handin_2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "UPNGQgO6aXtZ"
   },
   "source": [
    "# 2. házi feladat"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "BJ7TTytUaUwW"
   },
   "source": [
    "###**A**: Az adatbázis letöltése\n",
    "\n",
    "**A/1 - A/6**: A tesztelőszkript a számodra szükséges adatbázist tölti le és a részedre kiosztott feladatok helyességét fogja tesztelni. Ehhez az alábbi kódblokk tetején, a Tester példányosításakor paraméterként a Neptun-kódodat kell megadni egy sztringként, pl: \n",
    "`tester = Tester('DK7UAQ')`\n",
    "\n",
    "**Információ az adatbázisokról**:\n",
    "Az adatbázisok a UCI Machine Learning Repository-ról származnak.\n",
    "- **A/1**: Különböző vegyületekre (a vegyületek tulajdonságai alapján) próbáljuk megbecsülni azt a mennyiséget, ami az esetek 50%-ában megöli a _Pimephales promelas_ fajtájú halat. http://archive.ics.uci.edu/ml/datasets/QSAR+fish+toxicity\n",
    "- **A/2**: Két portugál középiskolában a tanulók évvégi matematika jegyét (0-20) próbáljuk megbecsülni a tanulók szociális és pénzügyi háttere alapján. http://archive.ics.uci.edu/ml/datasets/Student+Performance \n",
    "- **A/3**: Lásd A/2, matematika helyett portugál nyelv tárgyból.\n",
    "- **A/4**: Lásd A/2. Az A/2. és A/3.-ban használt adatbázisok esetében az input változók közt megtalálhatók a tanulók első- és második harmadévvégi osztályzata az adott tárgyból. Ezekkel nyilvánvalóan erősen korrelál az év végi jegy, így nehezíti a feladatot, ha ezeket az input változókat elhagyjuk. Az A/4 és A/5. adatbázisaiból ezek a változók hiányoznak, így jelentősen gyengébb eredmény várható.\n",
    "- **A/5**: Lásd A/4, matematika helyett portugál nyelv tárgyból.\n",
    "- **A/6**: Betontömbök nyomószilárdságát (~1-100 MPa) próbáljuk megbecsülni a betontömbök összetevőinek mennyiségéből, előállítási körülményeikből és korukból. http://archive.ics.uci.edu/ml/datasets/Concrete+Compressive+Strength\n",
    "\n",
    "A tesztelőszkript letölti a szöveges formában tárolt adatbázist és elhelyezi a `content` sztringben. Alább, kiírjuk a hosszát és az első 500 karaktert is a sztringből. Az adatbázis fájlok egy-egy sora adja meg az egy-egy mintaelemhez tartozó adatokat, ahol az utolsó érték a címke, míg a többi érték az input változók értékeit kódolja. Az adatbázis első sora a változók neveit tartalmazza.\n",
    "\n",
    "A feladatod, hogy alakítsd át a `content` sztringet az adatbázis input változóit és címkéit tartalmazó tömbbé. A tömbök típusa np.float32 (lebegőpontos) típusú legyen! Az input változók értékeit helyezd el a `features` (n_mintaelem, n_valtozo) alakú tömbben, míg a címkéket a `labels` (n_mintaelem,) alakú tömbben!\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "q_kfEC4sEMtw",
    "outputId": "250ecb44-06dd-4a9c-e4a8-6a4c7137e658"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tester: Dataset preparation OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "lines = content.split(\"\\n\")\n",
    "words = [line.split(\";\") for line in lines[1:-1]]    # ignore attribute names and last (empty) line\n",
    "\n",
    "vals = [[float(item) for item in rec] for rec in words]\n",
    "vals = np.array(vals, dtype=np.float32)\n",
    "\n",
    "splitted = np.hsplit(vals, (vals.shape[1] - 1,))\n",
    "features, labels = splitted[0], splitted[1].flatten()\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "KrE7NLJTR7Aq"
   },
   "source": [
    "## **B**: Tanító-, validációs- és teszthalmaz szétválasztása\n",
    "\n",
    "Keverd meg véletlenszerűen a mintaelemeket a `features` és `labels` tömbökben (természetesen a két tömb permutálása legyen azonos)! Ez azért szükséges, mert az adatbázisban az elemek lehet, hogy rendezettek pl. a címkék szerint. Megkeverés nélkül, a szétválasztott halmazokban jelentősen különbözne a címkék eloszlása. \n",
    "\n",
    "Ezt követően válaszd szét a tömböket tanító-, validációs- és teszthalmazra!\n",
    "\n",
    "A szétválasztás történjen az alábbi arányokkal:\n",
    "- **B/1**.: 70% / 15% / 15%\n",
    "- **B/2**: 60% / 20% / 20%\n",
    "- **B/3**: 50% / 25% / 25%\n",
    "\n",
    "A halmazokat tartalmazó három-három input- és címketömb változókat nevezd el `x_unnorm_train`, `x_unnorm_val`, `x_unnorm_test`, `y_train`, `y_val`, `y_test`-nek sorrendben."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "ls4niplvUgQ6",
    "outputId": "c2c8ece1-d8e0-48da-ac35-9c73d0d42560"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tester: Dataset split OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "# shuffle rows\n",
    "p = np.random.permutation(len(labels))\n",
    "features, labels = features[p], labels[p]\n",
    "\n",
    "# distribute elements\n",
    "n_samp = len(features)\n",
    "distr = (0.5, .25, .25)\n",
    "\n",
    "distr = tuple(np.add.accumulate(np.array(distr) * n_samp, dtype=np.int32) + 1)\n",
