核心判断：本文借助 DB-GPT 构建了一个名为“Chat Data with OceanBase”的应用。该方案充分利用了 DB-GPT 强大的工作流编排能力，使 OceanBase 数据库的使用体验大幅提升。更值得注意的是，操作流程十分简洁，对于所有 OceanBase 社区版的用户而言，这无疑是一次值得亲自实践的探索。
背景
首先简要介绍两位主角。DB-GPT 被称为“AI 原生数据应用开发框架”，可视为构建大模型应用的基础设施。它凭借多模型管理、Text2SQL 效果优化、RAG 框架、多智能体协作以及智能体工作流编排等能力，大幅降低了围绕数据库构建大模型应用的复杂度。
另一方面，OceanBase 自 4.3.3 版本起，正式支持向量数据类型的存储与检索。经过适配，它可无缝作为 DB-GPT 的向量数据库，统一承载结构化数据和向量数据的存取任务。
以往 AI Workshop 实验通常采用“OceanBase for AI”思路，即把 OceanBase 作为 AI 应用的后端数据库。而本次实验则尝试反向探索“AI for OceanBase”的潜力，检验 AI 技术究竟能多大程度提升数据库的用户体验。
实验介绍
以下是我个人对本次实验的理解，若有偏差，欢迎各位指正与探讨。
这里所说的“Chat Data”应用，其核心功能是允许用户通过自然语言向大模型描述所需数据，大模型随即自动生成相应的 SQL 查询语句。该应用不仅能直接在数据库中执行 SQL 并返回结果，还能将结果数据转化为可视化图表。具体实验步骤可参考《OceanBase X DB-GPT 实验教程》（附录1）。
为了让读者更直观地理解实验架构，尤其是 OceanBase 在此方案中的角色，我绘制了以下示意图。虽然画工略粗糙，但大致能说明问题。
图中展示的 OceanBase 租户里，主要分了三大类库：
- 用于存储用户业务数据的数据库（图中 User Data 库）；
- 专门存储向量数据的数据库（图中 Vector 库）；
- 以及其他一些数据库（图中 Others 库）。
Chat Data 应用所服务的数据库正是 OceanBase 中的 User Data 库。
同时，Chat Data 应用需要根据用户的自然语言请求，从数据库中检索相关的元数据（如表名、列名），并进行相似度匹配。这一过程依赖向量数据库提供支持，而该向量服务同样由 OceanBase 承担，即图中的 Vector 库。
换言之，此次 OceanBase 与 DB-GPT 的整合，让用户无需额外搭建独立的向量数据库。OceanBase 能够凭借自身向量能力，实现“自给自足”，为其存储的用户数据提供智能检索服务。
上面这张图信息量较大，我们可以将其拆分为两部分来理解。图中数据库的左侧部分展示了向量的生成流程。
如上图所示，DB-GPT 在搭建 Chat Data 应用时会执行以下三个步骤：
- 首先，创建一个指向 User Data 库（实际名称为
dbgpt_test_db）的连接。建立连接的同时，DB-GPT 会立即提取该库中所有用户数据的元信息（包括表名、列名、字段类型等）。 - 接着，将这些元信息文本通过模型转换为向量形式。
- 最后，将生成的向量存入 Vector 库中名为
dbgpt_test_db_profile的表内。
每当在 DB-GPT 上新建一个 User Data 库连接，系统便会在 OceanBase 的 Vector 库中自动创建一张名为 数据库名_profile 的表。该表包含一个 document 列，用于存储元数据的文本；以及一个 embedding 列，用于存储由 document 列转换而来的 1024 维向量。
obclient [dbgpt_vec_db]> desc dbgpt_test_db_profile;
+-----------+---------------+------+-----+---------+-------+
| Field | Type | Null | Key | Default | Extra |
+-----------+---------------+------+-----+---------+-------+
| id | varchar(4096) | NO | PRI | NULL | |
| embedding | VECTOR(1024) | YES | | NULL | |
| document | longtext | YES | | NULL | |
| metadata | json | YES | | NULL | |
+-----------+---------------+------+-----+---------+-------+
4 rows in set (0.003 sec)
obclient [dbgpt_vec_db]> select document from dbgpt_test_db_profile;
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| document |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ |
| plant_and_animal_table(id, name, name_embedding) |
| lineitem(l_orderkey, l_partkey, l_suppkey, l_linenumber, l_quantity, l_extendedprice, l_discount, l_tax, l_returnflag, l_linestatus, l_shipdate, l_commitdate, l_receiptdate, l_shipinstruct, l_shipmode, l_comment) |
| t1(c1, c2, c3, c4) |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
3 rows in set (0.003 sec)
obclient [dbgpt_vec_db]> select embedding from dbgpt_test_db_profile limit 1G
