"AI \u80fd\u81ea\u52a8\u67e5\u6570\u3001\u80fd\u81ea\u52a8\u5199 SQL\u3001\u80fd\u81ea\u52a8\u751f\u6210\u62a5\u544a\u2014\u2014\u90a3\u8fd8\u8981\u6570\u636e\u5206\u6790\u5e08\u5e72\u561b？"<\/p>\n
\u8fd9\u4e2a\u95ee\u9898\u6211\u4ece 2023 \u5e74\u88ab\u95ee\u5230 2026 \u5e74，\u6bcf\u6b21\u6211\u7684\u56de\u7b54\u90fd\u4e00\u6837：\u6570\u636e\u5206\u6790\u5e08\u4e0d\u662f\u5728\u548c AI \u62a2\u8dd1\u6570\u636e\u7684\u6d3b，\u800c\u662f\u5728\u548c AI \u4e89"\u5224\u65ad"\u7684\u6d3b\u3002\u800c\u5224\u65ad\u8fd9\u4ef6\u4e8b，\u6070\u6070\u662f\u6280\u672f\u8fdb\u6b65\u8d8a\u591a，\u4eba\u7684\u4ef7\u503c\u8d8a\u5927\u7684\u8d5b\u9053\u3002<\/p>\n
\u4e3a\u4ec0\u4e48？\u56e0\u4e3a AI \u89e3\u51b3\u7684\u6c38\u8fdc\u662f"\u6548\u7387"\u95ee\u9898\u2014\u2014\u5b83\u8ba9\u83b7\u53d6\u4fe1\u606f\u66f4\u5feb\u3001\u66f4\u4fbf\u5b9c\u3002\u4f46\u5b83\u89e3\u51b3\u4e0d\u4e86"\u5728\u5927\u91cf\u4fe1\u606f\u4e2d，\u54ea\u4e2a\u771f\u6b63\u91cd\u8981"\u7684\u4f18\u5148\u7ea7\u95ee\u9898\u3002<\/p>\n
\u4e3e\u4e2a\u4f8b\u5b50：AI \u53ef\u4ee5\u5728\u4e00\u79d2\u5185\u7ed9\u4f60 200 \u6761\u4e1a\u52a1\u5f02\u5e38\u9884\u8b66，\u4f46\u4f5c\u4e3a\u4e00\u4e2a\u6570\u636e\u5206\u6790\u5e08，\u4f60\u7684\u4ef7\u503c\u5728\u4e8e\u80fd\u51c6\u786e\u5224\u65ad\u51fa"\u8fd9 200 \u6761\u91cc，\u54ea 3 \u6761\u4f1a\u5bfc\u81f4\u4e0b\u5468 GMV \u6389 30%，\u54ea 197 \u6761\u53ea\u662f\u81ea\u7136\u6ce2\u52a8，\u4e0d\u7528\u7ba1"\u3002\u8fd9\u4e2a\u5224\u65ad\u529b，\u9760\u7684\u662f\u4e1a\u52a1\u7406\u89e3\u3001\u7ecf\u9a8c\u79ef\u7d2f\u548c\u5bf9\u516c\u53f8\u6218\u7565\u65b9\u5411\u7684\u628a\u63e1\u2014\u2014\u8fd9\u4e9b\u90fd\u662f AI \u4e0d\u64c5\u957f\u7684\u3002<\/p>\n
\u6280\u672f\u8d8a\u5f3a，\u4fe1\u606f\u8d8a\u591a\u3001\u566a\u97f3\u8d8a\u591a，"\u7b5b\u9009\u548c\u5224\u65ad"\u7684\u4ef7\u503c\u5c31\u8d8a\u5927\u3002\u8fd9\u5c31\u662f\u4e3a\u4ec0\u4e48\u6211\u8bf4"\u6280\u672f\u8d8a\u5f3a，\u4eba\u7684\u4ef7\u503c\u53cd\u800c\u8d8a\u5927"\u3002<\/p>\n
\u6211\u8ba4\u4e3a AI \u65f6\u4ee3\u7684\u6570\u636e\u5206\u6790\u5e08\u89d2\u8272\u4f1a\u7ecf\u5386\u4e09\u5c42\u9012\u8fdb：<\/p>\n
"""
\nAI \u65f6\u4ee3\u6570\u636e\u5206\u6790\u5e08\u7684\u534f\u4f5c\u6a21\u5f0f：AI \u5feb\u901f\u63a2\u7d22 + \u4eba\u6df1\u5ea6\u9a8c\u8bc1
\n"""
\nimport pandas as pd
\nimport numpy as np
\nfrom scipy import stats<\/p>\n
# \u6a21\u62df\u4e00\u4efd\u4e1a\u52a1\u6570\u636e
\nnp.random.seed(42)
\nn = 2000
\ndf = pd.DataFrame({