    "distr = np.insert(distr, 0, 0)  # indexing start at 0\n",
    "ind1, ind2, ind3 = (np.s_[distr[i]:distr[i+1]] for i in range(len(distr) - 1))  # ngrams\n",
    "\n",
    "x_unnorm_train, y_train = features[ind1], labels[ind1]\n",
    "\n",
    "x_unnorm_val, y_val = features[ind2], labels[ind2]\n",
    "\n",
    "x_unnorm_test, y_test = features[ind3], labels[ind3]\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "zjm2zVN92yGq"
   },
   "source": [
    "## **C**: Az input változók azonos nagyságrendre hozása\n",
    "\n",
    "A változók azonos nagyságrendre hozása az alábbi technikával történjen:\n",
    "- **C/1**: sztenderdizáció (0 átlag, 1 szórás)\n",
    "- **C/2**: min-max skálázás (0 minimum, 1 maximum)\n",
    "\n",
    "A változókat egymástól függetlenül kell skálázni. A skálázáshoz szükséges átlagot és szórást / minimumot és maximumot a tanítóhalmazon számold ki, majd alkalmazd őket a validációs és a teszt halmazra. Ez azért szükséges, mert a teszthalmazról azt feltételezzük, hogy ismeretlen a betanítás során, pontosan azért, hogy semmi esetre se tudjuk hozzáigazítani a tanított modellünket. \n",
    "\n",
    "**A feladatot *NumPy* segítségével kell megoldani. Ennek a részfeladatnak a megoldásához nem megengedett a *Keras* és a *TensorFlow* könyvtárnak és azok almoduljainak használata.**\n",
    "\n",
    "A skálázott halmazokat tartalmazó három változót nevezd el `x_train`, `x_val`, `x_test`-nek sorrendben!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "Q9SLPxkO3RZe",
    "outputId": "c6260fb0-bfb7-4629-b5c9-470a14d3fa7d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(32,)\n",
      "Tester: Feature rescale OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "per_feature_min = x_unnorm_train.min(axis=0)\n",
    "per_feature_max = x_unnorm_train.max(axis=0)\n",
    "per_feature_diff = (per_feature_max - per_feature_min)\n",
    "\n",
    "x_train = (x_unnorm_train - per_feature_min) / per_feature_diff\n",
    "\n",
    "x_val = (x_unnorm_val - per_feature_min) / per_feature_diff\n",
    "\n",
    "x_test = (x_unnorm_test - per_feature_min) / per_feature_diff\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "dohOMs4NlVXG"
   },
   "source": [
    "## **D**: A regressziós neuronháló definiálása Keras-ban \n",
    "\n",
    "Használd a `keras.models.Sequential()` osztályt a háló definiálásához! A háló architektúráját az alábbiak alapján alakítsd ki:\n",
    "- **D/1**: Kétrétegű háló, az első réteg neuronjainak száma 50, aktivációs függvénye ReLU. Az első réteg után alkalmazz dropout-ot neurononként 30%-os valószínűséggel!\n",
    "- **D/2**: Négyrétegű háló, a neuronok száma rendre 20, 20, 10, aktivációs függvényük ReLU. Az egyes rétegek közé tégy dropout rétegeket, melyek 20%-os valószínűséggel nulláznak ki egy-egy elemet!\n",
    "- **D/3**: Háromrétegű háló, a neuronok száma rendre 50, 30. Aktivációs függvényként használj sorban tanh-t, majd ReLU-t!\n",
    "\n",
    "Minden esetben csak teljesen összekötött (dense) rétegeket, valamint dropout rétegeket kell használnod. Az aktivációs függvényeket a teljesen összekötött rétegek `activation` paramétereként add meg (azaz most ne használj külön Acitvation rétegeket). A rétegek számába beleértendő a kimenetre képző, utolsó réteg is, melynek méretét a címke mérete határozza meg. Alkalmazd a tanult aktivációs függvényt az utolsó rétegen, függően a feladat típusától.\n",
    "A modellt fordítsd le a `compile()` függvény segítségével: a költségfüggvény legyen a szokásos, regresszióhoz használatos átlagos négyzetes eltérés (MSE), optimizer algoritmusként pedig alkalmazd az SGD algoritmust megfelelő tanulási rátával! \n",
    "A regressziós neuronhálót tartalmazó keras.models.Sequential() típusú változót nevezd el `reg_model`-nek!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "rAywpNOml4VL",
    "outputId": "c64a563c-358e-4eb9-81e7-96224c3da67d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tester: Regression model architecture OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "reg_model = tf.keras.models.Sequential()\n",
    "reg_model.add(tf.keras.layers.Dense(50, activation=\"tanh\", input_dim=x_train.shape[1]))\n",
    "reg_model.add(tf.keras.layers.Dense(30, activation=\"relu\"))\n",
    "reg_model.add(tf.keras.layers.Dense(1))\n",
    "\n",
    "reg_model.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=0.001),\n",
    "                  loss=\"mse\",\n",
    "                  metrics=[tf.keras.metrics.mean_absolute_error])\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Pz4fFtBUhUXV"
   },
   "source": [
    "## **E**: A háló betanítása regressziós feladatra\n",
    "\n",
    "Tanítsd be a neuronhálót a tanítóhalmazon! Használj early stopping-ot a validációs halmazzal! A tanulási rátát, epoch-ok maximális számát, a batch méretet, az early stopping `patience` paraméterét szabadon átállíthatod. Próbálgathatsz különböző kombinációkat, hogy jobb eredményt érj el.\n",
