{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "scratch_model_weight_changes_affect_accuracy.ipynb",
      "provenance": [],
      "authorship_tag": "ABX9TyPdI1D7PI0pSgwlrQULZj1P",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/raqueeb/TensorFlow2/blob/master/scratch_model_weight_changes_affect_accuracy.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HomHZrXgZHtm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "\n",
        "# আমাদের প্রেডিকশন করার জন্য ডেটা পয়েন্ট, ছবির সাথে মিলিয়ে দেখুন\n",
        "input_data = np.array([2, 3])\n",
        "\n",
        "# আমাদের ডিকশনারী\n",
        "weights =   {'node_0': np.array([1, 1]),\n",
        "             'node_1': np.array([-1, 1]),\n",
        "             'output': np.array([2, -1])\n",
        "            }"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v1b80TK6ZSWL",
        "colab_type": "code",
        "outputId": "7ac3581a-e48e-41f4-a39c-856b6603f0b0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "# node 0 এর ভ্যালু ক্যালকুলেট করি: node_0_value\n",
        "node_0_value = (input_data * weights['node_0']).sum()\n",
        "\n",
        "# node ১ এর ভ্যালু ক্যালকুলেট করি: node_1_value\n",
        "node_1_value = (input_data * weights['node_1']).sum()\n",
        "\n",
        "# নোডগুলোর ভ্য়ালুগুলোকে অ্যারেতে রাখি: hidden_layer_outputs\n",
        "hidden_layer_outputs = np.array([node_0_value, node_1_value])\n",
        "\n",
        "# আউটপুট ক্যালকুলেট করি: output\n",
        "output = (hidden_layer_outputs * weights['output']).sum()\n",
        "\n",
        "# আউটপুট প্রিন্ট করে দেখি\n",
        "print(hidden_layer_outputs)\n",
        "print(output)"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[5 1]\n",
            "9\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kNrRxANoZiet",
        "colab_type": "code",
        "outputId": "5af736d6-45ae-43d8-f7eb-f369a55213ca",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "# নতুন ওয়েট এবং ইনপুট ডেটা \n",
        "weights = np.array([1, 2])\n",
        "input_data = np.array([3, 4]) \n",
        "\n",
        "# প্রেডিকশন ক্যাল্কুলেট করি: preds\n",
        "preds = (weights * input_data).sum()\n",
        "\n",
        "# ধরি আমাদের টার্গেট ৬\n",
        "target = 6\n",
        "\n",
        "# এরর ক্যালকুলেট করি: error\n",
        "error = preds - target\n",
        "\n",
        "# স্লোপ ক্যালকুলেট করি: slope\n",
        "slope = 2 * input_data * error\n",
        "\n",
        "# স্লোপ প্রিন্ট করি \n",
        "print(slope)\n"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[30 40]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "B_U_OVOkaJXp",
        "colab_type": "code",
        "outputId": "038ce0a4-e45f-4e6e-ab0d-a5bb886c9bd8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "# লার্নিং রেট ঠিক করি: learning_rate\n",
        "learning_rate = 0.01\n",
        "\n",
        "# স্লোপ/গ্রেডিয়েন্ট ক্যালকুলেট করি: gradient\n",
        "gradient = 2 * input_data * error\n",
        "\n",
        "# ওয়েট আপডেট করি: weights_updated\n",
        "weights_updated = weights - learning_rate * gradient\n",
        "\n",
        "# প্রেডিকশন আপডেট নেই : preds_updated\n",
        "preds_updated = (weights_updated * input_data).sum()\n",
        "\n",
        "# এররের আপডেট নেই: error_updated\n",
        "error_updated = preds_updated - target\n",
        "\n",
        "# শুরুর এরর প্রিন্ট করি \n",
        "print(error)\n",
        "\n",
        "# নতুন এরর প্রিন্ট করি \n",
        "print(error_updated)"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5\n",
            "2.5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sBmSDPfm7B__",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        },
        "outputId": "9d4c1158-9f32-4476-fdf0-9c6368226f49"
      },
      "source": [
        "import numpy as np\n",
        "\n",
        "# আমাদের প্রেডিকশন করার জন্য ডেটা পয়েন্ট, ছবির সাথে মিলিয়ে দেখুন\n",
        "input_data = np.array([0, 3])\n",
        "\n",
        "# স্যাম্পল ওয়েট যা পাল্টে দিয়েছি আমরা  \n",
        "weights_0 = {'node_0': [2, 1],\n",
        "             'node_1': [1, 2],\n",