*************************** 1. row ***************************
embedding: [-0.0513874,0.000697833,-0.102775,-0.0497347,-0.0157103,-0.0107329,0.0206203,0.0187754,0.0106561,0.0389345,0.0656468,0.0567683,-0.0221961,-0.00616399,0.0408178,-0.0155181,0.0164501,-0.0248097,-0.0175071,-0.0595356,-0.0593819,0.0459297,-0.0351679,0.0226189,0.0173149,0.0114632,-0.0527711,-0.0593819,-0.00333663,-0.0568452,0.00423985,-0.00812417,0.044546,-0.0516949,-0.0157487,-0.0108675,0.0104255,0.0232531,-0.10585,0.0384541,0.016969,-0.0193808,0.0441617,-0.0598815,0.000535986,-0.10093,0.0264816,-0.044546,0.0286148,0.0269236,-0.0222346,0.0150088,-0.00247665,-0.0367245,0.0186505,0.0361096,-0.0278653,-0.021389,-0.0364939,0.0295756,-0.0236374,-0.0082683,0.0217349,0.047813,0.0134618,0.0860173,0.039857,-0.0203897,-0.027577,-0.0603428,-0.0315935,0.015816,-0.0129622,-0.00887846,-0.0944729,-0.00964235,0.0505419,0.0249442,-0.0161523,-0.0521561,0.0411638,0.00171035,0.034649,-0.0128853,0.0461603,0.0481205,-0.0361672,0.0744484,-0.0117899,0.0119821,-0.0108579,-0.0530401,0.0298639,-0.0359366,-0.0123664,-0.00706241,-0.026251,0.0128469,0.0286532,0.0503881,0.00205026,0.0502344,0.0279229,-0.0545391,0.0137693,0.0219271,-0.00136083,0.00645706,0.0418171,-0.00944057,0.0251556,0.0504266,-0.0140383,-0.00412935,-0.00361288,-0.0144707,-0.0108867,-0.00345914,-0.0216388,0.0337074,0.00217878,0.00368494,-0.00970481,0.0149223,-0.00225445,-0.0133849,0.0128372,-0.00510223,-0.043662,0.0237528,0.050734,0.0173341,-0.0025343,-0.0108867,-0.0475824,-0.00122331,0.0273464,0.0113479,-0.0102141,-0.0526942,0.0372434,0.00913789,-0.0231954,0.00786954,-0.0148166,-0.00404527,0.0190925,0.0080473,-0.019131,-0.0298255,0.0324198,-0.0273656,0.0558459,-0.00638019,-0.0267315,-0.0659158,-0.0124913,0.0436236,0.00781669,0.0108771,0.0349181,0.0212737,-0.0210239,-0.00966637,-0.0323429,0.00598143,0.0335152,-0.0170363,-0.0249827,-0.0180644,0.0360711,-0.0171227,-0.02996,-0.027308,0.0713351,0.0340917,0.107387,0.0300561,0.00941654,-0.0621492,-0.0364747,0.00725458,-0.0430855,-0.0232339,-0.0204666,0.0565761,0.0278076,-0.010166,0.03438,0.039857,-0.0137309,-0.013702,-0.0268275,-0.0263279,0.0348604,-0.00645706,0.0118572,0.0200822,-0.016969,-0.0235414,-0.0817894,0.0167768,0.00241179,-0.0265969,0.00168393,0.0159216,-0.00533284,-0.00671169,0.0160754,-0.0370512,-0.0146149,-0.00293787,-0.0108579,-0.0258282,-0.0520408,-0.0126162,-0.040741,0.0262895,-0.00633695,-0.0158832,-0.00205747,0.0273849,-0.0244254,0.0549619,0.042509,-0.0193231,0.0481973,0.0483895,-0.00582288,-0.00549619,-0.0254823,-0.0374356,-0.0242716,-0.00371617,0.0390691,0.0192463,-0.0073795,-0.0421246,0.0170843,-0.0000144318,0.00335585,0.0261357