\u6700\u8fd1\u7814\u5b66\u8fc7\u7a0b\u4e2d\u53d1\u73b0\u4e86\u4e00\u4e2a\u5de8\u725b\u7684\u4eba\u5de5\u667a\u80fd\u5b66\u4e60\u7f51\u7ad9，\u901a\u4fd7\u6613\u61c2，\u98ce\u8da3\u5e7d\u9ed8，\u5fcd\u4e0d\u4f4f\u5206\u4eab\u4e00\u4e0b\u7ed9\u5927\u5bb6\u3002\u70b9\u51fb\u94fe\u63a5\u8df3\u8f6c\u5230\u7f51\u7ad9\u4eba\u5de5\u667a\u80fd\u53ca\u7f16\u7a0b\u8bed\u8a00\u5b66\u4e60\u6559\u7a0b\u3002\u8bfb\u8005\u4eec\u53ef\u4ee5\u901a\u8fc7\u91cc\u9762\u7684\u6587\u7ae0\u8be6\u7ec6\u4e86\u89e3\u4e00\u4e0b\u4eba\u5de5\u667a\u80fd\u53ca\u5176\u7f16\u7a0b\u7b49\u6559\u7a0b\u548c\u5b66\u4e60\u65b9\u6cd5\u3002\u4e0b\u9762\u5f00\u59cb\u5bf9\u6b63\u6587\u5185\u5bb9\u7684\u4ecb\u7ecd\u3002<\/p>\n
\u6807\u7b7e：\u5927\u6a21\u578b\u3001Text2SQL\u3001NL2Chart\u3001PandasAI\u3001LangChain\u3001\u53ef\u89c6\u5316\u3001\u96f6\u4ee3\u7801 —- 1. \u80cc\u666f：\u4e1a\u52a1\u4eba\u5458\u60f3\u8981\u201c\u8bf4\u53e5\u8bdd\u5c31\u51fa\u56fe\u201d \u516c\u53f8\u6bcf\u6708\u7ecf\u8425\u590d\u76d8\u4f1a\u524d，\u4e1a\u52a1\u540c\u5b66\u628a\u9700\u6c42\u7529\u7ed9\u6570\u636e\u56e2\u961f： \u201c\u7ed9\u6211\u4e00\u5f20\u6700\u8fd1 30 \u5929 GMV \u8d8b\u52bf\u56fe，\u6309\u6e20\u9053\u62c6\u5206\u3002\u201d \u7ecf\u5178\u6d41\u7a0b： \u9700\u6c42 \u2192 \u6570\u636e\u540c\u5b66\u5199 SQL \u2192 Excel \u900f\u89c6 \u2192 \u51fa\u56fe \u2192 \u8fd4\u5de5 \u2192 3 \u5c0f\u65f6\u6ca1\u4e86\u3002 \u76ee\u6807：\u4e00\u53e5\u8bdd\u76f4\u63a5\u51fa\u56fe，\u4e14 \u4e0d\u66b4\u9732\u6570\u636e\u5e93，\u4e0d\u6cc4\u9732\u539f\u59cb\u6570\u636e\u3002 —- 2. \u6280\u672f\u9009\u578b：LLM + PandasAI + \u672c\u5730\u7f13\u5b58 \u6a21\u5757\u00a0\u00a0 \u00a0\u9009\u578b\u00a0\u00a0 \u00a0\u539f\u56e0 \u5927\u6a21\u578b\u00a0\u00a0 \u00a0gpt-3.5-turbo-16k\u00a0\u00a0 \u00a0\u4fbf\u5b9c\u3001\u4ee3\u7801\u80fd\u529b\u5f3a \u6570\u636e\u5f15\u64ce\u00a0\u00a0 \u00a0DuckDB + PandasAI\u00a0\u00a0 \u00a0\u672c\u5730 parquet，\u65e0 SQL \u66b4\u9732 \u53ef\u89c6\u5316\u00a0\u00a0 \u00a0Altair\u00a0\u00a0 \u00a0\u8bed\u6cd5\u7b80\u6d01，\u81ea\u52a8\u751f\u6210\u4ea4\u4e92\u56fe \u4ea4\u4e92\u00a0\u00a0 \u00a0Streamlit\u00a0\u00a0 \u00a010 \u5206\u949f\u51fa\u754c\u9762 —- 3. \u7cfb\u7edf\u6d41\u7a0b：\u4e00\u53e5\u8bdd \u2192 \u56fe\u8868 \u7528\u6237\u8f93\u5165\u81ea\u7136\u8bed\u8a00 \u00a0 \u00a0 \u2193 LangChain Prompt LLM \u751f\u6210 DuckDB SQL + Altair Python \u00a0 \u00a0 \u2193 Streamlit exec \u56fe\u8868\u6e32\u67d3 + \u7f13\u5b58<\/p>\n