        "             'output': [1, 1]\n",
        "            }\n",
        "\n",
        "# আসল টার্গেট ভ্যালু, এরর বের করার জন্য লাগবে \n",
        "target_actual = 3\n",
        "\n",
        "# দুটো মেথড ডিফাইন করি \n",
        "def relu(input):\n",
        "    output = max(0, input)\n",
        "    return output\n",
        "\n",
        "def predict_with_network(input_data_row, weights):\n",
        "    node_0_input = (input_data_row * weights['node_0']).sum()\n",
        "    # print(node_0_input)\n",
        "    node_0_output = relu(node_0_input)\n",
        "    # print(node_0_output)\n",
        "\n",
        "    node_1_input = (input_data_row * weights['node_1']).sum()\n",
        "    node_1_output = relu(node_1_input)\n",
        "    \n",
        "    hidden_layer_outputs = np.array([node_0_output, node_1_output])\n",
        "    \n",
        "    input_to_final_layer = (hidden_layer_outputs * weights['output']).sum()\n",
        "    model_output = relu(input_to_final_layer)\n",
        "    return model_output\n",
        "       \n",
        "\n",
        "# শুরুর ওয়েট দিয়ে প্রেডিকশন করি \n",
        "model_output_0 = predict_with_network(input_data, weights_0)\n",
        "\n",
        "# এরর ক্যালকুলেট করি: error_0\n",
        "error_0 = model_output_0 - target_actual\n",
        "\n",
        "# নতুন ওয়েট দেই যাতে টার্গেট প্রেডিকশন (3) ধরতে পারে: weights_1\n",
        "weights_1 = {'node_0': [2, 1],\n",
        "             'node_1': [1, 2],\n",
        "             'output': [1, 0]\n",
        "            }\n",
        "\n",
        "# নতুন ওয়েট দিয়ে প্রেডিকশন: model_output_1\n",
        "model_output_1 = predict_with_network(input_data, weights_1)\n",
        "\n",
        "# আবার এরর ক্যালকুলেট করি: error_1\n",
        "error_1 = model_output_1 - target_actual\n",
        "\n",
        "# সবকিছু প্রিন্ট করে দেখি \n",
        "\n",
        "print(model_output_0)\n",
        "print(model_output_1)\n",
        "print(error_0)\n",
        "print(error_1)"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "9\n",
            "3\n",
            "6\n",
            "0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bNaR0dxG9MFW",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        },
        "outputId": "82ce5d31-8242-48f5-a935-eb50f7096f29"
      },
      "source": [
        "import numpy as np\n",
        "\n",
        "# আমাদের প্রেডিকশন করার জন্য ডেটা পয়েন্ট, ছবির সাথে মিলিয়ে দেখুন\n",
        "input_data = np.array([-1, 2])\n",
        "\n",
        "# আমাদের ডিকশনারী\n",
        "weights =   {'node_0': np.array([3, 3]),\n",
        "             'node_1': np.array([1, 5]),\n",
        "             'output': np.array([2, -1])\n",
        "            }\n",
        "\n",
        "def relu(input):\n",
        "    '''রেল্যু ফাংশনকে ডিফাইন করে দিচ্ছি এখানে'''\n",
        "    # ইনপুটে যা পাবো সেটাকে ম্যাক্সিমাম যা আসবে, অথবা ঋনাত্বক আসলে \"০\" : output\n",
        "    output = max(0, input)\n",
        "    \n",
        "    # Return the value just calculated\n",
        "    return(output)\n",
        "\n",
        "# নোড ১ এর ভ্যালু ক্যালকুলেট করি: node_0_output\n",
        "node_0_input = (input_data * weights['node_0']).sum()\n",
        "node_0_output = relu(node_0_input)\n",
        "\n",
        "# নোড ২ এর ভ্যালু ক্যালকুলেট করি: node_1_output\n",
        "node_1_input = (input_data * weights['node_1']).sum()\n",
        "node_1_output = relu(node_1_input)\n",
        "\n",
        "# নতুন ভ্যালুগুলোকে অ্যারেতে বসাই: hidden_layer_outputs\n",
        "hidden_layer_outputs = np.array([node_0_output, node_1_output])\n",
        "\n",
        "# মডেলের আউটপুট ক্যালকুলেট করি (রেল্যুকে সরাসরি ব্যবহার না করে)\n",
        "model_output = (hidden_layer_outputs * weights['output']).sum()\n",
        "\n",
        "# Print model output\n",
        "print(node_0_output)\n",
        "print(node_1_output)\n",
        "print(hidden_layer_outputs)\n",
        "print(model_output)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "3\n",
            "9\n",
            "[3 9]\n",
            "-3\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}