,0.040126,0.0279614,-0.037032,0.00482598,-0.00425426,0.0680297,-0.01547,0.0122319,-0.00172597,-0.017632,-0.00600065,0.0170651,-0.0359943,0.00524636,0.00296189,0.00167912,0.0287109,-0.00185809,0.00368735,-0.00884002,-0.0191886,0.0230801,0.0209662,-0.00214274,-0.0209086,0.0178146,0.0245599,0.031113,-0.00976726,-0.0290568,0.0136252,0.013212,0.0342839,0.0377431,-0.0259435,-0.0352448,0.0745637,-0.00756206,0.0146341,0.0379544,0.0117419,-0.00557786,0.0726419,-0.0380121,-0.0244446,-0.00902259,0.00736989,-0.0656083,-0.0203705,0.0461987,0.0744484,0.00908985,-0.0152394,-0.0104735,0.0216965,-0.172803,0.00589975,0.0101564,0.0493888,0.0188139,0.00830193,-0.0923974,0.00597662,-0.033669,-0.0121358,-0.0266162,-0.0669536,-0.0518486,-0.0139615,-0.0355907,-0.0118091,-0.04985,0.0324583,0.0346106,-0.0487354,0.00689426,-0.0277692,0.0152683,0.00349277,0.00480436,0.00267122,0.0330348,0.00889767,-0.0240795,-0.0332462,-0.0124817,-0.0507725,0.00258234,0.0361672,-0.00707202,-0.0200246,0.0288838,-0.0270966,0.0306518,-0.0458528,-0.0196979,0.0279422,0.0238873,0.0024214,-0.00681258,-0.0400107,-0.00597182,-0.0119148,-0.0551925,-0.0248866,-0.036417,-0.0320739,-0.00907544,-0.00951263,-0.0810976,0.00270966,-0.044546,0.0468521,0.0023181,0.00146653,-0.0292874,-0.0163636,0.030229,-0.0457759,-0.00612556,-0.0111942,-0.00677415,0.00810976,0.0296909,-0.00978168,0.053386,-0.00300993,-0.0245791,-0.0221193,-0.0292489,0.00562591,-0.0204281,-0.0152298,-0.0446229,-0.105619,-0.0452379,-0.011665,-0.0335536,0.0109155,-0.0402029,-0.0641478,0.0286148,0.0477746,0.0128853,0.210008,0.0278653,-0.0140864,-0.0037522,0.0511953,0.0088208,-0.0475824,-0.0252902,-0.00325495,-0.0374932,-0.0327273,-0.00734106,-0.00682219,-0.0215812,-0.0219463,0.0356676,-0.0591897,-0.0223499,0.0649165,-0.00948381,-0.000689426,-0.0112518,0.0059478,-0.0220424,-0.0408563,-0.0400876,0.0241948,-0.0279422,-0.0548081,0.0789452,0.0230609,0.0561918,0.0145956,-0.022004,-0.0148839,-0.0291913,0.00960391,0.0133369,-0.0214467,0.0541547,-0.037032,-0.0299408,-0.0132696,-0.0145764,0.00146413,0.000667806,0.0303828,0.042509,-0.00133081,0.0323814,-0.000711646,-0.0114344,-0.011444,0.0119725,0.0340917,-0.0118187,0.00825869,0.00711045,-0.022446,0.0184776,0.0586132,-0.00718252,-0.0123184,0.0269621,-0.00775904,0.0448151,0.00818182,0.019448,-0.00646186,0.014365,-0.0253862,0.0134234,0.00257994,0.00335585,0.021389,0.0329003,0.0115977,-0.0146917,0.0511184,0.0620723,0.0135003,-0.000102018,0.0143939,0.0196402,0.0277308,-0.0580367,-0.03659,0.0325351,-0.014144,-0.0212545,-0.0438158,-0.0228495,0.0211584,0.0176608,-0.0143939,-0.0173053,0.0205627,0.0371665,-0.0151145,-0.0183334,-0.0298062,-0.0338996,0.0252325,-0.00315166,-0.000766295,0.0