keypoint_classification_color.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "id": "igMyGnjE9hEp"
   },
   "outputs": [],
   "source": [
    "import csv\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "RANDOM_SEED = 42"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "t2HDvhIu9hEr"
   },
   "source": [
    "# Specify each path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "9NvZP2Zn9hEy"
   },
   "outputs": [],
   "source": [
    "dataset = 'model/keypoint_classifier_color/color.csv'\n",
    "model_save_path = 'model/keypoint_classifier_color/color.keras'\n",
    "tflite_save_path = 'model/keypoint_classifier_color/color.tflite'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "s5oMH7x19hEz"
   },
   "source": [
    "# Set number of classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "id": "du4kodXL9hEz"
   },
   "outputs": [],
   "source": [
    "NUM_CLASSES = 3"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "XjnL0uso9hEz"
   },
   "source": [
    "# Dataset reading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "id": "QT5ZqtEz9hE0"
   },
   "outputs": [],
   "source": [
    "X_dataset = np.loadtxt(dataset, delimiter=',', dtype='float32', usecols=list(range(1, 4 + 1)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "id": "QmoKFsp49hE0"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2301, 4)\n",
      "(2301,)\n",
      "[[ 57.27238   74.362854 112.1581     0.      ]\n",
      " [ 56.41714   73.17333  110.53048    0.      ]\n",
      " [ 58.575237  75.809525 113.75238    0.      ]\n",
      " ...\n",
      " [ 95.67809   39.286667  14.447619   0.      ]\n",
      " [ 96.13571   40.184284  16.093334   0.      ]\n",
      " [123.10762   52.218094  20.336191   0.      ]]\n",
      "[1 1 1 ... 2 2 2]\n"
     ]
    }
   ],
   "source": [
    "y_dataset = np.loadtxt(dataset, delimiter=',', dtype='int32', usecols=(0))\n",
    "print(X_dataset.shape)\n",
    "print(y_dataset.shape)\n",
    "print(X_dataset)\n",
    "print(y_dataset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "id": "xQU7JTZ_9hE0"
   },
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X_dataset, y_dataset, train_size=0.75, random_state=RANDOM_SEED)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "mxK_lETT9hE0"
   },
   "source": [
    "# Model building"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "id": "vHBmUf1t9hE1"
   },
   "outputs": [],
   "source": [
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.layers.Input((4, )), #number of additional para\n",
    "    tf.keras.layers.Dropout(0.2),\n",
    "    tf.keras.layers.Dense(20, activation='relu'),\n",
    "    tf.keras.layers.Dropout(0.4),\n",
    "    tf.keras.layers.Dense(10, activation='relu'),\n",
    "    tf.keras.layers.Dense(NUM_CLASSES, activation='softmax')\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "ypqky9tc9hE1",
    "outputId": "5db082bb-30e3-4110-bf63-a1ee777ecd46"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " dropout (Dropout)           (None, 4)                 0         \n",
      "                                                                 \n",
      " dense (Dense)               (None, 20)                100       \n",
      "                                                                 \n",
      " dropout_1 (Dropout)         (None, 20)                0         \n",
      "                                                                 \n",
      " dense_1 (Dense)             (None, 10)                210       \n",
      "                                                                 \n",
      " dense_2 (Dense)             (None, 3)                 33        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 343 (1.34 KB)\n",
      "Trainable params: 343 (1.34 KB)\n",
      "Non-trainable params: 0 (0.00 Byte)\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()  # tf.keras.utils.plot_model(model, show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "id": "MbMjOflQ9hE1"
   },
   "outputs": [],
   "source": [
    "# Model checkpoint callback\n",
    "cp_callback = tf.keras.callbacks.ModelCheckpoint(\n",
    "    model_save_path, verbose=1, save_weights_only=False)\n",
    "# Callback for early stopping\n",
    "es_callback = tf.keras.callbacks.EarlyStopping(patience=20, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "id": "c3Dac0M_9hE2"
   },
   "outputs": [],
   "source": [
    "# Model compilation\n",
    "model.compile(\n",
    "    optimizer='adam',\n",
    "    loss='sparse_categorical_crossentropy',\n",
    "    metrics=['accuracy']\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "7XI0j1Iu9hE2"
   },
   "source": [
    "# Model training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "WirBl-JE9hE3",
    "outputId": "71b30ca2-8294-4d9d-8aa2-800d90d399de",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1000\n",
