{"id":1237,"date":"2024-04-28T03:32:48","date_gmt":"2024-04-28T03:32:48","guid":{"rendered":"https:\/\/www.nicekj.com\/?p=1237"},"modified":"2024-04-28T03:34:32","modified_gmt":"2024-04-28T03:34:32","slug":"jinghuabijiwuendaxlangchainshiyonglangchaingoujianyushujuduihuadeliaotianjiqirenzhong","status":"publish","type":"post","link":"https:\/\/www.nicekj.com\/jinghuabijiwuendaxlangchainshiyonglangchaingoujianyushujuduihuadeliaotianjiqirenzhong.html","title":{"rendered":"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe x LangChain \u300a\u4f7f\u7528LangChain\u6784\u5efa\u4e0e\u6570\u636e\u5bf9\u8bdd\u7684\u804a\u5929\u673a\u5668\u4eba\u300b\uff08\u4e2d\uff09"},"content":{"rendered":"

\u6211\u4eec\u5df2\u7ecf\u5c06\u6587\u6863\u5206\u5272\u6210\u4e86\u66f4\u5c0f\u7684\u3001\u4e14\u8bed\u4e49\u660e\u786e\u7684\u5757\uff0c\u63a5\u4e0b\u6765\u8981\u505a\u7684\u5c31\u662f\u5c06\u8fd9\u4e9b\u5757\u653e\u5165\u5230\u4e00\u4e2a\u7d22\u5f15\u4e2d\uff0c\u8fd9\u6837\u5f53\u6211\u4eec\u8981\u56de\u7b54\u67d0\u4e2a\u6570\u636e\u96c6\u76f8\u5173\u7684\u95ee\u9898\u65f6\uff0c\u5c31\u80fd\u8f7b\u677e\u5730\u68c0\u7d22\u5230\u5bf9\u5e94\u7684\u5757\u3002<\/p>\n

\u8981\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\uff0c\u6211\u4eec\u9700\u8981\u7528\u5230\u4e24\u4e2a\u6280\u672f\uff1a\u5d4c\u5165(Embedding)\u548c\u5411\u91cf\u5b58\u50a8(Vector Store)\u3002<\/p>\n

\u6211\u4eec\u5728\u524d\u9762\u7684\u8bfe\u7a0b<\/a>\u91cc\u5df2\u7ecf\u4ecb\u7ecd\u8fc7\u8fd9\u4e24\u4e2a\u6280\u672f\uff0c\u8fd9\u91cc\u5148\u53ea\u505a\u7b80\u5355\u7684\u590d\u4e60\u3002<\/p>\n

\u5411\u91cf\u5b58\u50a8\u548c\u5d4c\u5165<\/h1>\n

\n

\"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n

\u5d4c\u5165\u662f\u5c06\u4e00\u6bb5\u6587\u672c\u8f6c\u5316\u4e3a\u6570\u503c\u5f62\u5f0f\u3002\u5177\u6709\u76f8\u4f3c\u5185\u5bb9\u7684\u6587\u672c\u5728\u6570\u503c\u7a7a\u95f4\u4e2d\u4f1a\u6709\u76f8\u4f3c\u7684\u5411\u91cf<\/strong>\uff0c\u8fd9\u5c31\u610f\u5473\u7740\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u6bd4\u8f83\u8fd9\u4e9b\u5411\u91cf\uff0c\u6765\u627e\u51fa\u76f8\u4f3c\u7684\u6587\u672c\u7247\u6bb5\u3002<\/p>\n

\n

\"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n

\u800c\u5411\u91cf\u5b58\u50a8\u662f\u4e00\u79cd\u6570\u636e\u5e93\uff0c\u5b83\u7528\u6765\u5b58\u50a8\u5206\u5272\u540e\u7684\u6587\u6863\u7247\u6bb5\u4ee5\u53ca\u5b83\u4eec\u5bf9\u5e94\u7684\u5d4c\u5165\uff0c\u65b9\u4fbf\u6211\u4eec\u540e\u7eed\u6839\u636e\u95ee\u9898\u67e5\u627e\u76f8\u5173\u7684\u6587\u6863\u3002<\/p>\n

\u6574\u4e2a\u8fc7\u7a0b\u5982\u4e0b\uff1a<\/p>\n

\n

\"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
    \n
  1. \u63d0\u51fa\u4e00\u4e2a\u95ee\u9898\uff0c\u5e76\u4e3a\u5b83\u751f\u6210\u4e00\u4e2a\u5d4c\u5165\uff1b<\/li>\n
  2. \u5c06\u5b83\u8ddf\u5411\u91cf\u5b58\u50a8\u91cc\u7684\u6240\u6709\u4e0d\u540c\u7684\u5411\u91cf\u8fdb\u884c\u6bd4\u8f83\uff1b<\/li>\n
  3. \u9009\u51fa\u6700\u76f8\u4f3c\u7684\u524dn\u4e2a\u7247\u6bb5\uff1b<\/li>\n
  4. \u5c06\u9009\u51fa\u7684\u7247\u6bb5\u548c\u95ee\u9898\u4e00\u8d77\u8f93\u5165\u5230LLM\u91cc\uff0c\u5f97\u5230\u4e00\u4e2a\u7b54\u6848\u3002<\/li>\n<\/ol>\n

    \u4e3a\u4e86\u5e2e\u52a9\u7406\u89e3\uff0c\u6211\u4eec\u5148\u770b\u4e00\u4e2a\u7b80\u5355\u7684\u4f8b\u5b50\uff1a<\/p>\n

