feat: Split text into semantic chunks and save as parquet file

dale_retriver_agent.ipynb

@@ -2,30 +2,33 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 3,
    "metadata": {},
    "outputs": [],
    "source": [
-    "from langchain.embeddings.cache import CacheBackedEmbeddings\n",
-    "from langchain.storage import LocalFileStore\n",
-    "from langchain_community.embeddings import HuggingFaceBgeEmbeddings\n",
-    "\n",
-    "underlying_embedder = HuggingFaceBgeEmbeddings(\n",
-    "    model_name=\"BAAI/bge-m3\",\n",
-    ")\n",
-    "\n",
-    "store = LocalFileStore(\"./.cache/embeddings\")\n",
-    "\n",
-    "embedder = CacheBackedEmbeddings.from_bytes_store(\n",
-    "    underlying_embedder, store, namespace=underlying_embedder.model_name\n",
-    ")"
+    "text = \"\"\n",
+    "with open(\n",
+    "    \"./raw_data/dale_carnegie/how_to_win_friends_and_influence_people.txt\", \"r\"\n",
+    ") as f:\n",
+    "    text = f.read()"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 26,
+   "execution_count": 4,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "139"
+      ]
+     },
+     "execution_count": 4,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "from langchain_text_splitters import RecursiveCharacterTextSplitter, Language\n",
     "\n",
@@ -35,36 +38,83 @@
     "    ),\n",
     "    chunk_size=4096,\n",
     "    keep_separator=True,\n",
-    ")"
+    ")\n",
+    "\n",
+    "splitted_text = text_splitter.split_text(text)\n",
+    "len(splitted_text)"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 28,
+   "execution_count": 10,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
-       "139"
+       "207"
       ]
      },
-     "execution_count": 28,
+     "execution_count": 10,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "text = \"\"\n",
-    "with open(\n",
-    "    \"./raw_data/dale_carnegie/how_to_win_friends_and_influence_people.txt\", \"r\"\n",
-    ") as f:\n",
-    "    text = f.read()\n",
+    "# Semantic Splitting\n",
+    "from langchain.storage import LocalFileStore\n",
+    "from langchain.embeddings.cache import CacheBackedEmbeddings\n",
+    "from langchain_experimental.text_splitter import SemanticChunker\n",
+    "from langchain_openai.embeddings import OpenAIEmbeddings\n",
+    "\n",
+    "underlying_embedder = OpenAIEmbeddings(model=\"text-embedding-3-small\")\n",
+    "\n",
+    "store = LocalFileStore(\"./.cache/embeddings\")\n",
     "\n",
+    "embedder = CacheBackedEmbeddings.from_bytes_store(\n",
+    "    underlying_embedder, store, namespace=underlying_embedder.model\n",
+    ")\n",
+    "\n",
+    "text_splitter = SemanticChunker(embedder)\n",
     "splitted_text = text_splitter.split_text(text)\n",
     "len(splitted_text)"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "\n",
+    "df = pd.DataFrame(splitted_text, columns=[\"text\"])\n",
+    "df.to_parquet(\n",
+    "    \"./raw_data/dale_carnegie/how_to_win_friends_and_influence_people.parquet\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain.storage import LocalFileStore\n",
+    "from langchain.embeddings.cache import CacheBackedEmbeddings\n",
+    "from langchain_community.embeddings import HuggingFaceBgeEmbeddings\n",
+    "\n",
+    "underlying_embedder = HuggingFaceBgeEmbeddings(\n",
+    "    model_name=\"BAAI/bge-m3\",\n",
+    ")\n",
+    "\n",
+    "store = LocalFileStore(\"./.cache/embeddings\")\n",
+    "\n",
+    "embedder = CacheBackedEmbeddings.from_bytes_store(\n",
+    "    underlying_embedder, store, namespace=underlying_embedder.model_name\n",
+    ")"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 29,

raw_data/dale_carnegie/how_to_win_friends_and_influence_people.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75b5718819389b15bf54e5edb329f888cf0b9f2dd69e418b29e7eedee8747c74
+size 259681