    "\n",
    "- Rajzold ki, hogyan alakult a tanulási és validációs költség a betanítás során! Ehhez használhatod a `matplotlib` könyvtárat, példát találsz a 8. előadás notebookjában. \n",
    "- A négyzetes (MSE) költség mellett az átlagos abszolút hiba (Mean Absolute Error, MAE) tanító- és validációs halmazon történő alakulását is rajzold ki egy másik grafikonon! A grafikonok y tengelyének megjelenített értékhatárait úgy állítsd be, hogy mindegyik idősor látható legyen és könnyen ki lehessen venni a költségek alakulását a tanítás vége felé is!\n",
    "- Számold ki a betanított modell négyzetes hibáját és az átlagos abszolút hibát a teszthalamzon, majd add értékül ezeket az értékeket a `test_mse` és `test_mae` változóknak!\n",
    "- Válassz néhány példát a teszthalmazból és becsülj hozzájuk címkét a betanított modellel, majd írasd ki a becsült és a megfelelő helyes címkéket! Így példákon is láthatjuk, hogy mennyire ad jó becsléseket a neuronhálónk.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "vQf90jJPkeUl",
    "outputId": "64631785-c1cc-4db1-c4e3-ff9383b57215"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/600\n",
      "21/21 [==============================] - 1s 11ms/step - loss: 57.2709 - mean_absolute_error: 6.3320 - val_loss: 9.9752 - val_mean_absolute_error: 2.3681\n",
      "Epoch 2/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 9.9567 - mean_absolute_error: 2.3402 - val_loss: 7.8427 - val_mean_absolute_error: 1.9933\n",
      "Epoch 3/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 9.1288 - mean_absolute_error: 2.2160 - val_loss: 7.4824 - val_mean_absolute_error: 1.9376\n",
      "Epoch 4/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 8.4297 - mean_absolute_error: 2.1157 - val_loss: 7.0830 - val_mean_absolute_error: 1.8691\n",
      "Epoch 5/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 7.8401 - mean_absolute_error: 2.0117 - val_loss: 6.5122 - val_mean_absolute_error: 1.7861\n",
      "Epoch 6/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 7.1874 - mean_absolute_error: 1.9352 - val_loss: 5.9271 - val_mean_absolute_error: 1.6862\n",
      "Epoch 7/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 6.7737 - mean_absolute_error: 1.8808 - val_loss: 5.6745 - val_mean_absolute_error: 1.6440\n",
      "Epoch 8/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 6.3553 - mean_absolute_error: 1.8119 - val_loss: 5.3044 - val_mean_absolute_error: 1.6158\n",
      "Epoch 9/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 5.9550 - mean_absolute_error: 1.7508 - val_loss: 4.8483 - val_mean_absolute_error: 1.5176\n",
      "Epoch 10/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 5.6172 - mean_absolute_error: 1.7178 - val_loss: 4.6038 - val_mean_absolute_error: 1.4862\n",
      "Epoch 11/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 5.2626 - mean_absolute_error: 1.6583 - val_loss: 4.3882 - val_mean_absolute_error: 1.4410\n",
      "Epoch 12/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 5.0654 - mean_absolute_error: 1.6403 - val_loss: 4.3527 - val_mean_absolute_error: 1.4884\n",
      "Epoch 13/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 4.7737 - mean_absolute_error: 1.5968 - val_loss: 4.6107 - val_mean_absolute_error: 1.5872\n",
      "Epoch 14/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 4.6519 - mean_absolute_error: 1.6017 - val_loss: 3.7223 - val_mean_absolute_error: 1.3430\n",
      "Epoch 15/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 4.2929 - mean_absolute_error: 1.5290 - val_loss: 3.5748 - val_mean_absolute_error: 1.3236\n",
      "Epoch 16/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 4.1077 - mean_absolute_error: 1.4905 - val_loss: 3.4183 - val_mean_absolute_error: 1.2957\n",
      "Epoch 17/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 3.9353 - mean_absolute_error: 1.4499 - val_loss: 3.2452 - val_mean_absolute_error: 1.2509\n",
      "Epoch 18/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 3.8027 - mean_absolute_error: 1.4368 - val_loss: 3.1116 - val_mean_absolute_error: 1.2340\n",
      "Epoch 19/600\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 3.6322 - mean_absolute_error: 1.4087 - val_loss: 2.9779 - val_mean_absolute_error: 1.1972\n",
      "Epoch 20/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 3.5046 - mean_absolute_error: 1.3623 - val_loss: 2.8924 - val_mean_absolute_error: 1.1905\n",
      "Epoch 21/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 3.3542 - mean_absolute_error: 1.3252 - val_loss: 2.8170 - val_mean_absolute_error: 1.1778\n",
      "Epoch 22/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 3.2537 - mean_absolute_error: 1.2984 - val_loss: 2.7101 - val_mean_absolute_error: 1.1162\n",