—- 4. 100 \u884c\u6838\u5fc3\u4ee3\u7801<\/p>\n
# app.py
\nimport streamlit as st
\nfrom pandasai import SmartDataframe
\nfrom pandasai.llm import OpenAI
\nimport duckdb, os, altair as alt<\/p>\n
@st.cache_data
\ndef load_data():
\n\u00a0 \u00a0 return duckdb.sql("SELECT * FROM 'sales.parquet'").df()<\/p>\n
llm = OpenAI(api_token=os.getenv("OPENAI_API_KEY"))
\ndf = load_data()
\nsdf = SmartDataframe(df, config={"llm": llm})<\/p>\n
query = st.text_input("\u8bf4\u53e5\u8bdd\u51fa\u56fe：")
\nif query:
\n\u00a0 \u00a0 try:
\n\u00a0 \u00a0 \u00a0 \u00a0 code = sdf.chat(query)
\n\u00a0 \u00a0 \u00a0 \u00a0 exec(code) \u00a0# \u751f\u6210 alt.Chart(…)
\n\u00a0 \u00a0 except Exception as e:
\n\u00a0 \u00a0 \u00a0 \u00a0 st.error("LLM \u5199\u9519\u4ee3\u7801\u5566：" + str(e)) <\/p>\n
—- 5. Prompt \u6a21\u677f（\u8ba9 LLM \u4e0d\u4e71\u9020）<\/p>\n
PROMPT = """
\n\u4f60\u662f\u4e00\u540d\u6570\u636e\u5206\u6790\u5e08，\u53ea\u4f7f\u7528 DuckDB SQL \u548c Altair\u3002
\n\u8868\u5b57\u6bb5：date, channel, gmv, orders\u3002
\n\u8981\u6c42：
\n1. \u751f\u6210 DuckDB SQL \u8fc7\u6ee4\u6570\u636e；
\n2. \u751f\u6210 alt.Chart \u4ee3\u7801；
\n3. \u4e0d\u8981\u66b4\u9732\u771f\u5b9e\u6570\u636e\u3002
\n""" <\/p>\n
—- 6. \u5b9e\u6d4b\u6848\u4f8b \u7528\u6237\u8f93\u5165\u00a0\u00a0 \u00a0\u751f\u6210\u56fe\u8868\u00a0\u00a0 \u00a0\u8017\u65f6 \u201c\u6700\u8fd1 30 \u5929 GMV \u8d8b\u52bf\u201d\u00a0\u00a0 \u00a0\u6298\u7ebf\u56fe\u00a0\u00a0 \u00a01.2 s \u201c\u5404\u6e20\u9053 GMV \u5360\u6bd4\u201d\u00a0\u00a0 \u00a0\u73af\u5f62\u56fe\u00a0\u00a0 \u00a01.1 s \u201cGMV \u4e0e\u8ba2\u5355\u91cf\u6563\u70b9\u201d\u00a0\u00a0 \u00a0\u6c14\u6ce1\u56fe\u00a0\u00a0 \u00a01.3 s —- 7. \u6027\u80fd & \u5b89\u5168 \u2022 \u00a0\u6570\u636e\u672c\u5730：sales.parquet \u4ec5 50 MB，\u4e0d\u843d\u76d8\u3002 \u2022 \u00a0\u6743\u9650\u9694\u79bb：LLM \u62ff\u4e0d\u5230\u539f\u59cb\u6570\u636e，\u4ec5\u89c1\u805a\u5408\u7ed3\u679c\u3002 \u2022 \u00a0\u7f13\u5b58：\u76f8\u540c\u63d0\u95ee 0.2 s \u8fd4\u56de\u7ed3\u679c\u3002 —- 8. \u8e29\u5751\u65e5\u8bb0 \u5751\u00a0\u00a0 \u00a0\u89e3\u51b3 LLM \u5199 SQL \u5b57\u6bb5\u540d\u5927\u5c0f\u5199\u9519\u00a0\u00a0 \u00a0DuckDB \u4e0d\u533a\u5206\u5927\u5c0f\u5199，\u81ea\u52a8\u5bb9\u9519 \u56fe\u8868\u4e2d\u6587\u4e71\u7801\u00a0\u00a0 \u00a0Altair \u9ed8\u8ba4\u5b57\u4f53 \u2192 \u6307\u5b9a\u601d\u6e90\u9ed1\u4f53 \u591a\u6b21\u63d0\u95ee\u4e0a\u4e0b\u6587\u51b2\u7a81\u00a0\u00a0 \u00a0\u6bcf\u6b21\u91cd\u7f6e SmartDataframe \u5b9e\u4f8b —- 9. \u4e00\u952e\u90e8\u7f72<\/p>\n