      " 1/14 [=>............................] - ETA: 9s - loss: 14.9036 - accuracy: 0.2891\n",
      "Epoch 1: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 1s 21ms/step - loss: 14.6653 - accuracy: 0.2765 - val_loss: 5.3734 - val_accuracy: 0.3368\n",
      "Epoch 2/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 9.9935 - accuracy: 0.3516\n",
      "Epoch 2: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 9.4274 - accuracy: 0.3629 - val_loss: 1.8046 - val_accuracy: 0.6476\n",
      "Epoch 3/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 7.0946 - accuracy: 0.3750\n",
      "Epoch 3: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 6.4401 - accuracy: 0.3994 - val_loss: 0.4759 - val_accuracy: 0.6632\n",
      "Epoch 4/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 5.0007 - accuracy: 0.3984\n",
      "Epoch 4: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 6ms/step - loss: 4.8223 - accuracy: 0.4609 - val_loss: 0.3561 - val_accuracy: 0.9931\n",
      "Epoch 5/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 3.3329 - accuracy: 0.5391\n",
      "Epoch 5: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 3.7991 - accuracy: 0.5061 - val_loss: 0.3299 - val_accuracy: 0.9983\n",
      "Epoch 6/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 3.3933 - accuracy: 0.4453\n",
      "Epoch 6: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 3.0340 - accuracy: 0.5368 - val_loss: 0.3352 - val_accuracy: 0.9983\n",
      "Epoch 7/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 2.5016 - accuracy: 0.5469\n",
      "Epoch 7: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 2.3530 - accuracy: 0.5809 - val_loss: 0.3328 - val_accuracy: 0.9965\n",
      "Epoch 8/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 2.5339 - accuracy: 0.5469\n",
      "Epoch 8: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 6ms/step - loss: 2.1533 - accuracy: 0.6000 - val_loss: 0.3308 - val_accuracy: 0.9983\n",
      "Epoch 9/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 2.3974 - accuracy: 0.6484\n",
      "Epoch 9: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 2.0820 - accuracy: 0.6017 - val_loss: 0.3250 - val_accuracy: 0.9983\n",
      "Epoch 10/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 2.1913 - accuracy: 0.6016\n",
      "Epoch 10: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 1.7850 - accuracy: 0.6093 - val_loss: 0.3257 - val_accuracy: 0.9983\n",
      "Epoch 11/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 2.1263 - accuracy: 0.5703\n",
      "Epoch 11: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 1.6748 - accuracy: 0.6116 - val_loss: 0.3218 - val_accuracy: 0.9983\n",
      "Epoch 12/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.5783 - accuracy: 0.6094\n",
      "Epoch 12: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 1.4899 - accuracy: 0.6220 - val_loss: 0.2882 - val_accuracy: 0.9861\n",
      "Epoch 13/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.0945 - accuracy: 0.6719\n",
      "Epoch 13: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 1.3476 - accuracy: 0.6406 - val_loss: 0.2654 - val_accuracy: 0.9861\n",
      "Epoch 14/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.2765 - accuracy: 0.6250\n",
      "Epoch 14: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 7ms/step - loss: 1.3127 - accuracy: 0.6278 - val_loss: 0.2522 - val_accuracy: 0.9913\n",
      "Epoch 15/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.3471 - accuracy: 0.7031\n",
      "Epoch 15: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 7ms/step - loss: 1.1784 - accuracy: 0.6452 - val_loss: 0.2515 - val_accuracy: 0.9948\n",
      "Epoch 16/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.1579 - accuracy: 0.5938\n",
      "Epoch 16: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 6ms/step - loss: 1.1178 - accuracy: 0.6446 - val_loss: 0.2649 - val_accuracy: 0.9983\n",
      "Epoch 17/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.8941 - accuracy: 0.7188\n",
      "Epoch 17: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 1.0386 - accuracy: 0.6672 - val_loss: 0.2608 - val_accuracy: 0.9983\n",
      "Epoch 18/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.8332 - accuracy: 0.6719\n",
      "Epoch 18: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.9106 - accuracy: 0.6800 - val_loss: 0.2526 - val_accuracy: 0.9983\n",
      "Epoch 19/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.2596 - accuracy: 0.6328\n",
      "Epoch 19: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 1.0052 - accuracy: 0.6800 - val_loss: 0.2709 - val_accuracy: 0.9983\n",
      "Epoch 20/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.8715 - accuracy: 0.7188\n",