\n 'region': np.random.choice(['\u534e\u4e1c', '\u534e\u5357', '\u534e\u5317', '\u897f\u5357'], n),
\n 'channel': np.random.choice(['App', 'PC', '\u5c0f\u7a0b\u5e8f', '\u7ebf\u4e0b'], n),
\n 'promo_active': np.random.choice([0, 1], n), # \u662f\u5426\u53c2\u4e0e\u4fc3\u9500
\n 'is_weekend': np.random.choice([0, 1], n),
\n 'gmv': np.random.uniform(100, 5000, n)
\n})<\/p>\n
# === AI \u505a\u7684：\u5feb\u901f\u63a2\u7d22，\u63d0\u4f9b\u5019\u9009\u5047\u8bf4 ===
\n# （\u5b9e\u9645\u573a\u666f\u4e2d，\u8fd9\u6b65\u7531\u5927\u6a21\u578b\u751f\u6210）
\nai_hypotheses = {
\n '\u5047\u8bf41': '\u4fc3\u9500\u6d3b\u52a8\u62c9\u5347\u4e86 GMV',
\n '\u5047\u8bf42': '\u5468\u672b GMV \u663e\u8457\u9ad8\u4e8e\u5de5\u4f5c\u65e5',
\n '\u5047\u8bf43': '\u534e\u4e1c\u533a\u8868\u73b0\u5f02\u5e38\u7a81\u51fa',
\n '\u5047\u8bf44': 'App \u6e20\u9053\u8d21\u732e\u6700\u5927',
\n}<\/p>\n
# === \u4eba\u505a\u7684：\u6df1\u5ea6\u9a8c\u8bc1，\u7b5b\u9009\u771f\u56e0 ===
\nclass HypothesisValidator:
\n """\u6570\u636e\u5206\u6790\u5e08\u7684\u6838\u5fc3\u4ef7\u503c：\u9a8c\u8bc1\u5047\u8bf4，\u800c\u975e\u751f\u6210\u5047\u8bf4"""<\/p>\n
def __init__(self, df: pd.DataFrame):
\n self.df = df
\n self.results = {}<\/p>\n
def validate_promo(self) -> dict:
\n """\u9a8c\u8bc1\u5047\u8bf41：\u4fc3\u9500\u662f\u5426\u771f\u7684\u62c9\u5347\u4e86 GMV"""
\n promo_on = self.df[self.df['promo_active'] == 1]['gmv']
\n promo_off = self.df[self.df['promo_active'] == 0]['gmv']<\/p>\n
# t \u68c0\u9a8c：\u4e24\u7ec4 GMV \u662f\u5426\u6709\u663e\u8457\u5dee\u5f02
\n t_stat, p_value = stats.ttest_ind(promo_on, promo_off)<\/p>\n
# \u6548\u5e94\u91cf（Cohen's d）：\u5dee\u5f02\u7684"\u5b9e\u9645\u5927\u5c0f"
\n pooled_std = np.sqrt((promo_on.std()**2 + promo_off.std()**2) \/ 2)
\n cohen_d = (promo_on.mean() – promo_off.mean()) \/ pooled_std<\/p>\n
return {
\n '\u5047\u8bf4': '\u4fc3\u9500\u62c9\u5347GMV',
\n '\u6709\u4fc3\u9500\u5747\u503c': f"{promo_on.mean():.1f}",
\n '\u65e0\u4fc3\u9500\u5747\u503c': f"{promo_off.mean():.1f}",
\n '\u63d0\u5347\u5e45\u5ea6': f"{(promo_on.mean() – promo_off.mean()) \/ promo_off.mean() * 100:.1f}%",
\n 'p\u503c': f"{p_value:.4f}",
\n '\u662f\u5426\u663e\u8457': '\u662f' if p_value < 0.05 else '\u5426',
\n '\u6548\u5e94\u91cf': f"{cohen_d:.3f}",
\n '\u662f\u5426\u91cd\u8981': '\u662f' if abs(cohen_d) > 0.2 else '\u6548\u5e94\u91cf\u592a\u5c0f，\u5b9e\u9645\u610f\u4e49\u6709\u9650'
\n }<\/p>\n