      "Epoch 23/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 3.1446 - mean_absolute_error: 1.2725 - val_loss: 2.6115 - val_mean_absolute_error: 1.1086\n",
      "Epoch 24/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 3.0264 - mean_absolute_error: 1.2520 - val_loss: 2.5485 - val_mean_absolute_error: 1.0755\n",
      "Epoch 25/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.9418 - mean_absolute_error: 1.2143 - val_loss: 2.4692 - val_mean_absolute_error: 1.0629\n",
      "Epoch 26/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.8318 - mean_absolute_error: 1.1819 - val_loss: 2.8191 - val_mean_absolute_error: 1.2417\n",
      "Epoch 27/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 2.7251 - mean_absolute_error: 1.1678 - val_loss: 2.3856 - val_mean_absolute_error: 1.0552\n",
      "Epoch 28/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.6180 - mean_absolute_error: 1.1490 - val_loss: 2.3349 - val_mean_absolute_error: 1.0263\n",
      "Epoch 29/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.5726 - mean_absolute_error: 1.1135 - val_loss: 2.3171 - val_mean_absolute_error: 1.0394\n",
      "Epoch 30/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.5204 - mean_absolute_error: 1.0965 - val_loss: 2.3216 - val_mean_absolute_error: 1.0479\n",
      "Epoch 31/600\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.4366 - mean_absolute_error: 1.0721 - val_loss: 2.2624 - val_mean_absolute_error: 0.9875\n",
      "Epoch 32/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 2.3522 - mean_absolute_error: 1.0423 - val_loss: 2.2858 - val_mean_absolute_error: 1.0396\n",
      "Epoch 33/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 2.3537 - mean_absolute_error: 1.0452 - val_loss: 2.5838 - val_mean_absolute_error: 1.0792\n",
      "Epoch 34/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.3761 - mean_absolute_error: 1.0483 - val_loss: 2.1699 - val_mean_absolute_error: 0.9860\n",
      "Epoch 35/600\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 2.2020 - mean_absolute_error: 0.9968 - val_loss: 2.1883 - val_mean_absolute_error: 1.0039\n",
      "Epoch 36/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 2.2005 - mean_absolute_error: 0.9994 - val_loss: 2.3546 - val_mean_absolute_error: 1.0764\n",
      "Epoch 37/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 2.1183 - mean_absolute_error: 0.9806 - val_loss: 2.1098 - val_mean_absolute_error: 0.9565\n",
      "Epoch 38/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.1044 - mean_absolute_error: 0.9627 - val_loss: 2.1992 - val_mean_absolute_error: 1.0149\n",
      "Epoch 39/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.0582 - mean_absolute_error: 0.9581 - val_loss: 2.1464 - val_mean_absolute_error: 0.9906\n",
      "Epoch 40/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.0429 - mean_absolute_error: 0.9508 - val_loss: 2.1464 - val_mean_absolute_error: 0.9902\n",
      "Epoch 41/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.0084 - mean_absolute_error: 0.9560 - val_loss: 2.1011 - val_mean_absolute_error: 0.9624\n",
      "Epoch 42/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 2.0383 - mean_absolute_error: 0.9687 - val_loss: 2.6629 - val_mean_absolute_error: 1.1264\n",
      "Epoch 43/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 2.0125 - mean_absolute_error: 0.9215 - val_loss: 2.1134 - val_mean_absolute_error: 0.9613\n",
      "Epoch 44/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.9855 - mean_absolute_error: 0.9399 - val_loss: 2.1922 - val_mean_absolute_error: 0.9813\n",
      "Epoch 45/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.9255 - mean_absolute_error: 0.9160 - val_loss: 2.0636 - val_mean_absolute_error: 0.9547\n",
      "Epoch 46/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.9070 - mean_absolute_error: 0.9150 - val_loss: 2.0651 - val_mean_absolute_error: 0.9525\n",
      "Epoch 47/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.9312 - mean_absolute_error: 0.9322 - val_loss: 2.1902 - val_mean_absolute_error: 1.0061\n",
      "Epoch 48/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.8935 - mean_absolute_error: 0.9131 - val_loss: 2.1016 - val_mean_absolute_error: 0.9620\n",
      "Epoch 49/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.8515 - mean_absolute_error: 0.9001 - val_loss: 2.0889 - val_mean_absolute_error: 0.9595\n",
      "Epoch 50/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.8222 - mean_absolute_error: 0.8817 - val_loss: 2.0593 - val_mean_absolute_error: 0.9608\n",
      "Epoch 51/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.8271 - mean_absolute_error: 0.9048 - val_loss: 2.2126 - val_mean_absolute_error: 0.9935\n",
      "Epoch 52/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.8275 - mean_absolute_error: 0.8908 - val_loss: 2.1257 - val_mean_absolute_error: 0.9710\n",
