 'bvr_matru_sneha_yoga', 'eka_puthra_yoga': 'bvr_eka_puthra_precise', 'guru_mangala_yoga': 'bvr_103_guru_mangala_yoga', 'vanchana_chora_bheethi_yoga': 'bvr_vanchana_chora_bheethi_yoga', 'bahu_sthree_yoga': 'bvr_bahu_sthree_yoga', 'dama_yoga': 'bvr_dama_yoga', 'utthama_graha_yoga': 'bvr_utthama_graha_yoga', 'chandra_mangala_yoga': 'bvr_105_chandra_mangala_yoga', 'bhaga_chumbana_yoga': 'bvr_bhaga_chumbana_yoga', 'yuddha_marana_yoga': 'bvr_yuddha_marana_yoga', 'andha_yoga': 'bvr_andha_yoga', 'matru_satrutwa_yoga': 'bvr_matru_satrutwa_yoga', 'apakeerthi_yoga': 'bvr_apakeerthi_yoga', } RULE_ID_ALIASES = { # Historical/numbered rule ids in yoga_rules.json 'bvr_vosi_precise': 'bvr_017_vosi_precise', 'bvr_sunaphaa_precise': 'bvr_002_sunapha_precise', 'bvr_anaphaa_precise': 'bvr_003_anapha_precise', 'bvr_duradhara_yoga': 'bvr_004_duradhara_precise', 'bvr_ubhayachara_yoga': 'bvr_018_ubhayachara_precise', 'bvr_dharidhra_precise': 'bvr_dharidhra_11_precise', 'bvr_sara_yoga': 'bvr_072_sara', 'bvr_ishu_yoga': 'bvr_072_ishu', 'bvr_naukaa_yoga': 'bvr_075_naukaa_precise', 'bvr_vihaga_yoga': 'vihaga_nabhasa', 'bvr_veenaa_yoga': 'veena_yoga', 'bvr_sumukha_yoga': 'bvr_sumukha_precise', } def canonical_rule_id(rule_id, valid_rule_ids): """Map historical validation ids to actual enabled yoga_rules.json ids.""" if not rule_id: return None if rule_id in valid_rule_ids: return rule_id alias = RULE_ID_ALIASES.get(rule_id) if alias in valid_rule_ids: return alias if rule_id.startswith('bvr_'): without_bvr = rule_id[4:] candidates = [without_bvr] if without_bvr.endswith('_precise'): base = without_bvr[:-8] candidates.extend([base, f'{base}_yoga']) for candidate in candidates: if candidate in valid_rule_ids: return candidate return None def extract_skill_rule_ids(skill_results): """Extract comparable rule ids while ignoring algorithmic Yoga rows.""" ids = set() for row in skill_results: if not isinstance(row, dict): continue rule_id = row.get('rule_id') or row.get('id') if isinstance(rule_id, str) and rule_id: ids.add(rule_id) return ids # ---- 内联结束 ---- def main(): # 1. 加载位置数据。固定使用 skill 内部 references，避免从其他 cwd 运行时读写错目录。 planet_positions_path = os.path.join(SKILL_DIR, 'references', 'planet_positions_60.json') standard_charts_path = os.path.join(SKILL_DIR, 'references', 'standard_test_charts.json') report_path = os.path.join(SKILL_DIR, 'references', 'validation_logic_report.json') with open(planet_positions_path) as f: pos_data = json.load(f) # 2. 加载 PyJhora Yoga 结果 with open(standard_charts_path) as f: pyj_data = json.load(f) # 构建 name -> pyjhora yoga list 映射 pyj_charts = {c['name']: c for c in pyj_data['charts']} # 3. 加载 Skill 引擎 engine = YogaEngine(RULES_PATH) # 构建 rule_id -> rule_name 映射 rule_id_to_name = {r['id']: r['name'] for r in engine.rules} # 以及 name -> rule_id 反向映射（用于去重匹配） name_to_rule_ids = {} for r in engine.rules: name = r['name'] name_to_rule_ids.setdefault(name, []).append(r['id']) valid_rule_ids = set(rule_id_to_name.keys()) # 建立"可对比规则集"：仅保留真实存在且启用的 Skill rule_id，避免历史 ID 别名造成 60/60 假性 FN。 variant_to_rule_id = {} missing_mappings = {} for py_name, raw_id in PYJHORA_VARIANTS.items(): canonical = canonical_rule_id(raw_id, valid_rule_ids) if canonical: variant_to_rule_id[py_name] = canonical else: missing_mappings.setdefault(raw_id, []).append(py_name) comparable_rule_ids = set(variant_to_rule_id.values()) print(f"可对比规则数量: {len(comparable_rule_ids)}") if missing_mappings: print(f"暂不可对比映射数量: {len(missing_mappings)}") # 统计（只统计可对比规则） total_skill_comp = 0 # Skill在可对比规则集中的检测数 total_pyj_comp = 0 # PyJhora能映射到可对比规则的检测数 total_agreements = 0 total_false_positives = 0 total_false_negatives = 0 total_unmapped_pyj = 0 fp_details = [] # Skill说有，PyJhora说没有（都在可对比集内） fn_details = [] # PyJhora说有，Skill说没有（都在可对比集内） for chart in pos_data['charts']: name = chart['name'] planets = chart['planets'] ascendant = chart['ascendant'] # 