def validate_weekend(self) -> dict:
\n """\u9a8c\u8bc1\u5047\u8bf42：\u5468\u672b GMV \u662f\u5426\u771f\u7684\u66f4\u9ad8"""
\n weekend = self.df[self.df['is_weekend'] == 1]['gmv']
\n weekday = self.df[self.df['is_weekend'] == 0]['gmv']<\/p>\n
t_stat, p_value = stats.ttest_ind(weekend, weekday)<\/p>\n
return {
\n '\u5047\u8bf4': '\u5468\u672bGMV\u66f4\u9ad8',
\n '\u5468\u672b\u5747\u503c': f"{weekend.mean():.1f}",
\n '\u5de5\u4f5c\u65e5\u5747\u503c': f"{weekday.mean():.1f}",
\n '\u5dee\u5f02': f"{(weekend.mean() – weekday.mean()) \/ weekday.mean() * 100:.1f}%",
\n 'p\u503c': f"{p_value:.4f}",
\n '\u662f\u5426\u663e\u8457': '\u662f' if p_value < 0.05 else '\u5426'
\n }<\/p>\n
def validate_region(self) -> dict:
\n """\u9a8c\u8bc1\u5047\u8bf43：\u7528 ANOVA \u68c0\u9a8c\u533a\u57df\u5dee\u5f02"""
\n regions = {}
\n for region in self.df['region'].unique():
\n regions[region] = self.df[self.df['region'] == region]['gmv'].values<\/p>\n
f_stat, p_value = stats.f_oneway(*regions.values())<\/p>\n
return {
\n '\u5047\u8bf4': '\u533a\u57dfGMV\u5dee\u5f02\u663e\u8457',
\n '\u5404\u533a\u57df\u5747\u503c': {r: f"{v.mean():.1f}" for r, v in regions.items()},
\n 'F\u7edf\u8ba1\u91cf': f"{f_stat:.2f}",
\n 'p\u503c': f"{p_value:.4f}",
\n '\u662f\u5426\u663e\u8457': '\u662f' if p_value < 0.05 else '\u5426'
\n }<\/p>\n
# \u6267\u884c\u9a8c\u8bc1
\nvalidator = HypothesisValidator(df)<\/p>\n
print("=== AI \u63d0\u51fa\u4e86 4 \u4e2a\u5047\u8bf4，\u6570\u636e\u5206\u6790\u5e08\u9010\u4e00\u9a8c\u8bc1 ===\\\\n")<\/p>\n
for name, func in [
\n ('\u5047\u8bf41', validator.validate_promo),
\n ('\u5047\u8bf42', validator.validate_weekend),
\n ('\u5047\u8bf43', validator.validate_region),
\n]:
\n result = func()
\n print(f"\u3010{result['\u5047\u8bf4']}\u3011")
\n for k, v in result.items():
\n if k != '\u5047\u8bf4':
\n print(f" {k}: {v}")
\n print()<\/p>\n
\u8fd9\u662f\u6700\u9ad8\u5c42\u6b21，\u4e5f\u662f\u672a\u6765\u6570\u636e\u5206\u6790\u5e08\u771f\u6b63\u7684\u4ef7\u503c\u6240\u5728：<\/p>\n
\u5728 AI \u65f6\u4ee3，\u6211\u4e0d\u7126\u8651\u88ab\u66ff\u4ee3，\u4f46\u56e0\u4e3a\u6211\u77e5\u9053\u81ea\u5df1\u9700\u8981"\u6362\u6b66\u5668"：<\/p>\n
AI \u80fd\u5e2e\u4f60\u8dd1\u56de\u5f52，\u4f46\u5b83\u5206\u4e0d\u6e05 p-hacking \u548c\u771f\u6b63\u7684\u7edf\u8ba1\u63a8\u7406\u3002\u4f60\u9700\u8981\u80fd\u4ece\u65b9\u6cd5\u8bba\u5c42\u9762\u5224\u65ad AI \u7684\u5206\u6790\u662f\u5426\u9760\u8c31\u3002<\/p>\n