159409,0.0313629,-0.00393958,-0.00293546,-0.0130871,-0.0192078,0.0208894,-0.00347595,0.025982,-0.0126643,0.0166327,0.00214875,0.0310362,0.056384,0.0186025,0.00562591,-0.0320739,-0.0245215,-0.0251748,-0.00640421,-0.0254631,-0.0261165,0.00251028,-0.0271158,0.0264048,0.0120205,-0.0279422,0.017584,0.0370897,0.035264,-0.00645706,-0.0237335,0.00351199,-0.00321412,-0.0222346,-0.0151433,-0.0306518,-0.0273272,-0.00458336,-0.00679336,0.00131519,-0.0177953,0.058421,0.00516469,0.00997385,0.0189676,0.0349757,-0.0348028,-0.059036,0.0308248,-0.00433353,0.027135,-0.00417018,0.0163636,-0.0493119,-0.00536647,-0.0298447,0.0388,0.0794065,-0.034207,-0.0320547,-0.0087103,0.0219079,-0.0157967,-0.00572199,0.0585748,-0.0199477,-0.0579214,0.0599968,0.00114764,-0.00159144,-0.0234261,-0.00372818,-0.0411638,-0.0368014,0.0102045,-0.00333903,-0.0254823,-0.0238873,0.00625047,0.0277115,0.0212545,-0.0325351,0.0288838,-0.0222538,-0.0337074,0.0871703,-0.000805931,-0.0272311,0.00569797,0.00741793,-0.036417,0.0108771,0.010118,-0.00163949,-0.0208317,0.0410484,0.0226381,-0.0188043,-0.000978888,-0.0360711,-0.014586,0.0473133,0.0127027,0.00589495,-0.0154604,0.0707586,-0.033227,-0.00934928,0.0288069,0.0220808,0.0516949,0.0151722,-0.031728,-0.0417787,-0.0230033,0.00832115,-0.00904661,-0.0460065,-0.0345337,-0.0330156,0.019006,-0.0253862,0.0161907,-0.000805331,-0.00144611,0.0128276,-0.0310746,-0.0348028,0.0463909,-0.0274425,0.00368014,0.0431624,0.0144323,-0.0242332,0.0191598,-0.0477746,0.0485048,0.0574217,0.0277115,0.00368014,-0.0642247,-0.000802328,-0.00952705,-0.0117034,-0.000449208,0.0023109,0.0497347,-0.0370512,-0.00344713,0.0438542,0.012991,0.0448535,-0.0286148,0.0175743,-0.0100507,0.0373203,-0.0329771,-0.031728,-0.0234453,-0.0388,0.0536935,-0.0644937,-0.0223114,-0.019006,0.0139134,-0.00516949,-0.0532707,-0.0310938,0.0179203,-0.00197459,0.00822987,0.0135675,-0.00421102,0.00768698,-0.0560381,0.00568356,-0.00909465,-0.0135483,0.0195634,0.0108482,0.0242332,0.00714408,0.0063898,-0.0456222,-0.0424321,-0.0117515,-0.0692212,0.0261357,-0.0322276,0.0145764,-0.0038603,0.0246944,-0.00605349,0.0309401,0.00933968,0.0169978,-0.0431239,-0.033227,0.0285187,0.0294988,0.00252709,-0.0334768,0.00656276,0.0194673,0.0108579,-0.00819143,-0.015643,-0.00469866,0.0172573,0.00202023,0.0131639,0.03438,-0.00695671,0.0127892,0.010608,-0.004629,-0.00406689,0.0224652,-0.0214082,0.00350718,0.014096,-0.0260973,0.014365,-0.000807132,-0.00978168,0.033054,-0.0203897,-0.029518,-0.0385117,0.0157775,0.032612,-0.0404719,-0.00106116,0.025367,0.0114824,-0.00314686,0.00748039,-0.0237143,0.0244446,0.0196402,-0.0287109,-0.00856617,-0.00333423,-0.0123664,0.0265008,0.0180163,-0.0115016,-0.0435083,0.049581,-0