      "Epoch 53/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.8006 - mean_absolute_error: 0.8817 - val_loss: 2.0505 - val_mean_absolute_error: 0.9520\n",
      "Epoch 54/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7561 - mean_absolute_error: 0.8834 - val_loss: 2.0451 - val_mean_absolute_error: 0.9524\n",
      "Epoch 55/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7553 - mean_absolute_error: 0.8681 - val_loss: 2.2449 - val_mean_absolute_error: 1.0088\n",
      "Epoch 56/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.7730 - mean_absolute_error: 0.8740 - val_loss: 2.0575 - val_mean_absolute_error: 0.9613\n",
      "Epoch 57/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7558 - mean_absolute_error: 0.8655 - val_loss: 2.4726 - val_mean_absolute_error: 1.0704\n",
      "Epoch 58/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7596 - mean_absolute_error: 0.8525 - val_loss: 2.3456 - val_mean_absolute_error: 1.0597\n",
      "Epoch 59/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6930 - mean_absolute_error: 0.8590 - val_loss: 2.0696 - val_mean_absolute_error: 0.9589\n",
      "Epoch 60/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7209 - mean_absolute_error: 0.8697 - val_loss: 2.1682 - val_mean_absolute_error: 0.9986\n",
      "Epoch 61/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.7083 - mean_absolute_error: 0.8630 - val_loss: 2.0526 - val_mean_absolute_error: 0.9476\n",
      "Epoch 62/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6886 - mean_absolute_error: 0.8583 - val_loss: 2.0519 - val_mean_absolute_error: 0.9476\n",
      "Epoch 63/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7080 - mean_absolute_error: 0.8557 - val_loss: 2.0526 - val_mean_absolute_error: 0.9477\n",
      "Epoch 64/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6762 - mean_absolute_error: 0.8505 - val_loss: 2.0626 - val_mean_absolute_error: 0.9560\n",
      "Epoch 65/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6524 - mean_absolute_error: 0.8595 - val_loss: 2.3192 - val_mean_absolute_error: 1.0374\n",
      "Epoch 66/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6876 - mean_absolute_error: 0.8406 - val_loss: 2.0420 - val_mean_absolute_error: 0.9459\n",
      "Epoch 67/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.7455 - mean_absolute_error: 0.8574 - val_loss: 2.0694 - val_mean_absolute_error: 0.9602\n",
      "Epoch 68/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.6485 - mean_absolute_error: 0.8360 - val_loss: 2.0474 - val_mean_absolute_error: 0.9440\n",
      "Epoch 69/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6863 - mean_absolute_error: 0.8507 - val_loss: 2.0250 - val_mean_absolute_error: 0.9399\n",
      "Epoch 70/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6613 - mean_absolute_error: 0.8377 - val_loss: 2.0306 - val_mean_absolute_error: 0.9432\n",
      "Epoch 71/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6720 - mean_absolute_error: 0.8358 - val_loss: 2.0531 - val_mean_absolute_error: 0.9486\n",
      "Epoch 72/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6407 - mean_absolute_error: 0.8341 - val_loss: 2.1799 - val_mean_absolute_error: 1.0021\n",
      "Epoch 73/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6276 - mean_absolute_error: 0.8430 - val_loss: 2.0324 - val_mean_absolute_error: 0.9460\n",
      "Epoch 74/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6014 - mean_absolute_error: 0.8223 - val_loss: 2.0814 - val_mean_absolute_error: 0.9590\n",
      "Epoch 75/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6179 - mean_absolute_error: 0.8243 - val_loss: 2.0431 - val_mean_absolute_error: 0.9479\n",
      "Epoch 76/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.6304 - mean_absolute_error: 0.8428 - val_loss: 2.0171 - val_mean_absolute_error: 0.9404\n",
      "Epoch 77/600\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 1.6365 - mean_absolute_error: 0.8342 - val_loss: 2.0161 - val_mean_absolute_error: 0.9419\n",
      "Epoch 78/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6077 - mean_absolute_error: 0.8368 - val_loss: 2.0181 - val_mean_absolute_error: 0.9416\n",
      "Epoch 79/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6047 - mean_absolute_error: 0.8347 - val_loss: 2.1283 - val_mean_absolute_error: 0.9924\n",
      "Epoch 80/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6158 - mean_absolute_error: 0.8489 - val_loss: 2.0142 - val_mean_absolute_error: 0.9422\n",
      "Epoch 81/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.6114 - mean_absolute_error: 0.8330 - val_loss: 2.0912 - val_mean_absolute_error: 0.9600\n",
      "Epoch 82/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5660 - mean_absolute_error: 0.8092 - val_loss: 2.0038 - val_mean_absolute_error: 0.9362\n",