    \u6b65\u9aa41\uff1a\u63d0\u4f9b\u4e00\u4e9b\u4f8b\u53e5\uff0c\u5176\u4e2d\u524d\u4e24\u53e5\u975e\u5e38\u76f8\u4f3c\uff0c\u7b2c\u4e09\u53e5\u5219\u4e0e\u524d\u4e24\u53e5\u5173\u8054\u4e0d\u5927<\/h3>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>sentence1<\/span> = \"i like dogs\"<\/span><\/span>\nsentence2<\/span> = \"i like canines\"<\/span><\/span>\nsentence3<\/span> = \"the weather is ugly outside\"<\/span><\/span>\n<\/code><\/pre>\n
    \u6b65\u9aa42\uff1a\u4f7f\u7528Embbeding\u7c7b\u4e3a\u6bcf\u4e2a\u53e5\u5b50\u751f\u6210\u4e00\u4e2a\u5d4c\u5165<\/h3>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.embeddings.openai import OpenAIEmbeddings<\/span>\nembedding<\/span> = OpenAIEmbeddings()<\/span>\n<\/span>\nembedding1<\/span> = embedding.embed_query(sentence1)<\/span>\nembedding2<\/span> = embedding.embed_query(sentence2)<\/span>\nembedding3<\/span> = embedding.embed_query(sentence3)<\/span>\n<\/code><\/pre>\n
    \u6b65\u9aa43\uff1a\u7528\u70b9\u79ef(dot product)\u6765\u8ba1\u7b97\u4e24\u4e24\u4e4b\u95f4\u7684\u5d4c\u5165\u76f8\u4f3c\u5ea6<\/h3>\n
    <\/div>
    python<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>import<\/span> numpy as<\/span> np<\/span>\n<\/span>\nnp.dot(embedding1, embedding2)<\/span>\n# 0.9631853877103518<\/span><\/span>\n<\/span>\nnp.dot(embedding1, embedding3)<\/span>\n# 0.7709997651294672<\/span><\/span>\n<\/span>\nnp.dot(embedding2, embedding3)<\/span>\n# 0.7596334120325523<\/span><\/span>\n<\/code><\/pre>\n
    \u70b9\u79ef\u7684\u503c\u8d8a\u5927\uff0c\u4ee3\u8868\u76f8\u4f3c\u5ea6\u5c31\u8d8a\u9ad8\u3002<\/p>\n
    \u518d\u6765\u770b\u4e00\u4e2a\u5b9e\u9645\u7684\u4f8b\u5b50\uff1a<\/p>\n
    \u76ee\u6807\u662f\u4e3a\u63d0\u4f9b\u7684\u6240\u6709PDF\u6587\u6863\u751f\u6210\u5d4c\u5165\uff0c\u5e76\u628a\u5b83\u4eec\u5b58\u50a8\u5728\u4e00\u4e2a\u5411\u91cf\u5b58\u50a8\u91cc\u3002<\/p>\n
    \u6b65\u9aa41\uff1a\u52a0\u8f7dPDF\u6587\u6863<\/h3>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.document_loaders import PyPDFLoader<\/span>\n<\/span>\n# \u52a0\u8f7d PDF<\/span><\/span>\nloaders<\/span> = [<\/span>\n    # \u91cd\u590d\u52a0\u8f7d\u7b2c\u4e00\u4e2a\u6587\u6863\uff0c\u6a21\u62df\u4e00\u4e9b\u810f\u6570\u636e<\/span><\/span>\n    PyPDFLoader(\"docs\/cs229_lectures\/MachineLearning-Lecture01.pdf\"<\/span>),<\/span>\n    PyPDFLoader(\"docs\/cs229_lectures\/MachineLearning-Lecture01.pdf\"<\/span>),<\/span>\n    PyPDFLoader(\"docs\/cs229_lectures\/MachineLearning-Lecture02.pdf\"<\/span>),<\/span>\n    PyPDFLoader(\"docs\/cs229_lectures\/MachineLearning-Lecture03.pdf\"<\/span>)<\/span>\n]<\/span>\ndocs<\/span> = []<\/span>\nfor loader in loaders:<\/span>\n    docs.extend(loader.load())<\/span>\n<\/code><\/pre>\n
    \u6b65\u9aa42\uff1a\u7528\u9012\u5f52\u5b57\u7b26\u6587\u672c\u5206\u5272\u5668\u6765\u628a\u6587\u6863\u5206\u6210\u5757<\/h3>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div># \u5206\u5272<\/span><\/span>\nfrom langchain.text_splitter import RecursiveCharacterTextSplitter<\/span>\ntext_splitter<\/span> = RecursiveCharacterTextSplitter(<\/span>\n    chunk_size<\/span> = 1500<\/span>,<\/span>\n    chunk_overlap<\/span> = 150<\/span><\/span>\n)<\/span>\n<\/span>\nsplits<\/span> = text_splitter.split_documents(docs)<\/span>\n<\/code><\/pre>\n
    \u6b65\u9aa43\uff1a\u4e3a\u6bcf\u4e2a\u5757\u751f\u6210\u5d4c\u5165\uff0c\u5e76\u521b\u5efaChroma\u5411\u91cf\u5b58\u50a8<\/h3>\n
    \u8fd9\u91cc\u7528\u5230\u7684\u5411\u91cf\u5b58\u50a8\u662fChroma\u3002Chroma\u662f\u4e00\u79cd\u8f7b\u91cf\u7ea7\u3001\u57fa\u4e8e\u5185\u5b58\u7684\u5411\u91cf\u5b58\u50a8\uff0c\u4f7f\u7528\u8d77\u6765\u5f88\u65b9\u4fbf\u3002<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.vectorstores import Chroma<\/span>\n<\/span>\n# \u53ef\u5148\u7528[rm -rf .\/docs\/chroma]\u79fb\u9664\u53ef\u80fd\u5b58\u5728\u7684\u65e7\u6570\u636e\u5e93\u6570\u636e<\/span><\/span>\npersist_directory<\/span> = 'docs\/chroma\/'<\/span><\/span>\n<\/span>\n# \u4f20\u5165\u4e4b\u524d\u521b\u5efa\u7684\u5206\u5272\u548c\u5d4c\u5165\uff0c\u4ee5\u53ca\u6301\u4e45\u5316\u76ee\u5f55<\/span><\/span>\nvectordb<\/span> = Chroma.from_documents(<\/span>\n    documents<\/span>=splits,<\/span>\n    embedding<\/span>=embedding,<\/span>\n    persist_directory<\/span>=persist_directory<\/span>\n)<\/span>\n<\/code><\/pre>\n