.00377142,0.0109251,0.0204666,0.00391796,-0.0286724,0.00906102,0.056384,-0.020678,-0.00802809,0.0548081,0.000265291,-0.009479,0.0141633,-0.00453051,-0.0316703,0.000943456,0.0301137,-0.0181124,0.0770235,-0.0660696,0.0175647,0.0221577,0.00155901,0.0370897,-0.054616,-0.0111845,-0.0148743,-0.0014377,-0.0143458,0.0221577,0.0226574,-0.00138486,-0.0256361,0.0270005,0.0260204,0.0312283,0.00761971,0.0313052,-0.0230225,-0.134753,0.00486922,0.0105696,0.00130919,-0.0495426,0.0478899,-0.00567875,-0.0174206,0.0171708,0.0059478,-0.0225997,0.0197171,0.0360327,-0.00557306,0.0285571,0.00150857,0.0270966,-0.0533092,-0.0343223,-00382187,0.00495329,0.0234261,0.0517333,0.0146917,0.0348989,0.000468425,-0.0104735,0.0503881,-0.0371665,-0.0268468,0.0338611,-0.00337026,0.0362057,0.00248866,0.0357637,0.0107714,0.00746117,0.0426243,-0.0382235,-0.000316787,0.0128565,-0.00701917,0.00811456,-0.0484664,0.0763317,0.0313821,-0.0263663,-0.0255976,-0.0206203,0.0044128,0.0470443,0.0237335,-0.0312668,0.00723056,0.0445845,-0.00385069,-0.0330732,0.0231762,-0.00895052,0.035437,0.000472028,0.0294027,-0.0283265,-0.0136156,-0.0233876,0.0298639,-0.0383388,-0.0192751,0.00844126,0.0137693,0.0120974,-0.0278269,0.0215812,-0.0353793,0.0349181,-0.0063946,0.0335921,0.000581027,-0.0330732,-0.00557306,0.0146245,0.0268468,-0.00780228,0.0596125,0.00445604,0.0333999,0.018737,-0.015864,-0.0503113,-0.0109539,-0.0189388,-0.00527038,0.016479,-0.0234837,-0.0126739,0.00749961,-0.0234068,0.0120974,0.0198708,-0.00701917,-0.009258,-0.0554615,0.00828272,0.0271735,-0.00616399,-0.0125009,0.0404719,-0.00423024,-0.00977687,0.00232891,-0.037301,0.0124625,-0.0243677,0.0051791,-0.0393957,-0.0284418,-0.00994502,0.0313052,-0.0115881,0.0233108,0.0120782,0.0014377,-0.0263279,-0.0324967,0.0205819,0.0403182,0.0358213,0.0154508,-0.00830193,-0.00961833,0.031113,0.0099162,0.0102525,0.0235221,0.0531554,-0.00775424,-0.0210431,0.026693,-0.0225613,-0.0592281,-0.0108771,0.00173678,-0.0120686,-0.00459057,-0.0353024,0.0245599,-0.013212,0.00196018,-0.00167192,-0.0321123,0.0228688,0.0365131,0.00295228,-0.00869109,-0.0618033,-0.0138846,-0.0118379,0.0532323,0.0097,-0.0263279,-0.0530786,-0.016085,-0.0145188,-0.0121646,0.0069471,-0.00858539,-0.0270966,-0.016921,0.0459297,-0.00492927,0.00928202,-0.00408611,-0.0236567,0.0105215,0.048082,-0.00917152,0.0172669,0.00185448,0.0435467,-0.0144035,0.0212929,0.0148166,-0.0140479,-0.00855656,0.0242332,0.0120589,0.0181701,-0.00252469,0.0131543,-0.0379737,-0.0458912,-0.00235294,0.00921957,-0.026866,0.0458912,-0.00790317,-0.0348797,0.0288262,0.00254631,0.0404335,-0.0364363,-0.00645706,-0.0181124,-0.0247905,0.00669728,0.0403566,-0.0561534,0.0560765,-0.00116025,0.0485433,0.0183719,0.0133946,-0.0417018,-0.00701437,0.00632734,-0.00184968,-0.0142305,0.0284995,0.00176921,0.0204666,0.0286916,-0.0175167,0.00338948,-0.0262702,-0.00534725,-0.00160105,-0.0139038,-0.0124721,0.0156334,0.0166423,-0.0147686,-0.0295564,0.0192751,0.00454012,-0.036321,-0.0150761,0.0124337,-0.0126451,0.0052944,-0.0490813,0.0329195,0.0187274,-0.0213506,-0.0204858]