      "Epoch 83/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.5945 - mean_absolute_error: 0.8169 - val_loss: 3.2755 - val_mean_absolute_error: 1.3932\n",
      "Epoch 84/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.6860 - mean_absolute_error: 0.8575 - val_loss: 2.4085 - val_mean_absolute_error: 1.0817\n",
      "Epoch 85/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5728 - mean_absolute_error: 0.8244 - val_loss: 2.1857 - val_mean_absolute_error: 0.9856\n",
      "Epoch 86/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.5730 - mean_absolute_error: 0.8248 - val_loss: 2.0349 - val_mean_absolute_error: 0.9492\n",
      "Epoch 87/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5463 - mean_absolute_error: 0.8065 - val_loss: 2.0597 - val_mean_absolute_error: 0.9438\n",
      "Epoch 88/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5755 - mean_absolute_error: 0.8264 - val_loss: 2.0230 - val_mean_absolute_error: 0.9355\n",
      "Epoch 89/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5615 - mean_absolute_error: 0.8048 - val_loss: 2.0390 - val_mean_absolute_error: 0.9500\n",
      "Epoch 90/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.5444 - mean_absolute_error: 0.8028 - val_loss: 2.1681 - val_mean_absolute_error: 0.9784\n",
      "Epoch 91/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5224 - mean_absolute_error: 0.7996 - val_loss: 2.0522 - val_mean_absolute_error: 0.9588\n",
      "Epoch 92/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5266 - mean_absolute_error: 0.8163 - val_loss: 2.0833 - val_mean_absolute_error: 0.9714\n",
      "Epoch 93/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.5455 - mean_absolute_error: 0.8328 - val_loss: 2.0101 - val_mean_absolute_error: 0.9312\n",
      "Epoch 94/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5458 - mean_absolute_error: 0.8118 - val_loss: 2.0050 - val_mean_absolute_error: 0.9301\n",
      "Epoch 95/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.5174 - mean_absolute_error: 0.7888 - val_loss: 2.0535 - val_mean_absolute_error: 0.9627\n",
      "Epoch 96/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5659 - mean_absolute_error: 0.8276 - val_loss: 2.0504 - val_mean_absolute_error: 0.9609\n",
      "Epoch 97/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4877 - mean_absolute_error: 0.7866 - val_loss: 2.0373 - val_mean_absolute_error: 0.9393\n",
      "Epoch 98/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4617 - mean_absolute_error: 0.8001 - val_loss: 2.0011 - val_mean_absolute_error: 0.9415\n",
      "Epoch 99/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.5031 - mean_absolute_error: 0.8120 - val_loss: 2.0562 - val_mean_absolute_error: 0.9479\n",
      "Epoch 100/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5082 - mean_absolute_error: 0.8063 - val_loss: 2.2440 - val_mean_absolute_error: 1.0040\n",
      "Epoch 101/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4839 - mean_absolute_error: 0.8141 - val_loss: 2.0620 - val_mean_absolute_error: 0.9661\n",
      "Epoch 102/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.4773 - mean_absolute_error: 0.8039 - val_loss: 2.0548 - val_mean_absolute_error: 0.9431\n",
      "Epoch 103/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4765 - mean_absolute_error: 0.7809 - val_loss: 1.9661 - val_mean_absolute_error: 0.9248\n",
      "Epoch 104/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4568 - mean_absolute_error: 0.7829 - val_loss: 2.0361 - val_mean_absolute_error: 0.9383\n",
      "Epoch 105/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4757 - mean_absolute_error: 0.7869 - val_loss: 1.9779 - val_mean_absolute_error: 0.9235\n",
      "Epoch 106/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4898 - mean_absolute_error: 0.7893 - val_loss: 1.9860 - val_mean_absolute_error: 0.9234\n",
      "Epoch 107/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.5056 - mean_absolute_error: 0.8024 - val_loss: 2.0686 - val_mean_absolute_error: 0.9678\n",
      "Epoch 108/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.4450 - mean_absolute_error: 0.7809 - val_loss: 2.0355 - val_mean_absolute_error: 0.9574\n",
      "Epoch 109/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.4501 - mean_absolute_error: 0.7908 - val_loss: 1.9857 - val_mean_absolute_error: 0.9200\n",
      "Epoch 110/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.4909 - mean_absolute_error: 0.7894 - val_loss: 2.1367 - val_mean_absolute_error: 0.9651\n",
      "Epoch 111/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4793 - mean_absolute_error: 0.8056 - val_loss: 2.1148 - val_mean_absolute_error: 0.9586\n",
      "Epoch 112/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4743 - mean_absolute_error: 0.7812 - val_loss: 1.9975 - val_mean_absolute_error: 0.9438\n",
      "Epoch 113/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4215 - mean_absolute_error: 0.7797 - val_loss: 1.9949 - val_mean_absolute_error: 0.9273\n",
