liquidMNIST.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": [
        "**MNIST Dataset Over a Liquid Neural Network**\n",
        "\n",
        "**assembled by babycommando, 2024**\n"
      ],
      "metadata": {
        "id": "NbenCMqlewah"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install ncps"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "iwvy3DlCaHA8",
        "outputId": "34afb531-522a-4cb1-b602-e3e65d0156c2"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Collecting ncps\n",
            "  Downloading ncps-1.0.0-py3-none-any.whl (59 kB)\n",
            "\u001b[?25l     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m0.0/59.1 kB\u001b[0m \u001b[31m?\u001b[0m eta \u001b[36m-:--:--\u001b[0m\r\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m59.1/59.1 kB\u001b[0m \u001b[31m2.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hRequirement already satisfied: future in /usr/local/lib/python3.10/dist-packages (from ncps) (0.18.3)\n",
            "Requirement already satisfied: packaging in /usr/local/lib/python3.10/dist-packages (from ncps) (24.1)\n",
            "Installing collected packages: ncps\n",
            "Successfully installed ncps-1.0.0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "eVRH8mqgZruK"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow.keras.layers import Dense, Flatten, Reshape\n",
        "from tensorflow.keras.datasets import mnist\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "from ncps.tf import LTC\n",
        "from ncps.wirings import AutoNCP\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Load and preprocess the MNIST dataset\n",
        "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
        "x_train = x_train / 255.0\n",
        "x_test = x_test / 255.0\n",
        "y_train = to_categorical(y_train, 10)\n",
        "y_test = to_categorical(y_test, 10)\n",
        "\n",
        "# Reshape the data to fit the model input requirements\n",
        "x_train = x_train.reshape(-1, 28, 28, 1)\n",
        "x_test = x_test.reshape(-1, 28, 28, 1)"
      ],
      "metadata": {
        "id": "9_GZuAgJZtwb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Define the wiring for the NCP with 30 neurons and 10 outputs\n",
        "wiring = AutoNCP(30, 10)\n",
        "\n",
        "# Define the model with the LTC layer\n",
        "model = tf.keras.models.Sequential([\n",
        "    tf.keras.layers.InputLayer(input_shape=(28, 28, 1)),  # input shape based on the MNIST images\n",
        "    Reshape((28, 28)),  # Reshape input to match expected input for LTC\n",
        "    LTC(wiring, return_sequences=False),  # define a single Liquid Time-Constant layer\n",
        "    Dense(10, activation='softmax')  # output layer for classification\n",
        "])\n"
      ],
      "metadata": {
        "id": "F9NMC6RnZvHf"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Compile the model\n",
        "model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])"
      ],
      "metadata": {
        "id": "0CzPL0LlZ21b"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Train the model\n",
        "model.fit(x_train, y_train, epochs=5, batch_size=64, validation_split=0.2)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4oR1-mLPZ31a",
        "outputId": "a80835df-bed9-4d1e-9c04-d8fdbab57b2d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/5\n",
            "750/750 [==============================] - 78s 94ms/step - loss: 1.7385 - accuracy: 0.3789 - val_loss: 1.1722 - val_accuracy: 0.5886\n",
            "Epoch 2/5\n",
            "750/750 [==============================] - 69s 92ms/step - loss: 1.0499 - accuracy: 0.6390 - val_loss: 0.8855 - val_accuracy: 0.6913\n",
            "Epoch 3/5\n",
            "750/750 [==============================] - 69s 92ms/step - loss: 0.8658 - accuracy: 0.6974 - val_loss: 0.7791 - val_accuracy: 0.7166\n",
            "Epoch 4/5\n",
            "750/750 [==============================] - 71s 94ms/step - loss: 0.7880 - accuracy: 0.7247 - val_loss: 0.7328 - val_accuracy: 0.7296\n",
            "Epoch 5/5\n",
            "750/750 [==============================] - 72s 97ms/step - loss: 0.7382 - accuracy: 0.7426 - val_loss: 0.6754 - val_accuracy: 0.7577\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.src.callbacks.History at 0x79ba8c128370>"
            ]
          },
          "metadata": {},
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Evaluate the model\n",
        "test_loss, test_acc = model.evaluate(x_test, y_test, verbose=2)\n",
        "print(f'Test accuracy: {test_acc}')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rr6TcZa6Z4mS",
        "outputId": "14c664cf-0219-4a40-83d3-b92bdea67a3f"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 - 7s - loss: 0.6984 - accuracy: 0.7576 - 7s/epoch - 24ms/step\n",
            "Test accuracy: 0.7576000094413757\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Inference on a single image\n",
        "def infer_single_image(model, image, true_label):\n",
        "    image = image.reshape(1, 28, 28, 1)\n",
        "    prediction = model.predict(image)\n",
        "    predicted_label = np.argmax(prediction)\n",
        "\n",
        "    plt.figure()\n",
        "    plt.imshow(image.reshape(28, 28), cmap='gray')\n",
        "    plt.title(f'Predicted: {predicted_label}, True: {true_label}')\n",
        "    plt.axis('off')\n",
        "    plt.show()\n",
        "\n",
        "# Select an image from the test set\n",
        "index = 9  # You can change this index to test other images\n",
        "test_image = x_test[index]\n",
        "true_label = np.argmax(y_test[index])\n",
        "\n",
        "# Perform inference and display the result\n",
        "infer_single_image(model, test_image, true_label)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 446
        },
        "id": "tTfzmIpieaC9",
        "outputId": "a63f4189-e1f8-4ba9-b7ac-b08dce47c8ad"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1/1 [==============================] - 0s 35ms/step\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    }
  ]
}