从 standard_test_charts.json 取完整 context（含 D9/Navamsa/Upagraha 数据） pyj_chart = pyj_charts.get(name) if not pyj_chart: print(f" WARNING: {name} not in PyJhora data") continue context = pyj_chart.get('context', {}) # Skill 引擎检测（使用含 D9 的完整 context） skill_results = engine.detect(planets, ascendant, context=context) skill_yoga_ids = extract_skill_rule_ids(skill_results) skill_comp = skill_yoga_ids & comparable_rule_ids # 只保留可对比的 total_skill_comp += len(skill_comp) pyj_yoga_names = set(pyj_chart.get('expected_yogas', [])) # 将 PyJhora Yoga名映射到 Skill rule_id pyj_mapped_ids = set() unmapped = 0 for pyj_name in pyj_yoga_names: mapped = variant_to_rule_id.get(pyj_name) if not mapped: norm = pyj_name.lower().strip() mapped = variant_to_rule_id.get(norm) if mapped: pyj_mapped_ids.add(mapped) else: unmapped += 1 total_unmapped_pyj += unmapped pyj_comp = pyj_mapped_ids & comparable_rule_ids total_pyj_comp += len(pyj_comp) # 对比（只对比可对比规则集） agreements = skill_comp & pyj_comp false_positives = skill_comp - pyj_comp # Skill说有，PyJhora说没有 false_negatives = pyj_comp - skill_comp # PyJhora说有，Skill说没有 total_agreements += len(agreements) total_false_positives += len(false_positives) total_false_negatives += len(false_negatives) if false_positives: for rid in sorted(false_positives): fp_details.append({ 'chart': name, 'rule_id': rid, 'rule_name': rule_id_to_name.get(rid, '?'), }) if false_negatives: for rid in sorted(false_negatives): orig_names = sorted(pn for pn, sid in variant_to_rule_id.items() if sid == rid) fn_details.append({ 'chart': name, 'rule_id': rid, 'rule_name': rule_id_to_name.get(rid, '?'), 'external_benchmark_names': orig_names, }) # ==== 输出报告 ==== print("=" * 70) print("Skill vs PyJhora 逻辑正确性验证报告（仅可对比规则）") print("=" * 70) print(f"测试星盘数量: 60") print(f"可对比规则集大小: {len(comparable_rule_ids)}") print(f"PyJhora未映射Yoga: {total_unmapped_pyj}") print() print(f"Skill 可对比检测总数: {total_skill_comp}") print(f"PyJhora 可对比检测总数: {total_pyj_comp}") print(f"一致 (Agreements): {total_agreements}") print(f"False Positives (Skill过宽): {total_false_positives}") print(f"False Negatives (Skill过严): {total_false_negatives}") print() precision = total_agreements / total_skill_comp if total_skill_comp else 0 recall = total_agreements / total_pyj_comp if total_pyj_comp else 0 f1 = 2 * precision * recall / (precision + recall) if (precision + recall) else 0 print(f"Precision (准确率): {precision:.2%}") print(f"Recall (召回率): {recall:.2%}") print(f"F1 Score: {f1:.2%}") print() # 按规则汇总 FP if fp_details: from collections import Counter fp_by_rule = Counter((d['rule_id'], d['rule_name']) for d in fp_details) print("--- False Positives 最多的规则 (Top 20) ---") for (rid, rname), count in fp_by_rule.most_common(20): print(f" {count:2d} 次: {rid} ({rname})") print() # 按规则汇总 FN if fn_details: from collections import Counter fn_by_rule = Counter((d['rule_id'], d['rule_name']) for d in fn_details) print("--- False Negatives 最多的规则 (Top 20) ---") for (rid, rname), count in fn_by_rule.most_common(20): print(f" {count:2d} 次: {rid} ({rname})") print() # 保存详细报告 report = { "summary": { "charts_tested": 60, "comparable_rules": len(comparable_rule_ids), "skill_total": total_skill_comp, "external_benchmark_total": total_pyj_comp, "unmapped_external_benchmark": total_unmapped_pyj, "missing_mappings": {k: v for k, v in sorted(missing_mappings.items())}, "agreements": total_agreements, "false_positives": total_false_positives, "false_negatives": total_false_negatives, "precision": round(precision, 4), "recall": round(recall, 4), "f1": round(f1, 4), }, "false_positives": sorted(fp_details, key=lambda row: (row["chart"], row["rule_id"])), "false_negatives": sorted(fn_details, key=lambda row: (row["chart"], row["rule_id"])), } with open(report_path, 'w') as f: json.dump(report, f, indent=2, ensure_ascii=False) print(f"详细报告已保存至: {report_path}") if __name__ == "__main__": main()