    \u6b65\u9aa44\uff1a\u7528\u76f8\u4f3c\u6027\u641c\u7d22\u65b9\u6cd5\u6765\u67e5\u627e\u6587\u6863<\/h3>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question<\/span> = \"is there an email i can ask for help\"<\/span><\/span>\n# K=3\u7528\u4e8e\u6307\u5b9a\u8fd4\u56de\u7684\u6587\u6863\u6570\u91cf<\/span><\/span>\ndocs<\/span> = vectordb.similarity_search(question,k=3<\/span>)<\/span>\n<\/code><\/pre>\n
    \u53ef\u4ee5\u6253\u5370\u6587\u6863\u7684\u957f\u5ea6\u548c\u5185\u5bb9\u6765\u68c0\u67e5\uff1a<\/p>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>len<\/span>(docs)<\/span>\n# 3<\/span><\/span>\n<\/span>\ndocs[0]<\/span>.page_content<\/span><\/span>\n<\/code><\/pre>\n
    \n
    \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
    \u6b65\u9aa45\uff1a\u6301\u4e45\u5316\u5411\u91cf\u6570\u636e\u5e93\uff0c\u4ee5\u4fbf\u4ee5\u540e\u4f7f\u7528<\/h3>\n
    <\/div>
    scss<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>vectordb.persist<\/span>()<\/span>\n<\/code><\/pre>\n
    \u63a5\u4e0b\u6765\u6211\u4eec\u5c06\u8ba8\u8bba\u4e00\u4e9b\u8fb9\u7f18\u6848\u4f8b\uff0c\u5c55\u793a\u51e0\u79cd\u53ef\u80fd\u51fa\u73b0\u5931\u8d25\u60c5\u51b5\uff1a<\/p>\n
    \u5931\u8d25\u60c5\u51b51\uff1a\u91cd\u590d\u7684\u5757\u5bfc\u81f4\u91cd\u590d\u7684\u5197\u4f59\u4fe1\u606f<\/h3>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question<\/span> = \"what did they say about matlab?\"<\/span><\/span>\ndocs<\/span> = vectordb.similarity_search(question,k=5<\/span>)<\/span>\n<\/span>\ndocs[0]<\/span><\/span>\ndocs[1]<\/span><\/span>\n<\/code><\/pre>\n
    \n
    \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
    \u5176\u4e2d\uff0cdocs[0] \u548c docs[1] \u5f97\u5230\u7684\u7ed3\u679c\u662f\u76f8\u540c\u7684\uff0c\u8fd9\u662f\u56e0\u4e3a\u6211\u4eec\u5728\u4e00\u5f00\u59cb\u5c31\u6709\u610f\u91cd\u590d\u52a0\u8f7d\u4e86\u7b2c\u4e00\u4e2a\u6587\u6863\u3002<\/p>\n
    \u8fd9\u6837\u505a\u7684\u7ed3\u679c\u662f\uff0c\u6211\u4eec\u628a\u4e24\u4e2a\u5185\u5bb9\u76f8\u540c\u7684\u5206\u5757\u90fd\u4f20\u7ed9\u4e86\u8bed\u8a00\u6a21\u578b\u3002\u800c\u7b2c\u4e8c\u4e2a\u5206\u5757\u662f\u6ca1\u6709\u4ef7\u503c\u7684\uff0c\u5982\u679c\u6362\u6210\u4e00\u4e2a\u5185\u5bb9\u4e0d\u540c\u7684\u5206\u5757\u4f1a\u66f4\u597d\uff0c\u8fd9\u6837\u81f3\u5c11\u8bed\u8a00\u6a21\u578b\u53ef\u4ee5\u4ece\u4e2d\u83b7\u53d6\u66f4\u591a\u4fe1\u606f\u3002<\/p>\n
    \u5728\u4e0b\u4e00\u8bfe\u4e2d\uff0c\u6211\u4eec\u5c06\u8ba8\u8bba\u5982\u4f55\u5728\u4fdd\u8bc1\u68c0\u7d22\u5230\u76f8\u5173\u7684\u5757\u7684\u540c\u65f6\uff0c\u4e5f\u80fd\u4fdd\u8bc1\u6bcf\u4e2a\u5757\u90fd\u662f\u552f\u4e00\u7684\u3002<\/p>\n
    \u5931\u8d25\u60c5\u51b52\uff1a\u65e0\u6cd5\u5b8c\u6574\u6355\u6349\u5230\u95ee\u9898\u4e2d\u7684\u5173\u952e\u4fe1\u606f<\/h3>\n
    \u6bd4\u5982\u4e0b\u9762\u8fd9\u4e2a\u95ee\u9898\uff0c\u201c\u7b2c\u4e09\u5802\u8bfe\u91cc\u4ed6\u4eec\u8bb2\u4e86\u4ec0\u4e48\u5173\u4e8e\u56de\u5f52\u7684\u5185\u5bb9\uff1f\u201d<\/p>\n
    <\/div>
    ini<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question<\/span> = \"what did they say about regression in the third lecture?\"<\/span><\/span>\ndocs<\/span> = vectordb.similarity_search(question,k=5<\/span>)<\/span>\n<\/code><\/pre>\n
    \u4e00\u822c\u6765\u8bf4\uff0c\u6211\u4eec\u5e94\u8be5\u80fd\u770b\u51fa\uff0c\u95ee\u9898\u7684\u63d0\u95ee\u8005\u662f\u60f3\u8981\u4ece\u7b2c\u4e09\u5802\u8bfe\u91cc\u627e\u5230\u7b54\u6848\u7684\u3002<\/p>\n
    \u4f46\u5b9e\u9645\u4e0a\uff0c\u5f53\u6211\u4eec\u904d\u5386\u6240\u6709\u6587\u6863\uff0c\u5e76\u6253\u5370\u51fa\u5143\u6570\u636e\u540e\u4f1a\u53d1\u73b0\uff0c\u7ed3\u679c\u91cc\u5b9e\u9645\u4e0a\u6df7\u5408\u4e86\u591a\u4e2a\u6587\u6863\u7684\u5185\u5bb9\u3002<\/p>\n
    <\/div>
    shell<\/span>
