{"id":721,"date":"2024-05-24T23:19:36","date_gmt":"2024-05-24T14:19:36","guid":{"rendered":"https:\/\/hirabay.net\/?p=721"},"modified":"2024-05-24T23:19:40","modified_gmt":"2024-05-24T14:19:40","slug":"%e3%80%90openai-embedding%e3%80%91%e7%b2%be%e5%ba%a6%e3%82%92%e3%81%82%e3%81%92%e3%82%8b%e3%81%9f%e3%82%81%e3%81%ae%e4%ba%8b%e5%89%8d%e5%87%a6%e7%90%86","status":"publish","type":"post","link":"https:\/\/hirabay.net\/?p=721","title":{"rendered":"\u3010OpenAI Embedding\u3011\u7cbe\u5ea6\u3092\u3042\u3052\u308b\u305f\u3081\u306e\u4e8b\u524d\u51e6\u7406"},"content":{"rendered":"\n<p>\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u306e\u5185\u5bb9\u3092\u3082\u3068\u306bAI\u3092\u4f7f\u3063\u3066QA\u3092\u3057\u3088\u3046\u3068\u3059\u308b\u3068\u3001<br>OpenAI \u306e\u5834\u5408Embedding\u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u306b\u306a\u308b\u3068\u601d\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u305f\u3060\u3001\u3053\u306eEmbedding\u306f\u305f\u3060\u30d5\u30a1\u30a4\u30eb\u3092\u305d\u306e\u307e\u307e\u98df\u308f\u305b\u3066\u3082\u306a\u304b\u306a\u304b\u7cbe\u5ea6\u304c\u51fa\u307e\u305b\u3093\u3002<\/p>\n\n\n\n<p>\u3069\u3046\u3057\u305f\u3089\u7cbe\u5ea6\u304c\u3067\u308b\u304b\u3068\u8a66\u884c\u932f\u8aa4\u3057\u305f\u7d50\u679c\u3001\u4f55\u500b\u304b\u7cbe\u5ea6\u306b\u95a2\u4fc2\u3057\u305d\u3046\u306a\u4e8b\u524d\u51e6\u7406\u3092\u767a\u898b\u3057\u305f\u306e\u3067\u3054\u7d39\u4ecb\u3057\u305f\u3044\u3068\u601d\u3044\u307e\u3059\uff01<\/p>\n\n\n\n<div class=\"wp-block-sgb-say\"><div class=\"sgb-block-say sgb-block-say--left\"><div class=\"sgb-block-say-avatar\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/hirabay.net\/wp-content\/uploads\/2023\/01\/c90be45c419f1b74db0f1d5a262d000d.png\" alt=\"\u3072\u3089\u3079\u30fc\" width=\"80\" height=\"80\" style=\"border-color:#eaedf2\"><div class=\"sgb-block-say-avatar__name\">\u3072\u3089\u3079\u30fc<\/div><\/div><div class=\"sgb-block-say-text\"><div class=\"sgb-block-say-text__content\" style=\"color:#333;border-color:#d5d5d5;background-color:#FFF\">\n<p><strong><span class=\"keiko_yellow\">\u3053\u306e\u8a18\u4e8b\u306f\u3053\u3093\u306a\u4eba\u306b\u30aa\u30b9\u30b9\u30e1\uff01<\/span><\/strong><\/p>\n\n\n\n<p>\u30fbEmbedding\u3092\u3053\u308c\u304b\u3089\u8a66\u305d\u3046\u3068\u601d\u3063\u3066\u3044\u308b<\/p>\n\n\n\n<p>\u30fbEmbedding\u3092\u8a66\u3057\u305f\u3089\u7cbe\u5ea6\u304c\u60aa\u3044<\/p>\n<span class=\"sgb-block-say-text__before\" style=\"border-right-color:#d5d5d5\"><\/span><span class=\"sgb-block-say-text__after\" style=\"border-right-color:#FFF\"><\/span><\/div><\/div><\/div><\/div>\n\n\n\n<p>\u203b \u672c\u8a18\u4e8b\u306e\u7cbe\u5ea6\u306b\u95a2\u3059\u308b\u8a00\u53ca\u306f\u3001\u5ba2\u89b3\u7684\u306a\u4e8b\u5b9f\u3067\u306f\u306a\u304f\u7d4c\u9a13\u5247\u304b\u3089\u306e\u4e3b\u89b3\u7684\u306a\u3082\u306e\u3068\u306a\u308a\u307e\u3059\u3002<br>\u3054\u4e86\u627f\u306e\u4e0a\u3054\u53c2\u8003\u307e\u3067\u306b\u3054\u89a7\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n<div