1 row in set (0.001 sec)
现在，我们将目光转向图中数据库的右侧部分。这一部分涉及用户交互，并借助大模型利用向量数据来生成答案。
如图所示，当用户开始与 Chat Data 应用交互时，流程如下：
- 步骤 1 - 3： 用户的自然语言请求首先被模型转换为向量，随后在
dbgpt_test_db_profile表中进行相似度检索，以匹配最相关的元数据。 - 步骤 4 - 5： 大语言模型根据向量数据库返回的相似结果及其对应的文本元数据，将用户自然语言“翻译”为一条 SQL 语句。该 SQL 随后在 User Data 库中执行，获取结果数据。最后，应用可根据用户需求将结果数据渲染为合适的图表。
网络上展示的 Chat Data 应用效果如下：它能够根据用户自然语言直接查询数据库，并生成最适合呈现结果的图表。
经验与问题记录
坦率地说，本次实验过程相当顺畅，几乎没遇到什么实质性困难。但为了与前两篇文章保持一致的体例，这里依然“鸡蛋里挑骨头”，为 DB-GPT 和 OceanBase 指出几处小问题。
第一个问题是，OceanBase 为支持单集群多实例，引入了“租户”概念，这在许多数据库中并不常见。因此，若遇到连接串只要求填写用户名而无租户名时，记得将用户名格式设置为 user_name@tenant_name（后续在 DB-GPT 中创建连接时也会类似，灵活处理即可）。
第二个问题是拉取 DB-GPT 镜像的速度较慢，但这很可能是我个人网络环境所致。晚上九点开始拉取，等了十分钟只下载了百分之一二，无奈之下先回家打游戏。第二天到公司发现已拉取完成。公司周末的网络速度极快，因此这个下载速度可能不具代表性。如果你在公司环境下需要下载较大镜像，建议选择周末或深夜时段。
如果你使用的是 MacOS，可能会看到一个操作系统兼容性警告（WARNING）。不过既然只是警告而非错误（ERROR），可以忽略，实际使用中完全没问题。
第三个问题涉及 DB-GPT 的数据源更新机制。如果数据库中的元数据发生变化，Chat Data 应用不会自动感知并更新，需要手动删除该数据源并重新连接。据称旧版本中删除和重建操作合并为一个“刷新”按钮，但新版本不知为何移除了此功能。好在仍有变通方法，影响不大。（DB-GPT 新版本据说即将优化该功能）
数据准备
如果仅用于体验 Chat Data 应用，无需准备特殊数据，在数据库中随意添加一些测试数据即可。我的做法是导入了一个 TPCH 1G 的测试集，仅使用了 lineitem 这一张大表的数据。如果你已有 OceanBase 测试环境，直接使用现有数据即可。
效果展示
我的测试数据集中内容较为朴素，可能无法生成绚丽的图表。但各位可以通过以下简单示例，以小见大，感受该应用的内在潜力。
示例一
首先让 Chat Data 编写一条简单 SQL，对 TPCH 测试集进行查询并生成可视化图表。效果如下：
Chat Data 会自动执行生成的 SQL 并返回查询结果。这种体验几乎可以直接当作 SQL 执行客户端来使用……
生成图表时，它并未明确标注横纵坐标含义。可能是因为 SQL 过于简单，Chat Data 自动选择的柱状图已是最优呈现。此外，您可以直接在柱状图上选择需要展示的数据系列。
示例二
接下来“使点坏”，测试窗口函数。确实难以想象，Chat Data 会如何处理窗口函数的计算结果并生成图表。
事实证明，只要自然语言描述足够清晰，生成的 SQL 就能准确无误。
画图的结果略出乎意料。Chat Data 似乎感知到这条 SQL 是在故意刁难，直接放弃生成图表。这与那些不懂装懂、一本正经胡说八道的问答机器人不同，Chat Data 懂得“量力而行”。对于 AI 而言，“自知之明”或许是个值得肯定的品质。
如果换成其他适合可视化展示结果的 SQL，它就能一次性生成多种图表供您选择。
示例三
最后，我发现数据库中自带一张名为 plant_and_animal_table 的表，很可能是有人特意准备的。其中包含一个向量类型的列。
desc plant_and_animal_table; +----------------+--------------+------+-----+---------+----------------+ | Field | Type | Null | Key | Default | Extra | +----------------+--------------+------+-----+---------+----------------+ | id | int(11) | NO | PRI | NULL | auto_increment | | name | text | YES | | NULL | | | name_embedding | VECTOR(1536) | YES | | NULL | | +----------------+--------------+------+-----+---------+----------------+ 3 rows in set (0.013 sec)
我猜测，图中将原本 1536 维的向量简化为了 2 维。不同维度的向量值可以反映生物的不同属性。例如，X 轴的值可能与生物是动物还是植物具有强相关性。
有同学建议，在坐标系中生成散点图，可以更直观地根据点与点之间的距离判断两种生物的相似度。不过图中的数据点过多，区分颜色恐怕是个挑战。