    \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>for doc in docs:<\/span>\n    print(doc.metadata)<\/span>\n    <\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf'<\/span>, 'page'<\/span>: 0}<\/span><\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf'<\/span>, 'page'<\/span>: 14}<\/span><\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture02.pdf'<\/span>, 'page'<\/span>: 0}<\/span><\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf'<\/span>, 'page'<\/span>: 6}<\/span><\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture01.pdf'<\/span>, 'page'<\/span>: 8}<\/span><\/span>\n<\/code><\/pre>\n
    \u8fd9\u662f\u56e0\u4e3a\uff0c\u6211\u4eec\u53ea\u662f\u57fa\u4e8e\u5d4c\u5165\u505a\u4e86\u4e00\u4e2a\u8bed\u4e49\u641c\u7d22\uff0c\u5b83\u4e3a\u6574\u4e2a\u53e5\u5b50\u751f\u6210\u4e86\u4e00\u4e2a\u5d4c\u5165\uff0c\u5e76\u4e14\u53ef\u80fd\u4f1a\u66f4\u5173\u6ce8\u4e8e\u201c\u56de\u5f52\u201d\u8fd9\u4e2a\u8bcd\u3002<\/p>\n
    \u5f53\u6211\u4eec\u67e5\u770b\u7b2c\u4e94\u4e2a\u6587\u6863\u65f6\uff0c\u5c31\u4f1a\u53d1\u73b0\u5b83\u786e\u5b9e\u63d0\u5230\u4e86\u201c\u56de\u5f52\u201d\u8fd9\u4e2a\u8bcd\u3002<\/p>\n
    \n
    \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
    \u5931\u8d25\u60c5\u51b53\uff1a\u968f\u7740\u68c0\u7d22\u6587\u6863\u6570\u91cf\u7684\u589e\u52a0\uff0c\u76f8\u5173\u6027\u9010\u6e10\u964d\u4f4e<\/h3>\n
    \u5f53\u6211\u4eec\u5c1d\u8bd5\u6539\u53d8k\u503c\uff0c\u4e5f\u5c31\u662f\u68c0\u7d22\u7684\u6587\u6863\u6570\u91cf\u65f6\uff0c\u6211\u4eec\u4f1a\u5f97\u5230\u66f4\u591a\u7684\u6587\u6863\uff0c\u4f46\u7ed3\u679c\u5217\u8868\u540e\u9762\u7684\u6587\u6863\u53ef\u80fd\u6ca1\u6709\u524d\u9762\u7684\u90a3\u4e9b\u76f8\u5173\u6027\u5f3a\u3002<\/p>\n
    \u68c0\u7d22<\/h1>\n
    \n
    \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
    \u5728\u8fd9\u4e00\u8bfe\u4e2d\uff0c\u6211\u4eec\u5c06\u6df1\u5165\u63a2\u8ba8\u300c\u68c0\u7d22\u300d\u6280\u672f\uff0c\u5e76\u4ecb\u7ecd\u4e00\u4e9b\u66f4\u5148\u8fdb\u7684\u65b9\u6cd5\u6765\u514b\u670d\u4e0a\u4e00\u8bfe\u7684\u8fb9\u7f18\u60c5\u51b5\u3002<\/p>\n
    \u68c0\u7d22\u662f\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08RAG\uff09\u6d41\u7a0b\u7684\u6838\u5fc3<\/strong>\u3002<\/p>\n
    \u89e3\u51b3\u591a\u6837\u6027\uff1a\u6700\u5927\u8fb9\u7f18\u76f8\u5173\u6027<\/h2>\n
    \n
    \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
    \u6700\u5927\u8fb9\u7f18\u76f8\u5173\u6027(MMR, Maximum Marginal Relevance)\u80cc\u540e\u7684\u7406\u5ff5\u662f\uff0c\u5982\u679c\u6211\u4eec\u603b\u662f\u9009\u62e9\u4e0e\u67e5\u8be2\u5728\u5d4c\u5165\u7a7a\u95f4\u4e2d\u6700\u76f8\u4f3c\u7684\u6587\u6863\uff0c\u6211\u4eec\u53ef\u80fd\u4f1a\u9519\u8fc7\u4e00\u4e9b\u591a\u5143\u5316\u7684\u4fe1\u606f<\/strong>\u3002<\/p>\n
    MMR\u53ef\u4ee5\u5e2e\u52a9\u6211\u4eec\u9009\u62e9\u4e00\u4e2a\u66f4\u591a\u6837\u5316\u7684\u6587\u6863\u96c6\u5408\u3002<\/p>\n
    MMR\u5728\u4fdd\u6301\u67e5\u8be2\u76f8\u5173\u6027\u7684\u540c\u65f6\uff0c\u5c3d\u91cf\u589e\u52a0\u7ed3\u679c\u4e4b\u95f4\u7684\u591a\u6837\u6027<\/strong>\uff0c\u5b83\u7684\u505a\u6cd5\u662f\uff1a<\/p>\n
    \n
    \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
      \n
    1. \u9996\u5148\u53d1\u9001\u4e00\u4e2a\u67e5\u8be2\uff0c\u5f97\u5230\u4e00\u7ec4\u56de\u7b54\uff1b<\/li>\n
    2. \u7528”fetch_k”\u53c2\u6570\u6307\u5b9a\u6211\u4eec\u60f3\u8981\u83b7\u53d6\u7684\u54cd\u5e94\u6570\u91cf\uff0c\u8fd9\u5b8c\u5168\u57fa\u4e8e\u8bed\u4e49\u76f8\u4f3c\u6027\uff1b<\/li>\n
    3. \u7136\u540e\uff0c\u9488\u5bf9\u8fd9\u4e2a\u8f83\u5c0f\u7684\u6587\u6863\u96c6\u5408\uff0c\u4ece\u591a\u6837\u6027\u65b9\u9762\u8fdb\u884c\u4f18\u5316\uff1b<\/li>\n
    4. \u6700\u540e\u4ece\u8fd9\u7ec4\u6587\u6863\u4e2d\uff0c\u9009\u62e9”k”\u4e2a\u54cd\u5e94\u8fd4\u56de\u7ed9\u7528\u6237\u3002<\/li>\n<\/ol>\n
      \u6211\u4eec\u7528\u4e00\u4e2a\u7b80\u5355\u7684\u4f8b\u5b50\u6765\u5e2e\u52a9\u7406\u89e3\uff1a<\/p>\n
      \u76ee\u6807\u662f\u67e5\u8be2\u6709\u6307\u5b9a\u7279\u5f81\u7684\u8611\u83c7\u4fe1\u606f\u3002<\/p>\n
      \u6b65\u9aa41\uff1a\u521b\u5efaChroma\u5411\u91cf\u5b58\u50a8<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.vectorstores import Chroma<\/span>\nfrom langchain.embeddings.openai import OpenAIEmbeddings<\/span>\npersist_directory<\/span> = 'docs\/chroma\/'<\/span><\/span>\n<\/span>\nembedding<\/span> = OpenAIEmbeddings()<\/span>\nvectordb<\/span> = Chroma(<\/span>\n    persist_directory<\/span>=persist_directory,<\/span>\n    embedding_function<\/span>=embedding<\/span>\n)<\/span>\n<\/code><\/pre>\n