class=\"toc\"><br \/>\n<b>Warning<\/b>:  Undefined array key \"is_admin\" in <b>\/home\/c4179872\/public_html\/hirabay.net\/wp-content\/themes\/sango-theme\/library\/gutenberg\/dist\/classes\/Toc.php<\/b> on line <b>110<\/b><br \/>\n<br \/>\n<b>Warning<\/b>:  Undefined array key \"is_category_top\" in <b>\/home\/c4179872\/public_html\/hirabay.net\/wp-content\/themes\/sango-theme\/library\/gutenberg\/dist\/classes\/Toc.php<\/b> on line <b>115<\/b><br \/>\n<br \/>\n<b>Warning<\/b>:  Undefined array key \"is_top\" in <b>\/home\/c4179872\/public_html\/hirabay.net\/wp-content\/themes\/sango-theme\/library\/gutenberg\/dist\/classes\/Toc.php<\/b> on line <b>122<\/b><br \/>\n    <div id=\"toc_container\" class=\"sgb-toc--bullets js-smooth-scroll\">\n      <p class=\"toc_title\">\u76ee\u6b21 <\/p>\n      <ul class=\"toc_list\">\n  <li class=\"first\">\n    <a href=\"#i-0\">\u8868\u5f62\u5f0f\u306e\u30c7\u30fc\u30bf\u306f\uff11\u884c\uff11\u884c\u3067\u5b8c\u7d50\u3055\u305b\u308b<\/a>\n  <\/li>\n  <li class=\"last\">\n    <a href=\"#i-1\">\u533a\u5207\u308a\u4f4d\u7f6e\u3092\u5236\u5fa1\u3059\u308b<\/a>\n  <\/li>\n<\/ul>\n\n      \n    <\/div><\/div><h2 class=\"wp-block-heading\" id=\"i-0\">\u8868\u5f62\u5f0f\u306e\u30c7\u30fc\u30bf\u306f\uff11\u884c\uff11\u884c\u3067\u5b8c\u7d50\u3055\u305b\u308b<\/h2>\n\n\n\n<p>\u5148\u306b\u3069\u3046\u3044\u3046\u3053\u3068\u304b\u4f8b\u793a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u307e\u305a\u305d\u306e\u307e\u307eEmbedding\u3059\u308b\u3068\u7cbe\u5ea6\u304c\u60aa\u3081\u306a\u4f8b\u3067\u3059\u3002<\/p>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-markdown\" data-lang=\"Markdown\"><code>| TABLE_NAME    | COLUMN_NAME    | EXPLAIN         |\n| ------------- | -------------- | --------------- |\n| SAMPLE_TABLE1 | SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n|               | SAMPLE_COLUMN2 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n|               | SAMPLE_COLUMN3 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_TABLE2 | SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|<\/code><\/pre><\/div>\n\n\n\n<p>\u6b21\u306b\u7cbe\u5ea6\u304c\u6bd4\u8f03\u7684\u826f\u3044\u4f8b\u3067\u3059\u3002<\/p>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-markdown\" data-lang=\"Markdown\"><code>| TABLE_NAME    | COLUMN_NAME    | EXPLAIN         |\n| ------------- | -------------- | --------------- |\n| SAMPLE_TABLE1 | SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_TABLE1 | SAMPLE_COLUMN2 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_TABLE1 | SAMPLE_COLUMN3 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_TABLE2 | SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|<\/code><\/pre><\/div>\n\n\n\n<p>\u524d\u8005\u306f\u3001<code>TABLE_NAME<\/code>\u304c\u5171\u901a\u306a\u306e\u3067\u8a18\u8f09\u3092\u7701\u3044\u3066\u3044\u307e\u3059\u304c\u3001\u5f8c\u8005\u3067\u306f\u3059\u3079\u3066\u306e\u30bb\u30eb\u304c\u57cb\u307e\u3063\u3066\u3044\u307e\u3059\u3002<br>\u5f8c\u8005\u3060\u3068\u3069\u306e\uff11\u884c\u3092\u5207\u308a\u53d6\u3063\u3066\u3082\u610f\u5473\u304c\u901a\u3058\u308b\u306e\u3067\u7cbe\u5ea6\u304c\u4e0a\u304c\u308b\u50be\u5411\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u3082\u3057\u304f\u306f\u8868\u3092\u5206\u5272\u3057\u3066\u3057\u307e\u3046\u3068\u3044\u3046\u306e\u3082\u3042\u308a\u3067\u3059\u3002<\/p>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-markdown\" data-lang=\"Markdown\"><code># SAMPLE_TABLE1\n\n| COLUMN_NAME    | EXPLAIN         |\n| -------------- | --------------- |\n| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n\n# SAMPLE_TABLE2\n\n| COLUMN_NAME    | EXPLAIN         |\n| -------------- | --------------- |\n| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|<\/code><\/pre><\/div>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"i-1\">\u533a\u5207\u308a\u4f4d\u7f6e\u3092\u5236\u5fa1\u3059\u308b<\/h2>\n\n\n\n<p>Embedding\u3092\u884c\u3046\u969b\u306f\u3001\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3092\u5168\u6587\u305d\u306e\u307e\u307e\u5b66\u7fd2\u3055\u305b\u308b\u306e\u3067\u306f\u306a\u304f\u3001\u4e00\u5b9a\u306e\u6587\u5b57\u6570\u4ee5\u5185\u306b\u304a\u3055\u307e\u308b\u3088\u3046\u5206\u5272\u3092\u3057\u307e\u3059\u3002<br>\u3053\u306e\u5206\u5272\u306e\u969b\u3001\u610f\u5473\u306e\u3042\u308b\u307e\u3068\u307e\u308a\u3067\u5206\u5272\u3067\u304d\u308b\u304b\u304c\u809d\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4f8b\u3048\u3070\u3001\u5148\u307b\u3069\u306e\u8868\u3092100\u6587\u5b57\u3067\u5206\u5272\u3059\u308b\u3068\u3053\u3093\u306a\u611f\u3058\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-markdown\" data-lang=\"Markdown\"><code># SAMPLE_TABLE1\n\n| COLUMN_NAME    | EXPLAIN         |\n| -------------- | --------------- |<\/code><\/pre><\/div>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-markdown\" data-lang=\"Markdown\"><code>| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|\n| SAMPLE_COLUMN1 | \u3053\u308c\u306f\u30b5\u30f3\u30d7\u30eb\u3067\u3059\u3002|<\/code><\/pre><\/div>\n\n\n\n<p>\u3053\u308c\u3067\u306f\u5206\u5272\u3055\u308c\u305f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u5358\u72ec\u3067\u306f\u610f\u5473\u3092\u306a\u3055\u306a\u304f\u306a\u308a\u3001\u7cbe\u5ea6\u304c\u4e0b\u304c\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u306a\u306e\u3067\u3001\u304d\u3061\u3093\u3068\u610f\u5473\u306e\u3042\u308b\u30d6\u30ed\u30c3\u30af\u304c\u307e\u3068\u307e\u3063\u3066\u5206\u5272\u3055\u308c\u308b\u3088\u3046\u306b\u300c\u3042\u308b\u7a0b\u5ea6\u306e\u5927\u304d\u3055\u306e\u30c1\u30e3\u30f3\u30af\u3092\u8a2d\u5b9a\u3059\u308b\u3053\u3068\u300d\u3068\u300c\u5206\u5272\u3055\u308c\u308b\u4f4d\u7f6e\u3092\u8abf\u6574\u3059\u308b\u3053\u3068\u300d\u304c\u5fc5\u8981\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u306e\u5206\u5272\u306b\u306fLangChain\u304c\u4f7f\u3048\u305d\u3046\u3060\u3063\u305