\u80fd\u628a\u4e00\u4e2a\u6a21\u7cca\u7684\u4e1a\u52a1\u95ee\u9898，\u8f6c\u5316\u4e3a\u7cbe\u786e\u7684\u6570\u636e\u95ee\u9898\u3002\u8fd9\u4e2a"\u7ffb\u8bd1"\u80fd\u529b，\u662f\u4efb\u4f55 ChatBI \u90fd\u505a\u4e0d\u597d\u7684\u3002<\/p>\n
\u6570\u636e+\u56fe\u8868\u4e0d\u7b49\u4e8e\u6d1e\u5bdf\u3002\u6d1e\u5bdf = \u6570\u636e + \u4e1a\u52a1\u8bed\u5883 + \u53ef\u884c\u52a8\u5efa\u8bae\u3002AI \u80fd\u7ed9\u524d\u4e24\u4e2a，\u7ed9\u4e0d\u4e86\u7b2c\u4e09\u4e2a\u3002<\/p>\n
\u77e5\u9053 AI \u80fd\u505a\u4ec0\u4e48\u3001\u4e0d\u80fd\u505a\u4ec0\u4e48\u3001\u600e\u4e48\u63d0\u95ee\u80fd\u5f97\u5230\u6700\u597d\u7684\u7ed3\u679c\u3002\u8fd9\u672c\u8eab\u662f\u4e00\u79cd"\u65b0\u578b\u7f16\u7a0b"\u2014\u2014\u4e0d\u662f\u4f60\u5199\u4ee3\u7801\u6307\u6325\u8ba1\u7b97\u673a，\u800c\u662f\u4f60\u5199 prompt \u6307\u6325\u5927\u6a21\u578b\u3002<\/p>\n
\u4e3a\u4ec0\u4e48\u6280\u672f\u8d8a\u5f3a，\u4eba\u7684\u4ef7\u503c\u53cd\u800c\u8d8a\u5927？<\/p>\n
\u56e0\u4e3a\u6280\u672f\u5f3a\u5927\u4e4b\u540e，"\u4fe1\u606f\u7684\u4e0d\u5bf9\u79f0"\u88ab\u5927\u5927\u964d\u4f4e\u4e86\u3002\u4ee5\u524d\u53ea\u6709\u4f60\u6570\u636e\u56e2\u961f\u77e5\u9053\u7684\u6570\u5b57，\u73b0\u5728\u8001\u677f\u81ea\u5df1\u5bf9\u8bdd AI \u4e5f\u80fd\u62ff\u5230\u3002\u4f46\u8fd9\u5c31\u66b4\u9732\u51fa\u4e00\u4e2a\u66f4\u6df1\u5c42\u7684 gap：\u62ff\u5230\u6570\u5b57\u4e4b\u540e\u600e\u4e48\u529e？<\/p>\n
\u4ee5\u524d\u6570\u636e\u56e2\u961f\u7684\u4ef7\u503c\u90e8\u5206\u85cf\u5728"\u6211\u77e5\u9053\u4f60\u4e0d\u77e5\u9053"\u91cc，\u672a\u6765\u6570\u636e\u56e2\u961f\u7684\u4ef7\u503c\u5168\u90e8\u4f53\u73b0\u5728"\u6211\u77e5\u9053\u4e4b\u540e\u8be5\u600e\u4e48\u505a"\u91cc\u3002\u800c"\u77e5\u9053\u8be5\u600e\u4e48\u505a"，\u672c\u8d28\u4e0a\u662f\u4e00\u7cfb\u5217\u5224\u65ad\u7684\u603b\u548c\u2014\u2014\u54ea\u4e9b\u503c\u5f97\u5173\u6ce8\u3001\u54ea\u4e9b\u503c\u5f97\u6295\u5165\u3001\u54ea\u4e9b\u5e94\u8be5\u653e\u5f03\u3002<\/p>\n
AI \u80fd\u7ed9\u4f60 1000 \u4e2a"\u53ef\u80fd\u7684\u6d1e\u5bdf"，\u4f46\u5b83\u7ed9\u4e0d\u4e86\u4f60"\u8fd9 1000 \u4e2a\u6d1e\u5bdf\u4e2d，\u6211\u4eec\u5e94\u8be5\u4f18\u5148\u6267\u884c\u54ea 3 \u4e2a"\u7684\u51b3\u65ad\u3002\u56e0\u4e3a\u8fd9\u4e2a\u51b3\u65ad，\u9700\u8981\u7406\u89e3\u516c\u53f8\u7684\u6218\u7565\u65b9\u5411\u3001\u56e2\u961f\u7684\u8d44\u6e90\u7ea6\u675f\u3001\u8001\u677f\u7684\u504f\u597d\u3001\u5e02\u573a\u7684\u98ce\u5411\u2014\u2014\u8fd9\u4e9b\u90fd\u662f\u786c\u90a6\u90a6\u7684\u6570\u636e\u4e4b\u5916\u7684"\u8f6f\u77e5\u8bc6"，\u76ee\u524d\u6ca1\u6709\u4efb\u4f55 AI \u7cfb\u7edf\u80fd\u638c\u63e1\u3002<\/p>\n