      \u6b65\u9aa42\uff1a\u7528\u5c11\u91cf\u4fe1\u606f\u521b\u5efa\u4e00\u4e2a\u5c0f\u578b\u6570\u636e\u5e93<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div># \u9e45\u818f\u83cc\u6709\u4e00\u4e2a\u5de8\u5927\u800c\u96c4\u4f1f\u7684\u5b50\u5b9e\u4f53(\u5730\u4e0a\u90e8\u5206)\u3002<\/span><\/span>\n# \u6709\u5927\u5b50\u5b9e\u4f53\u7684\u8611\u83c7\u662f\u9e45\u818f\u83cc\u3002\u6709\u4e9b\u54c1\u79cd\u662f\u5168\u767d\u8272\u7684\u3002<\/span><\/span>\n# \u9e45\u818f\u83cc\uff0c\u53c8\u53eb\u6b7b\u4ea1\u5e3d\uff0c\u662f\u6240\u6709\u5df2\u77e5\u8611\u83c7\u4e2d\u6bd2\u6027\u6700\u5f3a\u7684\u4e00\u79cd\u3002<\/span><\/span>\ntexts<\/span> = [<\/span>\n    \"\"\"The Amanita phalloides has a large and imposing epigeous (aboveground) fruiting body (basidiocarp).\"\"\"<\/span>,<\/span>\n    \"\"\"A mushroom with a large fruiting body is the Amanita phalloides. Some varieties are all-white.\"\"\"<\/span>,<\/span>\n    \"\"\"A. phalloides, a.k.a Death Cap, is one of the most poisonous of all known mushrooms.\"\"\"<\/span>,<\/span>\n]<\/span>\n<\/span>\nsmalldb<\/span> = Chroma.from_texts(texts, embedding=embedding)<\/span>\n<\/code><\/pre>\n
      \u6b65\u9aa43\uff1a\u8fdb\u884c\u76f8\u4f3c\u6027\u641c\u7d22<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div># \u544a\u8bc9\u6211\u6709\u5173\u5e26\u6709\u5927\u5b50\u5b9e\u4f53\u7684\u5168\u767d\u8611\u83c7\u7684\u4fe1\u606f<\/span><\/span>\nquestion<\/span> = \"Tell me about all-white mushrooms with large fruiting bodies\"<\/span><\/span>\n<\/span>\nsmalldb.similarity_search(question, k<\/span>=2<\/span>)<\/span>\n# [Document(page_content='A mushroom with a large fruiting body is the Amanita phalloides. Some varieties are all-white.', metadata={}),<\/span><\/span>\n# Document(page_content='The Amanita phalloides has a large and imposing epigeous (aboveground) fruiting body (basidiocarp).', metadata={})]<\/span><\/span>\n<\/code><\/pre>\n
      \u53ef\u4ee5\u770b\u5230\uff0c\u5b83\u6839\u636ek\u503c\u8fd4\u56de\u4e86\u4e24\u4e2a\u6700\u76f8\u5173\u7684\u6587\u6863\uff0c\u4f46\u6ca1\u6709\u63d0\u5230\u5b83\u4eec\u6709\u6bd2\u7684\u4e8b\u5b9e\u3002<\/p>\n
      \u6b65\u9aa44\uff1a\u8fdb\u884cMMR\u641c\u7d22<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>smalldb.max_marginal_relevance_search(question,k<\/span>=2<\/span>, fetch_k=3<\/span>)<\/span>\n# [Document(page_content='A mushroom with a large fruiting body is the Amanita phalloides. Some varieties are all-white.', metadata={}),<\/span><\/span>\n# Document(page_content='A. phalloides, a.k.a Death Cap, is one of the most poisonous of all known mushrooms.', metadata={})]<\/span><\/span>\n<\/code><\/pre>\n
      \u8fd9\u91cc\u6211\u4eec\u4f20\u5165\u4e86”k=2″\uff0c\u8868\u793a\u4ecd\u7136\u60f3\u8fd4\u56de\u4e24\u4e2a\u6587\u6863\uff0c\u4f46\u6211\u4eec\u8bbe\u7f6e\u4e86”fetch_k=3″\uff0c\u8868\u793a\u60f3\u83b7\u53d6\u4e09\u4e2a\u6587\u6863\u3002\u7136\u540e\u6211\u4eec\u5c31\u53ef\u4ee5\u770b\u5230\uff0c\u8fd4\u56de\u7684\u6587\u6863\u4e2d\u5305\u542b\u4e86\u5b83\u4eec\u6709\u6bd2\u7684\u4e8b\u5b9e\u3002<\/p>\n
      \u73b0\u5728\u6211\u4eec\u8bd5\u7740\u7528\u8fd9\u4e2a\u65b9\u6cd5\u6765\u5904\u7406\u4e0a\u4e00\u8282\u8bfe\u4e2d\u7684\u5931\u8d25\u60c5\u51b51\uff1a<\/p>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question<\/span> = \"what did they say about matlab?\"<\/span><\/span>\ndocs_mmr<\/span> = vectordb.max_marginal_relevance_search(question,k=3<\/span>)<\/span>\n<\/span>\ndocs_mmr[0]<\/span>.page_content[:100]<\/span><\/span>\n# 'those homeworks will be done in either MATLA B or in Octave, which is sort of \u2014 I nknow some people '<\/span><\/span>\n<\/span>\ndocs_mmr[1]<\/span>.page_content[:100]<\/span><\/span>\n# 'algorithm then? So what\u2019s different? How come  I was making all that noise earlier about nleast squa'<\/span><\/span>\n<\/code><\/pre>\n
      \u53ef\u4ee5\u770b\u5230\uff0c\u7b2c\u4e00\u4e2a\u6587\u6863\u8ddf\u4e4b\u524d\u4e00\u6837\uff0c\u56e0\u4e3a\u5b83\u6700\u76f8\u5173\u3002\u800c\u7b2c\u4e8c\u4e2a\u6587\u6863\u8fd9\u6b21\u5c31\u4e0d\u540c\u4e86\uff0c\u8fd9\u8bf4\u660eMMR\u8ba9\u56de\u7b54\u4e2d\u589e\u52a0\u4e86\u4e00\u4e9b\u591a\u6837\u6027\u3002<\/p>\n
