{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Notebook 3: Machine Learning from Examples & Recommendation Systems\n",
        "\n",
        "In Chapter 4, we learned the difference between rigid, rule-based software and **Machine Learning**: instead of writing every rule by hand, the machine discovers boundaries directly from labelled data.\n",
        "\n",
        "In this notebook, we experiment with threshold learning and build a Nearest Neighbor recommendation engine."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Experiment 1: The Computer Learns a Boundary from Examples\n",
        "\n",
        "We provide numeric examples labelled either `small` or `big`. Notice that we do not write an `if number < 20` rule ourselves. The computer calculates the averages of each group and places the decision boundary right in the middle!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {},
      "source": [
        "# Our labelled training dataset\n",
        "examples = [\n",
        "    (2, \"small\"), (5, \"small\"), (3, \"small\"), (7, \"small\"), (4, \"small\"),\n",
        "    (40, \"big\"), (35, \"big\"), (50, \"big\"), (42, \"big\"), (38, \"big\")\n",
        "]\n",
        "\n",
        "small_vals = [n for n, label in examples if label == \"small\"]\n",
        "big_vals = [n for n, label in examples if label == \"big\"]\n",
        "\n",
        "avg_small = sum(small_vals) / len(small_vals)\n",
        "avg_big = sum(big_vals) / len(big_vals)\n",
        "learned_boundary = (avg_small + avg_big) / 2\n",
        "\n",
        "print(f\"Average of small examples: {avg_small}\")\n",
        "print(f\"Average of big examples:   {avg_big}\")\n",
        "print(f\"AI learned this boundary automatically: {learned_boundary}\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {},
      "source": [
        "# Let us test our learned boundary on brand new numbers!\n",
        "def predict_size(n):\n",
        "    return \"small\" if n < learned_boundary else \"big\"\n",
        "\n",
        "for new_n in [1, 12, 21, 30, 60]:\n",
        "    print(f\"Number {new_n} -> AI predicts: {predict_size(new_n)}\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Experiment 2: Nearest Neighbors (`k-NN`) Recommendation System\n",
        "\n",
        "When you visit an online store or agricultural seed catalog, AI recommends items that are most *similar* to what you like (`Nearest Neighbors`).\n",
        "\n",
        "Let us build a simple Python recommender that finds the most suitable traditional crop variety based on soil moisture and sunlight scores!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {},
      "source": [
        "import math\n",
        "\n",
        "# Crop catalog with features: (moisture_score, sunlight_score)\n",
        "crops = {\n",
        "    \"Khejri Tree Pods (Sangri)\": (20, 90),\n",
        "    \"Pearl Millet (Bajra)\": (30, 85),\n",
        "    \"Green Gram (Moong)\": (45, 75),\n",
        "    \"Guar (Cluster Bean)\": (35, 80)\n",
        "}\n",
        "\n",
        "def recommend_crop(soil_moisture, sunlight_level):\n",
        "    best_crop = None\n",
        "    min_dist = float(\"inf\")\n",
        "    \n",
        "    for name, (m, s) in crops.items():\n",
        "        # Euclidean distance formula\n",
        "        dist = math.sqrt((soil_moisture - m)**2 + (sunlight_level - s)**2)\n",
        "        if dist < min_dist:\n",
        "            min_dist = dist\n",
        "            best_crop = name\n",
        "    return best_crop, min_dist\n",
        "\n",
        "# Let us recommend crops for a field with moisture=25 and sunlight=88\n",
        "chosen, dist = recommend_crop(25, 88)\n",
        "print(f\"For your soil (moisture=25, sunlight=88), AI recommends: '{chosen}'\")"
      ],
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}