      "Epoch 20: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 0.9134 - accuracy: 0.6951 - val_loss: 0.2650 - val_accuracy: 0.9983\n",
      "Epoch 21/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.8996 - accuracy: 0.6797\n",
      "Epoch 21: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 0.8761 - accuracy: 0.6754 - val_loss: 0.2689 - val_accuracy: 0.9983\n",
      "Epoch 22/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.8442 - accuracy: 0.7109\n",
      "Epoch 22: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.8843 - accuracy: 0.6765 - val_loss: 0.2835 - val_accuracy: 0.9983\n",
      "Epoch 23/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.7922 - accuracy: 0.7188\n",
      "Epoch 23: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.8899 - accuracy: 0.6817 - val_loss: 0.2712 - val_accuracy: 0.9983\n",
      "Epoch 24/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.7617 - accuracy: 0.7031\n",
      "Epoch 24: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.7874 - accuracy: 0.7113 - val_loss: 0.2808 - val_accuracy: 0.9983\n",
      "Epoch 25/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.9894 - accuracy: 0.7734\n",
      "Epoch 25: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 0.8713 - accuracy: 0.7107 - val_loss: 0.2807 - val_accuracy: 0.9983\n",
      "Epoch 26/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.6627 - accuracy: 0.7344\n",
      "Epoch 26: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 0.8356 - accuracy: 0.7038 - val_loss: 0.2840 - val_accuracy: 0.9983\n",
      "Epoch 27/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 1.0089 - accuracy: 0.6797\n",
      "Epoch 27: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 0.7726 - accuracy: 0.7322 - val_loss: 0.2830 - val_accuracy: 0.9983\n",
      "Epoch 28/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.7263 - accuracy: 0.7031\n",
      "Epoch 28: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 5ms/step - loss: 0.7902 - accuracy: 0.7177 - val_loss: 0.2781 - val_accuracy: 0.9983\n",
      "Epoch 29/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.5863 - accuracy: 0.7188\n",
      "Epoch 29: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.7352 - accuracy: 0.7328 - val_loss: 0.2893 - val_accuracy: 1.0000\n",
      "Epoch 30/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.5451 - accuracy: 0.7812\n",
      "Epoch 30: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.7308 - accuracy: 0.7438 - val_loss: 0.2755 - val_accuracy: 1.0000\n",
      "Epoch 31/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.6996 - accuracy: 0.7578\n",
      "Epoch 31: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.7582 - accuracy: 0.7345 - val_loss: 0.2846 - val_accuracy: 1.0000\n",
      "Epoch 32/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.8025 - accuracy: 0.6953\n",
      "Epoch 32: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.7647 - accuracy: 0.7159 - val_loss: 0.3032 - val_accuracy: 1.0000\n",
      "Epoch 33/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.7572 - accuracy: 0.7500\n",
      "Epoch 33: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.6728 - accuracy: 0.7443 - val_loss: 0.2901 - val_accuracy: 1.0000\n",
      "Epoch 34/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.6257 - accuracy: 0.7422\n",
      "Epoch 34: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.7156 - accuracy: 0.7345 - val_loss: 0.2756 - val_accuracy: 1.0000\n",
      "Epoch 35/1000\n",
      " 1/14 [=>............................] - ETA: 0s - loss: 0.5358 - accuracy: 0.7891\n",
      "Epoch 35: saving model to model/keypoint_classifier_color\\color.keras\n",
      "14/14 [==============================] - 0s 4ms/step - loss: 0.6980 - accuracy: 0.7519 - val_loss: 0.2726 - val_accuracy: 1.0000\n",
      "Epoch 35: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.History at 0x2a4912cf3d0>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    X_train,\n",
    "    y_train,\n",
    "    epochs=1000,\n",
    "    batch_size=128,\n",
    "    validation_data=(X_test, y_test),\n",
    "    callbacks=[cp_callback, es_callback]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "pxvb2Y299hE3",
    "outputId": "59eb3185-2e37-4b9e-bc9d-ab1b8ac29b7f"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5/5 [==============================] - 0s 1ms/step - loss: 0.2726 - accuracy: 1.0000\n"
     ]
    }
   ],
   "source": [
    "# Model evaluation\n",
    "val_loss, val_acc = model.evaluate(X_test, y_test, batch_size=128)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "id": "RBkmDeUW9hE4"
   },
   "outputs": [],
   "source": [
    "# Loading the saved model\n",
    "model = tf.keras.models.load_model(model_save_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "tFz9Tb0I9hE4",
    "outputId": "1c3b3528-54ae-4ee2-ab04-77429211cbef"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 58ms/step\n",