f\u306e\u3067\u3059\u304c\u3001\u3082\u3046\u5c11\u3057\u81ea\u5206\u3067\u52d5\u304d\u3092\u30b3\u30f3\u30c8\u30ed\u30fc\u30eb\u3057\u305f\u3044\u3068\u601d\u3044\u3001<br>\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u95a2\u6570\u3092\u5b9f\u88c5\u3057\u3066\u307f\u307e\u3057\u305f\u3002<\/p>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-python\" data-lang=\"Python\"><code># \u6587\u5b57\u30b3\u30fc\u30c9\u306e\u81ea\u52d5\u5224\u5225\u306e\u6a5f\u80fd\u304c\u4fbf\u5229\u306a\u306e\u3067File -&gt; \u30c6\u30ad\u30b9\u30c8\u306e\u90e8\u5206\u3060\u3051LangChain\u3092\u4f7f\u7528\nfrom langchain_community.document_loaders import TextLoader\n\n    # separators\u3067\u6307\u5b9a\u3055\u308c\u305f\u6587\u5b57\u5217\u3067\u9806\u306b\u30c6\u30ad\u30b9\u30c8\u3092\u5206\u5272\u3059\u308b\n    # chunk_size\u3067\u6307\u5b9a\u3055\u308c\u305f\u30b5\u30a4\u30ba\u4ee5\u4e0b\u306b\u306a\u3063\u305f\u3089\u305d\u3053\u3067\u5206\u5272\u3092\u7d42\u4e86\u3059\u308b\n    def split_text_recursive(\n            self,\n            file: str,\n            separate_regex_list: list[str],\n            chunk_size: int = 2000,\n    ) -&gt; list[str]:\n        split_text = [doc.page_content for doc in TextLoader(file, autodetect_encoding=True).load()]\n        for separator in separate_regex_list:\n            new_split_text = []\n            for text_block in split_text:\n                if len(text_block) &gt; chunk_size:\n                    # \u4f8b: \\n\u3067\u5206\u5272\u3059\u308b\u5834\u5408\n                    # \u524d: a\\nb\\nc\n                    # \u5f8c: [a, \\n, b, \\n, c]\n                    split_text = re.split(f'({separator})', text_block)\n                    # \u4f8b:\n                    # \u524d: [a, \\n, b, \\n, c]\n                    # \u5f8c: [a, \\nb, \\nc]\n                    new_split_text.append(split_text[0])\n                    for i in range(2, len(split_text), 2):\n                        new_split_text.append(split_text[i-1] + split_text[i])\n\n                else:\n                    new_split_text.append(text_block)\n            # \u7d30\u304b\u3059\u304e\u308b\u5206\u5272\u306f\u7d50\u5408\n            split_text = self.merge_text(new_split_text)\n\n        return split_text\n\n    def merge_text(self, text_list: list[str], chunk_size: int = 2000) -&gt; list[str]:\n        # chunk_size\u3067\u6307\u5b9a\u3055\u308c\u305f\u30b5\u30a4\u30ba\u4ee5\u4e0b\u3060\u3063\u305f\u3089\u6b21\u306e\u8981\u7d20\u3068\u306e\u7d50\u5408\u3092\u8a66\u307f\u308b\n        # \u2192\u3000\u65b0\u3057\u3044\u914d\u5217\u306b\u8ffd\u52a0\n        merged_text = ''\n        merged_text_list = []\n        for text in text_list:\n            # \u8981\u7d20\u3092\u7d50\u5408\u3057\u3066\u307f\u3066\u3001\u6307\u5b9a\u30b5\u30a4\u30ba\u4ee5\u4e0b\u3060\u3063\u305f\u3089\u7d50\u5408\n            if len(merged_text) + len(text) &lt; chunk_size:\n                merged_text += text\n            else:\n                