      \u89e3\u51b3\u7279\u6b8a\u6027\uff1a\u4f7f\u7528\u81ea\u67e5\u8be2\u68c0\u7d22\u5668\u5904\u7406\u5143\u6570\u636e<\/h2>\n
      \n
      \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
      \u81ea\u67e5\u8be2\u4f7f\u7528\u8bed\u8a00\u6a21\u578b\u5c06\u539f\u59cb\u95ee\u9898\u5206\u5272\u4e3a\u4e24\u4e2a\u72ec\u7acb\u7684\u90e8\u5206\uff0c\u4e00\u4e2a\u8fc7\u6ee4\u5668\u548c\u4e00\u4e2a\u641c\u7d22\u9879\u3002<\/p>\n
      \u641c\u7d22\u9879\u5c31\u662f\u6211\u4eec\u5728\u8bed\u4e49\u4e0a\u60f3\u8981\u67e5\u627e\u7684\u95ee\u9898\u5185\u5bb9\u3002<\/p>\n
      \u8fc7\u6ee4\u5668\u5219\u662f\u5305\u542b\u6211\u4eec\u60f3\u8981\u8fc7\u6ee4\u7684\u5143\u6570\u636e\u3002<\/p>\n
      \u6bd4\u5982\uff0c\u201c1980\u5e74\u5236\u4f5c\u7684\u5173\u4e8e\u5916\u661f\u4eba\u7684\u7535\u5f71\u6709\u54ea\u4e9b\u201d\uff0c\u8bed\u4e49\u90e8\u5206\u5c31\u662f\u201c\u5173\u4e8e\u5916\u661f\u4eba\u7684\u7535\u5f71\u201d\uff0c\u5143\u6570\u636e\u90e8\u5206\u5219\u662f\u201c\u7535\u5f71\u5e74\u4efd\u5e94\u4e3a1980\u5e74\u201d\u3002<\/p>\n
      \u6211\u4eec\u5148\u624b\u52a8\u6307\u5b9a\u4e00\u4e2a\u5143\u6570\u636e\u8fc7\u6ee4\u5668\u6765\u9a8c\u8bc1\u5b83\u7684\u6548\u679c\u3002<\/p>\n
      \u76ee\u6807\u662f\u5904\u7406\u4e0a\u4e00\u8282\u8bfe\u7684\u5931\u8d25\u60c5\u51b52\uff1a<\/p>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question<\/span> = \"what did they say about regression in the third lecture?\"<\/span><\/span>\n# \u6307\u5b9a\u6e90\u4e3a\u7b2c\u4e09\u5802\u8bfe\u7684PDF\u6587\u6863<\/span><\/span>\ndocs<\/span> = vectordb.similarity_search(<\/span>\n    question,<\/span>\n    k<\/span>=3<\/span>,<\/span>\n    filter<\/span>={\"source\"<\/span>:\"docs\/cs229_lectures\/MachineLearning-Lecture03.pdf\"<\/span>}<\/span>\n)<\/span>\n<\/span>\nfor d in docs:<\/span>\n    print(d.metadata)<\/span>\n# {'source': 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf', 'page': 0}<\/span><\/span>\n# {'source': 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf', 'page': 14}<\/span><\/span>\n# {'source': 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf', 'page': 4} <\/span><\/span>\n<\/code><\/pre>\n
      \u53ef\u4ee5\u770b\u5230\uff0c\u73b0\u5728\u68c0\u7d22\u5230\u7684\u6587\u6863\u90fd\u6765\u81ea\u90a3\u4e00\u5802\u8bfe\u4e86\u3002<\/p>\n
      \u6211\u4eec\u8fd8\u53ef\u4ee5\u4f7f\u7528SelfQueryRetriever<\/code>\uff0c\u4ece\u95ee\u9898\u672c\u8eab\u63a8\u65ad\u51fa\u5143\u6570\u636e\u3002<\/p>\n
      \u6b65\u9aa41\uff1a\u63d0\u4f9b\u5143\u6570\u636e\u5b57\u6bb5\u4fe1\u606f<\/h3>\n
      <\/div>
      python<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from<\/span> langchain.llms import<\/span> OpenAI<\/span>\nfrom<\/span> langchain.retrievers.self_query.base import<\/span> SelfQueryRetriever<\/span>\nfrom<\/span> langchain.chains.query_constructor.base import<\/span> AttributeInfo<\/span>\n<\/span>\nmetadata_field_info = [<\/span>\n    AttributeInfo(<\/span>\n        name=\"source\"<\/span>,<\/span>\n        description=\"The lecture the chunk is from, should be one of `docs\/cs229_lectures\/MachineLearning-Lecture01.pdf`, `docs\/cs229_lectures\/MachineLearning-Lecture02.pdf`, or `docs\/cs229_lectures\/MachineLearning-Lecture03.pdf`\"<\/span>,<\/span>\n        type<\/span>=\"string\"<\/span>,<\/span>\n    ),<\/span>\n    AttributeInfo(<\/span>\n        name=\"page\"<\/span>,<\/span>\n        description=\"The page from the lecture\"<\/span>,<\/span>\n        type<\/span>=\"integer\"<\/span>,<\/span>\n    ),<\/span>\n]<\/span>\n<\/code><\/pre>\n
      \u8fd9\u4e2a\u4f8b\u5b50\u4e2d\u7684\u5143\u6570\u636e\u53ea\u6709\u4e24\u4e2a\u5b57\u6bb5\uff0c\u6e90\uff08source\uff09\u548c\u9875\uff08page\uff09\u3002\u6211\u4eec\u9700\u8981\u586b\u5199\u6bcf\u4e2a\u5b57\u6bb5\u7684\u540d\u79f0\u3001\u63cf\u8ff0\u548c\u7c7b\u578b\u3002\u8fd9\u4e9b\u4fe1\u606f\u4f1a\u88ab\u4f20\u7ed9\u8bed\u8a00\u6a21\u578b\uff0c\u6240\u4ee5\u9700\u8981\u5c3d\u53ef\u80fd\u63cf\u8ff0\u5f97\u6e05\u695a\u3002<\/p>\n