      "[0.16240916 0.82135236 0.01623841]\n",
      "1\n"
     ]
    }
   ],
   "source": [
    "# Inference test\n",
    "predict_result = model.predict(np.array([X_test[0]]))\n",
    "print(np.squeeze(predict_result))\n",
    "print(np.argmax(np.squeeze(predict_result)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "S3U4yNWx9hE4"
   },
   "source": [
    "# Confusion matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 582
    },
    "id": "AP1V6SCk9hE5",
    "outputId": "08e41a80-7a4a-4619-8125-ecc371368d19"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "18/18 [==============================] - 0s 1ms/step\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x600 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Classification Report\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       1.00      1.00      1.00       192\n",
      "           1       1.00      1.00      1.00       181\n",
      "           2       1.00      1.00      1.00       203\n",
      "\n",
      "    accuracy                           1.00       576\n",
      "   macro avg       1.00      1.00      1.00       576\n",
      "weighted avg       1.00      1.00      1.00       576\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.metrics import confusion_matrix, classification_report\n",
    "\n",
    "def print_confusion_matrix(y_true, y_pred, report=True):\n",
    "    labels = sorted(list(set(y_true)))\n",
    "    cmx_data = confusion_matrix(y_true, y_pred, labels=labels)\n",
    "    \n",
    "    df_cmx = pd.DataFrame(cmx_data, index=labels, columns=labels)\n",
    " \n",
    "    fig, ax = plt.subplots(figsize=(7, 6))\n",
    "    sns.heatmap(df_cmx, annot=True, fmt='g' ,square=False)\n",
    "    ax.set_ylim(len(set(y_true)), 0)\n",
    "    plt.show()\n",
    "    \n",
    "    if report:\n",
    "        print('Classification Report')\n",
    "        print(classification_report(y_test, y_pred))\n",
    "\n",
    "Y_pred = model.predict(X_test)\n",
    "y_pred = np.argmax(Y_pred, axis=1)\n",
    "\n",
    "print_confusion_matrix(y_test, y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "FNP6aqzc9hE5"
   },
   "source": [
    "# Convert to model for Tensorflow-Lite"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "id": "ODjnYyld9hE6"
   },
   "outputs": [],
   "source": [
    "# Save as a model dedicated to inference\n",
    "model.save(model_save_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "zRfuK8Y59hE6",
    "outputId": "a4ca585c-b5d5-4244-8291-8674063209bb"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\samra\\AppData\\Local\\Temp\\tmphw71zkda\\assets\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\samra\\AppData\\Local\\Temp\\tmphw71zkda\\assets\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "3460"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Transform model (quantization)\n",
    "\n",
    "converter = tf.lite.TFLiteConverter.from_keras_model(model)\n",
    "converter.optimizations = [tf.lite.Optimize.DEFAULT]\n",
    "tflite_quantized_model = converter.convert()\n",
    "\n",
    "open(tflite_save_path, 'wb').write(tflite_quantized_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "CHBPBXdx9hE6"
   },
   "source": [
    "# Inference test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "id": "mGAzLocO9hE7"
   },
   "outputs": [],
   "source": [
    "interpreter = tf.lite.Interpreter(model_path=tflite_save_path)\n",
    "interpreter.allocate_tensors()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "id": "oQuDK8YS9hE7"
   },
   "outputs": [],
   "source": [
    "# Get I / O tensor\n",
    "input_details = interpreter.get_input_details()\n",
    "output_details = interpreter.get_output_details()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "id": "2_ixAf_l9hE7"
   },
   "outputs": [],
   "source": [
    "interpreter.set_tensor(input_details[0]['index'], np.array([X_test[0]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "s4FoAnuc9hE7",
    "outputId": "91f18257-8d8b-4ef3-c558-e9b5f94fabbf",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 0 ns\n",
      "Wall time: 1e+03 µs\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# Inference implementation\n",
    "interpreter.invoke()\n",
    "tflite_results = interpreter.get_tensor(output_details[0]['index'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "vONjp19J9hE8",
    "outputId": "77205e24-fd00-42c4-f7b6-e06e527c2cba"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.16240902 0.82135254 0.01623843]\n",
      "1\n"
     ]
    }
   ],
   "source": [
    "print(np.squeeze(tflite_results))\n",
    "print(np.argmax(np.squeeze(tflite_results)))"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "keypoint_classification_EN.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}