      "Epoch 114/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4085 - mean_absolute_error: 0.7972 - val_loss: 2.0120 - val_mean_absolute_error: 0.9276\n",
      "Epoch 115/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4291 - mean_absolute_error: 0.7865 - val_loss: 1.9719 - val_mean_absolute_error: 0.9197\n",
      "Epoch 116/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.3877 - mean_absolute_error: 0.7624 - val_loss: 2.4621 - val_mean_absolute_error: 1.1034\n",
      "Epoch 117/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.4518 - mean_absolute_error: 0.7971 - val_loss: 1.9918 - val_mean_absolute_error: 0.9253\n",
      "Epoch 118/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.3665 - mean_absolute_error: 0.7717 - val_loss: 2.0161 - val_mean_absolute_error: 0.9323\n",
      "Epoch 119/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.4324 - mean_absolute_error: 0.7901 - val_loss: 2.0232 - val_mean_absolute_error: 0.9337\n",
      "6/6 [==============================] - 0s 3ms/step - loss: 2.0107 - mean_absolute_error: 0.9814\n",
      "Pred. | Truth | Abs. err.\n",
      " 6.4  |   8.0 |  1.6\n",
      "15.5  |  15.0 |  0.5\n",
      "15.8  |  16.0 |  0.2\n",
      " 9.4  |   9.0 |  0.4\n",
      "18.2  |  17.0 |  1.2\n",
      " 8.9  |   9.0 |  0.1\n",
      "16.7  |  17.0 |  0.3\n",
      "14.1  |  16.0 |  1.9\n",
      "10.8  |  10.0 |  0.8\n",
      "10.5  |  10.0 |  0.5\n",
      "Tester: Regression model learning OK\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2304x1152 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "# training\n",
    "earlystopping_callback = tf.keras.callbacks.EarlyStopping(monitor=\"val_loss\", patience=16)\n",
    "\n",
    "history = reg_model.fit(x_train, y_train, validation_data=(x_val, y_val),\n",
    "              batch_size=16, epochs=600, verbose=1, callbacks=[earlystopping_callback])\n",
    "\n",
    "# metrics visualization\n",
    "tr_mse_losses = history.history['loss']\n",
    "val_mse_losses = history.history['val_loss']\n",
    "tr_mae = history.history['mean_absolute_error']\n",
    "val_mae = history.history['val_mean_absolute_error']\n",
    "\n",
    "# plots\n",
    "plt.figure(figsize=(32, 16))\n",
    "\n",
    "plt.subplot(3, 4, 1)\n",
    "plt.plot(tr_mse_losses, label=\"J_train\")\n",
    "plt.plot(val_mse_losses, label=\"J_val\")\n",
    "plt.ylim((0,5))\n",
    "plt.xlabel(\"Number of epochs\")\n",
    "plt.ylabel(\"Cost (J)\")\n",
    "plt.legend()\n",
    "plt.title(\"Mean Squared Error during training\")\n",
    "\n",
    "plt.subplot(3, 4, 2)\n",
    "plt.plot(tr_mae, label=\"J_train\")\n",
    "plt.plot(val_mae, label=\"J_val\")\n",
    "plt.ylim((0,5))\n",
    "plt.xlabel(\"Number of epochs\")\n",
    "plt.ylabel(\"Cost (J)\")\n",
    "plt.legend()\n",
    "plt.title(\"Mean Absolute Error during training\")\n",
    "\n",
    "# metrics\n",
    "test_mse, test_mae = reg_model.evaluate(x_test, y_test)\n",
    "\n",
    "\n",
    "# testing for a few, random values\n",
    "samp_size = 10\n",
    "rand_idxs = np.random.randint(len(x_test), size=samp_size)\n",
    "predictions = reg_model.predict(x_test[rand_idxs].reshape(samp_size,-1)).flatten()\n",
    "ground_t_labels = y_test[rand_idxs]\n",
    "\n",
    "print(\"Pred. | Truth | Abs. err.\")\n",
    "for (p, g) in zip(predictions, ground_t_labels):\n",
    "  print(f\"{p:4.1f}  |  {g:4.1f} | {abs(p - g):4.1f}\")\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "pTBqUmDzPiYI"
   },
   "source": [
    "## **F**: A regressziós adatbázis átalakítása klasszifikációs adatbázissá\n",
    "\n",
    "Alakítsd át a regressziós adatbázist, hogy alkalmas legyen \n",
    "klasszifikációs feladathoz! A kategóriákat az alábbiak alapján alakítsd ki : \n",
    "\n",
    "*   **F/1**: 3 kategória. A kategóriákat úgy alakítsd ki, hogy a regressziós adatbázis címkéinek intervallumán felveszel két küszöbértéket. Az első kategóriába a kisebb küszöbérték alatti, a másodikba a két küszöbérték közti és a harmadikba a nagyobb küszöbérték feletti címkével rendelkező mintaelemek kerüljenek. \n",
    "Tehát, például ha a regressziós adatbázis címkéi 3 és 8 közt vannak, akkor a két küszöbértéknek választhatjuk az 5.2-t és a 6.1-et. Ekkor az 1. kategóriába az 5.2-nél kisebb címkéjű elemek, a 2. kategóriába az 5.2 és 6.1 közti címkéjű elemek és a 3. kategóriába a 6.1-nél nagyobb címkéjű elemek kerülnek. Azonos címkéjű elemek mindenképpen azonos kategóriába kell, hogy kerüljenek.\n",
    "\n",
    "*   **F/2**: 4 kategória. A kategóriákat úgy alakítsd ki, hogy a regressziós adatbázis címkéinek intervallumán felveszel három küszöbértéket. Az első kategóriába a kisebb küszöbérték alatti, a másodikba az első és második küszöbérték közti, a harmadikba a második és a haramdik küszöbérték közti, és a negyedikbe a harmadik küszöbérték feletti címkével rendelkező mintaelemek kerüljenek. \n",