      \u6b65\u9aa42\uff1a\u521d\u59cb\u5316\u81ea\u67e5\u8be2\u68c0\u7d22\u5668<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div># \u6307\u5b9a\u6587\u6863\u5b9e\u9645\u5185\u5bb9\u7684\u4fe1\u606f<\/span><\/span>\ndocument_content_description<\/span> = \"Lecture notes\"<\/span><\/span>\nllm<\/span> = OpenAI(temperature=0<\/span>)<\/span>\nretriever<\/span> = SelfQueryRetriever.from_llm(<\/span>\n    llm,<\/span>\n    vectordb,<\/span>\n    document_content_description,<\/span>\n    metadata_field_info,<\/span>\n    verbose<\/span>=True<\/span><\/span>\n)<\/span>\n<\/code><\/pre>\n
      \u6b65\u9aa43\uff1a\u8fd0\u884c\u81ea\u67e5\u8be2\u68c0\u7d22\u5668\u641c\u7d22\u95ee\u9898<\/h3>\n
      <\/div>
      shell<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question = \"what did they say about regression in the third lecture?\"<\/span>\n<\/span>\ndocs = retriever.get_relevant_documents(question)<\/span>\n<\/span>\nfor d in docs:<\/span>\n    print(d.metadata)<\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf'<\/span>, 'page'<\/span>: 0}<\/span><\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf'<\/span>, 'page'<\/span>: 14}<\/span><\/span>\n# <\/span>{'source'<\/span>: 'docs\/cs229_lectures\/MachineLearning-Lecture03.pdf'<\/span>, 'page'<\/span>: 4}<\/span> <\/span>\n<\/code><\/pre>\n
      \n
      \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
      \u53ef\u4ee5\u770b\u5230\uff0c\u8bed\u4e49\u90e8\u5206\u8868\u660e\u8fd9\u662f\u4e00\u4e2a\u5173\u4e8e\u56de\u5f52\u7684\u67e5\u8be2\u3002\u8fc7\u6ee4\u5668\u90e8\u5206\u8868\u660e\u6211\u4eec\u53ea\u60f3\u770b\u90a3\u4e9bsource\u503c\u4e3a\u6307\u5b9a\u503c\u7684\u6587\u6863\u3002<\/p>\n
      \u800c\u4ece\u6253\u5370\u51fa\u7684\u5143\u6570\u636e\u770b\uff0c\u5b83\u4eec\u90fd\u6765\u81ea\u6307\u5b9a\u7684\u90a3\u4e00\u5802\u8bfe\uff0c\u8bf4\u660e\u81ea\u67e5\u8be2\u68c0\u7d22\u5668\u786e\u5b9e\u53ef\u4ee5\u7528\u6765\u7cbe\u786e\u5730\u8fdb\u884c\u5143\u6570\u636e\u8fc7\u6ee4\u3002<\/p>\n
      \u89e3\u51b3\u76f8\u5173\u6027\uff1a\u4f7f\u7528\u4e0a\u4e0b\u6587\u538b\u7f29\u63d0\u53d6\u51fa\u4e0e\u67e5\u8be2\u6700\u76f8\u5173\u7684\u90e8\u5206<\/h2>\n
      \n
      \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
      \u63d0\u9ad8\u68c0\u7d22\u6587\u6863\u8d28\u91cf\u7684\u53e6\u4e00\u79cd\u65b9\u6cd5\u662f\u538b\u7f29\u3002<\/p>\n
      \u5f53\u4f60\u63d0\u51fa\u4e00\u4e2a\u95ee\u9898\u65f6\uff0c\u4f60\u4f1a\u5f97\u5230\u6574\u4e2a\u5b58\u50a8\u7684\u6587\u6863\uff0c\u4f46\u53ef\u80fd\u53ea\u6709\u5176\u4e2d\u4e00\u5c0f\u90e8\u5206\u662f\u8ddf\u95ee\u9898\u76f8\u5173\u7684\u3002<\/p>\n
      \u4e5f\u5c31\u662f\u8bf4\uff0c\u4e0e\u67e5\u8be2\u6700\u76f8\u5173\u7684\u4fe1\u606f\u53ef\u80fd\u88ab\u9690\u85cf\u5728\u5305\u542b\u5927\u91cf\u65e0\u5173\u6587\u672c\u7684\u6587\u6863\u91cc\u3002<\/p>\n
      \u4e0a\u4e0b\u6587\u538b\u7f29\u5c31\u662f\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u7684\u3002<\/p>\n
      \u901a\u8fc7\u538b\u7f29\uff0c\u4f60\u53ef\u4ee5\u5148\u8ba9\u8bed\u8a00\u6a21\u578b\u63d0\u53d6\u51fa\u6700\u76f8\u5173\u7684\u7247\u6bb5\uff0c\u7136\u540e\u53ea\u628a\u6700\u76f8\u5173\u7684\u7247\u6bb5\u4f20\u7ed9\u6700\u7ec8\u7684\u8bed\u8a00\u6a21\u578b\u8c03\u7528<\/strong>\u3002<\/p>\n
      \u8fd9\u4f1a\u589e\u52a0\u8bed\u8a00\u6a21\u578b\u8c03\u7528\u7684\u6210\u672c\uff0c\u4f46\u4e5f\u4f1a\u8ba9\u6700\u7ec8\u7684\u7b54\u6848\u66f4\u96c6\u4e2d\u5728\u6700\u91cd\u8981\u7684\u5185\u5bb9\u4e0a\uff0c\u8fd9\u9700\u8981\u6211\u4eec\u81ea\u5df1\u6743\u8861\u3002<\/p>\n
      \u6b65\u9aa41\uff1a\u521b\u5efa\u4e0a\u4e0b\u6587\u538b\u7f29\u68c0\u7d22\u5668<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>from langchain.retrievers import ContextualCompressionRetriever<\/span>\nfrom langchain.retrievers.document_compressors import LLMChainExtractor<\/span>\n<\/span>\n# \u5305\u88c5\u6211\u4eec\u7684\u5411\u91cf\u5b58\u50a8<\/span><\/span>\nllm<\/span> = OpenAI(temperature=0<\/span>)<\/span>\ncompressor<\/span> = LLMChainExtractor.from_llm(llm)<\/span>\n<\/span>\ncompression_retriever<\/span> = ContextualCompressionRetriever(<\/span>\n    base_compressor<\/span>=compressor,<\/span>\n    base_retriever<\/span>=vectordb.as_retriever()<\/span>\n)<\/span>\n<\/code><\/pre>\n
