RAG_only_for_hsbc_uk_bank_local.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "gpuType": "T4"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "Get hsbc bank dataset from this google drive: https://drive.google.com/file/d/1gFVEzyTBgDmMP48SjySA3skq8nwRxZZw/view?usp=sharing"
      ],
      "metadata": {
        "id": "-6LGAl2AGj4j"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "filename = 'dataset_hsbc'\n",
        "with open(filename, 'r') as f:\n",
        "    document = f.read()"
      ],
      "metadata": {
        "id": "cWtP9m9unZNP"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "document_chunks = document.split('########')\n",
        "print('Number of texts:', len(document_chunks))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "cbWUXpnVobB9",
        "outputId": "7b25a9f2-af4a-49dd-b294-3064317883cf"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of texts: 565\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## RAG (Retrieval Augmented Generation)\n",
        "Retriving relevant HSBC information.\n",
        "1. Define query.\n",
        "2. Encode into embedding space. Retrieve most similar chunks of text.\n",
        "3. Improve retrieved chunks of text via reranker - rearrange it in an improved order.\n",
        "\n",
        "Excellent open source embeddings: https://github.com/FlagOpen/FlagEmbedding"
      ],
      "metadata": {
        "id": "3klGmAwEDxDY"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "pip install FlagEmbedding"
      ],
      "metadata": {
        "id": "OqVExC56qYXF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from FlagEmbedding import FlagModel, FlagReranker\n",
        "import numpy as np"
      ],
      "metadata": {
        "id": "nrBLEiWqq16F"
      },
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Initialize embedding model."
      ],
      "metadata": {
        "id": "1BlAqzrbTLi4"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "embeddings_model_bge = FlagModel('BAAI/bge-base-en-v1.5',\n",
        "                       use_fp16=True) # Setting use_fp16 to True speeds up computation with a slight performance degradation"
      ],
      "metadata": {
        "id": "maa9zPqp1jKR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "query = \"how do i get student loan\""
      ],
      "metadata": {
        "id": "3VwI3SmcZhca"
      },
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "## encode dataset\n",
        "embeddings_bge = embeddings_model_bge.encode(document_chunks)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "038d455b-b26f-4a10-e9ae-ccabe1310429",
        "id": "5mgQaML-3dYm"
      },
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Inference Embeddings: 100%|██████████| 3/3 [00:07<00:00,  2.38s/it]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "emb_bge_query = embeddings_model_bge.encode([query])"
      ],
      "metadata": {
        "id": "oROlBTnm0hYC"
      },
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Getting similarity scores via dot product\n",
        "scores = emb_bge_query @ embeddings_bge.T\n",
        "scores = np.squeeze(scores)"
      ],
      "metadata": {
        "id": "NOGFWBg-0rp7"
      },
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Sort scores from large to small and obtaining indexes of them\n",
        "max_idx = np.argsort(-scores)\n",
        "max_idx[:8]"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ggpWrw8N896o",
        "outputId": "6029604e-dc4e-4f79-ea14-6f81c80c6163"
      },
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([200, 452, 196, 253, 252, 176, 201,  72])"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(f\"Query: {query}\\n\")\n",
        "context_chunks_init = []\n",
        "context_scores = []\n",
        "for idx in max_idx[:8]:\n",
        "  print(f\"Score: {scores[idx]:.3f}\")\n",
        "  print(document_chunks[idx].split('\\n')[1])\n",
        "  print(\"--------\")\n",
        "  context_chunks_init.append(document_chunks[idx])\n",
        "  context_scores.append(scores[idx])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fuRPAA_v2Dfd",
        "outputId": "1ede5358-8e4d-4db9-d731-98fa29c1216a"
      },
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Query: how do i get student loan\n",
            "\n",
            "Score: 0.777\n",
            "Borrowing money as a student \n",
            "--------\n",
            "Score: 0.735\n",
            "Paying off your student loan early \n",
            "--------\n",
            "Score: 0.720\n",
            "How to budget as a student \n",
            "--------\n",
            "Score: 0.715\n",
            "What is a loan? \n",
            "--------\n",
            "Score: 0.709\n",
            "Tips to successfully apply for a loan \n",
            "--------\n",
            "Score: 0.708\n",
            "How much can I borrow? \n",
            "--------\n",
            "Score: 0.702\n",
            "Should you get a student credit card? \n",
            "--------\n",
            "Score: 0.701\n",
            "Ways to borrow \n",
            "--------\n"
          ]
        }
      ]
    }
  ]
}