pip install pandasai streamlit duckdb altair
\nstreamlit run app.py <\/p>\n
\u6d4f\u89c8\u5668\u6253\u5f00 http:\/\/localhost:8501，\u4e1a\u52a1\u540c\u5b66\u5373\u53ef\u81ea\u52a9\u51fa\u56fe\u3002 —- 10. \u7ed3\u8bed：\u8ba9 SQL \u6d88\u5931\u5728\u5bf9\u8bdd\u91cc \u5f53\u4e1a\u52a1\u4eba\u5458\u53ea\u9700\u8981\u8bf4 \u201c\u7ed9\u6211\u4e00\u5f20\u56fe\u201d， \u5f53\u6570\u636e\u56e2\u961f\u4e0d\u518d\u9700\u8981\u71ac\u591c\u5199 SQL， \u4f60\u4f1a\u53d1\u73b0 \u5927\u6a21\u578b\u771f\u6b63\u7684 killer app \u662f\u201c\u6d88\u706d\u4e2d\u95f4\u5c42\u201d\u3002 \u5982\u679c\u8fd9\u7bc7\u6587\u7ae0\u5e2e\u4f60\u8282\u7701\u4e86\u4e00\u6b21\u8fd4\u5de5，\u6b22\u8fce\u70b9\u4e2a Star； \u4e5f\u6b22\u8fce\u7559\u8a00\u804a\u804a\u4f60\u8ba9 LLM \u5199\u8fc7\u54ea\u4e9b\u201c\u79bb\u8c31\u201d\u7684 SQL！<\/p>\n","protected":false},"excerpt":{"rendered":"
\u6587\u7ae0\u6d4f\u89c8\u9605\u8bfb1.1k\u6b21\uff0c\u70b9\u8d5e45\u6b21\uff0c\u6536\u85cf7\u6b21\u3002\u6458\u8981\uff1a\u672c\u6587\u4ecb\u7ecd\u4e86\u5982\u4f55\u5229\u7528\u5927\u6a21\u578b\u6280\u672f\u5b9e\u73b0"\u81ea\u7136\u8bed\u8a00\u76f4\u63a5\u751f\u6210\u56fe\u8868"\u7684\u529f\u80fd\u3002\u901a\u8fc7\u6574\u5408LangChain\u3001PandasAI\u548cDuckDB\u7b49\u6280\u672f\uff0c\u6784\u5efa\u4e86\u4e00\u4e2a100\u884c\u4ee3\u7801\u4ee5\u5185\u7684\u89e3\u51b3\u65b9\u6848\uff0c\u4e1a\u52a1\u4eba\u5458\u53ea\u9700\u8f93\u5165\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\uff08\u5982"\u6700\u8fd130\u5929GMV\u8d8b\u52bf"\uff09\uff0c\u7cfb\u7edf\u5373\u53ef\u57281\u79d2\u5de6\u53f3\u81ea\u52a8\u751f\u6210\u76f8\u5e94\u53ef\u89c6\u5316\u56fe\u8868\u3002\u8be5\u7cfb\u7edf\u91c7\u7528\u672c\u5730parquet\u6587\u4ef6\u5b58\u50a8\u6570\u636e\uff0c\u4e0d\u66b4\u9732\u6570\u636e\u5e93\uff0c\u901a\u8fc7Prompt\u5de5\u7a0b\u7ea6\u675fLLM\u8f93\u51fa\u89c4\u8303\u7684DuckDB SQL\u548cAltair\u56fe\u8868\u4ee3\u7801\u3002<\/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":[99,187,81,101,768,100,5895,86],"topic":[],"class_list":["post-56637","post","type-post","status-publish","format-standard","hentry","category-server","tag-java","tag-javascript","tag-python","tag-spring","tag-sql","tag-100","tag-5895","tag-86"],"yoast_head":"\n