      \u6b65\u9aa42\uff1a\u63d0\u51fa\u95ee\u9898\uff0c\u68c0\u7d22\u538b\u7f29\u540e\u7684\u76f8\u5173\u6587\u6863<\/h3>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>question<\/span> = \"what did they say about matlab?\"<\/span><\/span>\ncompressed_docs<\/span> = compression_retriever.get_relevant_documents(question)<\/span>\npretty_print_docs(compressed_docs)<\/span>\n<\/code><\/pre>\n
      \n
      \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
      \u4ece\u538b\u7f29\u540e\u7684\u6587\u6863\u6211\u4eec\u53ef\u4ee5\u770b\u5230\uff0c \u4e00\uff0c\u5b83\u4eec\u6bd4\u666e\u901a\u7684\u6587\u6863\u77ed\u5f97\u591a\uff1b \u4e8c\uff0c\u4ecd\u7136\u6709\u91cd\u590d\u7684\u5185\u5bb9\uff0c\u8fd9\u662f\u56e0\u4e3a\u5e95\u5c42\u6211\u4eec\u8fd8\u662f\u7528\u7684\u8bed\u4e49\u641c\u7d22\u7b97\u6cd5\u3002<\/p>\n
      \u4e3a\u4e86\u89e3\u51b3\u5185\u5bb9\u91cd\u590d\u7684\u95ee\u9898\uff0c\u6211\u4eec\u53ef\u4ee5\u5728\u521b\u5efa\u68c0\u7d22\u5668\u65f6\uff0c\u7ed3\u5408\u524d\u9762\u7684\u5185\u5bb9\uff0c\u628a\u641c\u7d22\u7c7b\u578b\u8bbe\u7f6e\u4e3aMMR\u3002<\/p>\n
      <\/div>
      ini<\/span>
      \u590d\u5236\u4ee3\u7801<\/div><\/div><\/div>compression_retriever<\/span> = ContextualCompressionRetriever(<\/span>\n    base_compressor<\/span>=compressor,<\/span>\n    base_retriever<\/span>=vectordb.as_retriever(search_type = \"mmr\"<\/span>)<\/span>\n)<\/span>\n<\/span>\nquestion<\/span> = \"what did they say about matlab?\"<\/span><\/span>\ncompressed_docs<\/span> = compression_retriever.get_relevant_documents(question)<\/span>\npretty_print_docs(compressed_docs)<\/span>\n<\/code><\/pre>\n
      \u91cd\u65b0\u8fd0\u884c\u4e4b\u540e\uff0c\u6211\u4eec\u5f97\u5230\u7684\u5c31\u662f\u4e00\u4e2a\u6ca1\u6709\u4efb\u4f55\u91cd\u590d\u4fe1\u606f\u7684\u8fc7\u6ee4\u540e\u7684\u7ed3\u679c\u96c6\u4e86\u3002<\/p>\n
      \n
      \"\u7cbe\u534e\u7b14\u8bb0\uff1a\u5434\u6069\u8fbe<\/figure>\n<\/p>\n
      \u4e0b\u4e00\u6b65\uff0c\u6211\u4eec\u5c06\u8ba8\u8bba\u5982\u4f55\u4f7f\u7528\u8fd9\u4e9b\u68c0\u7d22\u5230\u7684\u6587\u6863\u6765\u56de\u7b54\u7528\u6237\u7684\u95ee\u9898\u3002<\/p>","protected":false},"excerpt":{"rendered":"
      \u6211\u4eec\u5df2\u7ecf\u5c06\u6587\u6863\u5206\u5272\u6210\u4e86\u66f4\u5c0f\u7684\u3001\u4e14\u8bed\u4e49\u660e\u786e\u7684\u5757\uff0c\u63a5\u4e0b\u6765\u8981\u505a\u7684\u5c31\u662f\u5c06\u8fd9\u4e9b\u5757\u653e\u5165\u5230\u4e00\u4e2a\u7d22\u5f15\u4e2d\uff0c\u8fd9\u6837\u5f53\u6211\u4eec\u8981\u56de\u7b54\u67d0\u4e2a\u6570\u636e\u96c6\u76f8\u5173\u7684\u95ee\u9898\u65f6\uff0c\u5c31\u80fd\u8f7b\u677e\u5730\u68c0\u7d22\u5230\u5bf9\u5e94\u7684\u5757\u3002 \u8981\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\uff0c\u6211\u4eec\u9700\u8981\u7528\u5230\u4e24\u4e2a\u6280\u672f\uff1a\u5d4c\u5165(<\/p>\n","protected":false},"author":1,"featured_media":2771,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"rank_math_title":"\u590d\u5236 | \u4ee3\u7801 | \u68c0\u7d22 - \u5948\u65af\u79d1\u6280\u793e\u533a","rank_math_description":"\u5948\u65af\u79d1\u6280\u793e\u533a\u4e13\u6ce8\u4e8e\u590d\u5236\u3001\u4ee3\u7801\u3001\u68c0\u7d22\u3001docscslecturesMachineLearningLecturepdf\u3001source\uff0c\u63d0\u4f9b\u4e13\u4e1a\u7684\u5206\u4eab\u548c\u8ba8\u8bba\u5e73\u53f0\u3002","rank_math_focus_keyword":"\u590d\u5236,\u4ee3\u7801,\u68c0\u7d22,docscslecturesMachineLearningLecturepdf,source","views":"40","footnotes":""},"categories":[3],"tags":[126,127,128,129,136],"collection":[],"class_list":["post-1237","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fenlei2","tag-gpt","tag-ai","tag-128","tag-129","tag-136"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1237","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/comments?post=1237"}],"version-history":[{"count":0,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/posts\/1237\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media\/2771"}],"wp:attachment":[{"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/media?parent=1237"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/categories?post=1237"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/tags?post=1237"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.nicekj.com\/nicekj2024\/wp\/v2\/collection?post=1237"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}