merged_text_list.append(merged_text)\n                merged_text = text\n\n        if merged_text:\n            merged_text_list.append(merged_text)\n\n        return merged_text_list\n<\/code><\/pre><\/div>\n\n\n\n<p>\u4f7f\u3044\u65b9\u306e\u30a4\u30e1\u30fc\u30b8\u306f\u3053\u3093\u306a\u611f\u3058\u3067\u3059\uff08SQL\u3084PLSQL\u30d5\u30a1\u30a4\u30eb\u3092\u5bfe\u8c61\u306b\u3057\u305f\u4f8b\uff09<\/p>\n\n\n\n<div class=\"hcb_wrap\"><pre class=\"prism undefined-numbers lang-plain\"><code>        separators = [\n            '\\n\\\/\\*\\n',  # PLSQL\u306e\u30b3\u30e1\u30f3\u30c8\n            '\\n *CREATE ', '\\n *create ',\n            '\\n *INSERT ', '\\n *insert ',\n            '\\n *UPDATE ', '\\n *update ',\n            '\\n *DELETE ', '\\n *delete ',\n            '\\n *ALTER ', '\\n *alter ',\n            '\\n *DROP ', '\\n *drop ',\n            '\\n *SELECT ', '\\n *select ',\n            '\\n *PROCEDURE ', '\\n *procedure ',\n            '\\n *BEGIN ', '\\n *begin ',\n            '\\n *EXCEPTION ', '\\n *exception ',\n        ]\n        split_text = split_text_recursive(file=file, separate_regex_list=separators, chunk_size=2000)<\/code><\/pre><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u306e\u5185\u5bb9\u3092\u3082\u3068\u306bAI\u3092\u4f7f\u3063\u3066QA\u3092\u3057\u3088\u3046\u3068\u3059\u308b\u3068\u3001OpenAI \u306e\u5834\u5408Embedding\u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u306b\u306a\u308b\u3068\u601d\u3044\u307e\u3059\u3002 \u305f\u3060\u3001\u3053\u306eEmbedding\u306f\u305f\u3060\u30d5\u30a1\u30a4\u30eb\u3092\u305d\u306e\u307e\u307e\u98df\u308f\u305b\u3066\u3082\u306a\u304b\u306a\u304b\u7cbe\u5ea6\u304c\u51fa\u307e\u305b\u3093\u3002  &#8230; <\/p>\n","protected":false},"author":1,"featured_media":725,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[111,110,51],"class_list":{"0":"post-721","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-6","8":"tag-embedding","9":"tag-openai","10":"tag-python"},"jetpack_featured_media_url":"https:\/\/hirabay.net\/wp-content\/uploads\/2024\/05\/OpenAI.png","_links":{"self":[{"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/posts\/721","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/hirabay.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=721"}],"version-history":[{"count":4,"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/posts\/721\/revisions"}],"predecessor-version":[{"id":726,"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/posts\/721\/revisions\/726"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/hirabay.net\/index.php?rest_route=\/wp\/v2\/media\/725"}],"wp:attachment":[{"href":"https:\/\/hirabay.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=721"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hirabay.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=721"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hirabay.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=721"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}