\u6211\u59cb\u7ec8\u76f8\u4fe1\u4e00\u4ef6\u4e8b：\u5de5\u5177\u53d8\u5f3a，\u4e0d\u4f1a\u6dd8\u6c70\u4f7f\u7528\u5de5\u5177\u7684\u4eba，\u53ea\u4f1a\u6dd8\u6c70\u62d2\u7edd\u5347\u7ea7\u7684\u4eba\u3002<\/p>\n
AI \u5728\u6570\u636e\u5206\u6790\u9886\u57df\u7684\u89d2\u8272，\u4e0d\u662f\u6570\u636e\u5206\u6790\u5e08\u7684"\u66ff\u4ee3\u8005"，\u800c\u662f"\u653e\u5927\u8005"\u3002\u5b83\u653e\u5927\u4e86\u4f60\u7684\u5206\u6790\u80fd\u529b（\u53ef\u4ee5\u505a\u66f4\u6df1\u5165\u7684\u63a2\u7d22），\u4e5f\u653e\u5927\u4e86\u4f60\u7684\u8d23\u4efb（\u4f60\u4e0d\u80fd\u518d\u8bf4"\u6570\u636e\u8fd8\u6ca1\u8dd1\u51fa\u6765"）\u3002<\/p>\n
\u672a\u6765\u4e09\u5e74，\u6211\u770b\u597d\u7684\u6570\u636e\u5206\u6790\u5e08\u753b\u50cf\u662f\u8fd9\u6837\u7684：<\/p>\n
\u6280\u672f\u8d8a\u5f3a，\u4eba\u7684\u4ef7\u503c\u53cd\u800c\u8d8a\u5927\u3002\u56e0\u4e3a\u6280\u672f\u6e05\u7a7a\u7684\u662f"\u4f4e\u4ef7\u503c\u91cd\u590d\u52b3\u52a8"，\u7559\u4e0b\u7684\u662f"\u9ad8\u4ef7\u503c\u7684\u5224\u65ad\u548c\u51b3\u7b56"\u2014\u2014\u800c\u8fd9，\u6c38\u8fdc\u662f\u4eba\u7c7b\u7684\u4e3b\u573a\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
AI \u6570\u636e\u5206\u6790\u5e08\u7684\u672a\u6765\u89d2\u8272：\u6280\u672f\u8d8a\u5f3a，\u4eba\u7684\u4ef7\u503c\u53cd\u800c\u8d8a\u5927
\n\u4e00\u3001\u5148\u56de\u7b54\u90a3\u4e2a\u6700\u8ba9\u4eba\u7126\u8651\u7684\u95ee\u9898
\n\\”AI \u80fd\u81ea\u52a8\u67e5\u6570\u3001\u80fd\u81ea\u52a8\u5199 SQL\u3001\u80fd\u81ea\u52a8\u751f\u6210\u62a5\u544a\u2014\u2014\u90a3\u8fd8\u8981\u6570\u636e\u5206\u6790\u5e08\u5e72\u561b？\\”
\n\u8fd9\u4e2a\u95ee\u9898\u6211\u4ece 2023 \u5e74\u88ab\u95ee\u5230 2026 \u5e74，\u6bcf\u6b21\u6211\u7684\u56de\u7b54\u90fd\u4e00\u6837：\u6570\u636e\u5206\u6790\u5e08\u4e0d\u662f\u5728\u548c AI \u62a2\u8dd1\u6570\u636e\u7684\u6d3b，\u800c\u662f\u5728\u548c AI \u4e89\\”\u5224\u65ad\\”\u7684\u6d3b\u3002\u800c\u5224\u65ad\u8fd9\u4ef6\u4e8b，\u6070\u6070\u662f\u6280\u672f\u8fdb\u6b65\u8d8a\u591a，\u4eba\u7684\u4ef7\u503c\u8d8a\u5927\u7684\u8d5b\u9053\u3002
\n\u4e3a<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50],"topic":[],"class_list":["post-88907","post","type-post","status-publish","format-standard","hentry","category-server","tag-50"],"yoast_head":"\n