    "Tehát, például ha a regressziós adatbázis címkéi 3 és 8 közt vannak, akkor a három küszöbértéknek választhatjuk a 3.8-at, a 6.1-et és a 7.5-öt. Ekkor az 1. kategóriába az 3.8-nál kisebb címkéjű elemek, a 2. kategóriába az 3.8 és 6.1 közti címkéjű elemek, a 3. kategóriába az 6.1 és 7.5 közti címkéjű elemek és a 4. kategóriába a 7.5-nél nagyobb címkéjű elemek kerülnek. Azonos címkéjű elemek mindenképpen azonos kategóriába kell, hogy kerüljenek.\n",
    "\n",
    "\n",
    "Válaszd ki a küszöbértékeket úgy, hogy a kategóriábákba **megközelítőleg** ugyanannyi elem kerüljön! Hozd létre az új kategóriacímkéket az `y_train`, `y_val`, `y_test` tömbökből. Kerüljenek ezek a `y_cat_train`, `y_cat_val`, `y_cat_test` változókba!\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 245,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "kCdOectqRsvz",
    "outputId": "4e75132e-830f-463c-cdfb-859fee03a47a"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tester: Classification dataset creation OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "cat_count = 3\n",
    "ideal_group_size = len(y_train) / cat_count\n",
    "ideal_group_counts = ideal_group_size * np.arange(1,cat_count)\n",
    "\n",
    "uniqes, counts = np.unique(y_train, return_counts=True)\n",
    "\n",
    "# lowest difference from ideal divison point\n",
    "div_idx = np.argmin(np.absolute(np.add.accumulate(counts) - ideal_group_counts[:,None]), axis=1)\n",
    "\n",
    "# matrix to compare which category a value falls into\n",
    "div_vals = uniqes[div_idx]\n",
    "div_vals_lower = np.insert(div_vals, 0, 0)\n",
    "div_vals_upper = np.append(div_vals, uniqes[-1])\n",
    "\n",
    "# used to categorize every value, based on the divison points\n",
    "# calculated from the 'train' dataset (y_train)\n",
    "def categorize(mx): return np.argmax( (div_vals_lower[:,None] < mx) &\n",
    "                                      (mx <= div_vals_upper[:,None]),\n",
    "                                    axis=0 )\n",
    "\n",
    "# categorize\n",
    "to_categorize = (y_train, y_val, y_test)\n",
    "y_cat_train, y_cat_val, y_cat_test = (categorize(mx) for mx in to_categorize)\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "z_OwgkhPxCuA"
   },
   "source": [
    "## **G**: Kategóriacímkék átalakítása one-hot kódolásra\n",
    "\n",
    "Kettőnél több kategóriás (multi-class) klasszifikációnál minden kategóriához egy valószínűséget becslünk, azaz a háló kimenete akkora méretű, ahány kategóriánk van. Így tehát az igazi címkéink ugyanekkora méretű one-hot kódolású vektorok lesznek, ezeket címkevektorokként használva fogjuk betanítani a hálót.\n",
    "\n",
    "Készítsd el a one-hot kódolású címkevektor tömböket a kategóriacímkéket tartalmazó `y_cat_train`, `y_cat_val`, `y_cat_test` tömbökből! Az új tömbök kerüljenek a `y_onehot_train`, `y_onehot_val`, `y_onehot_test` változókba! A tömbök alakja így (n_samples, n_categories), típusuk legyen `np.float32`!\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 246,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "iLAlR3qyxT13",
    "outputId": "0008d239-8443-42ca-bba1-a7e8bcca9b6d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tester: One-hot conversion OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "from keras.utils.np_utils import to_categorical   \n",
    "\n",
    "# encode\n",
    "to_encode = (y_cat_train, y_cat_val, y_cat_test)\n",
    "y_onehot_train, y_onehot_val, y_onehot_test = (to_categorical(mx, cat_count) for mx in to_encode)\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "9xdtHfyrwexF"
   },
   "source": [
    "## **H**: A klasszifikációs neuronháló definiálása Keras-ban \n",
    "\n",
    "Használd a `keras.models.Sequential()` osztályt a háló definiálásához! A háló architektúráját az alábbiak alapján alakítsd ki:\n",
    "- **H/1**: Kétrétegű háló, az első réteg neuronjainak száma 50, aktivációs függvénye ReLU. Az első réteg után alkalmazz dropout-ot neurononként 30%-os valószínűséggel!\n",
    "- **H/2**: Négyrétegű háló, a neuronok száma rendre 20, 20, 10, aktivációs függvényük ReLU. Az egyes rétegek közé tégy dropout rétegeket, melyek 20%-os valószínűséggel nulláznak ki egy-egy elemet!\n",
    "- **H/3**: Háromrétegű háló, a neuronok száma rendre 50, 30. Aktivációs függvényként használj sorban tanh-t, majd ReLU-t!\n",
    "\n",
    "Minden esetben csak teljesen összekötött (dense) rétegeket, valamint dropout rétegeket kell használnod. Az aktivációs függvényeket a telejesen összekötött rétegek `activation` paramétereként add meg (azaz most ne használj külön Acitvation rétegeket). A rétegek számába beleértendő a kimenetre képző, utolsó réteg is, melynek méretét a címke mérete határozza meg. Alkalmazd a tanult aktivációs függvényt az utolsó rétegen, függően a feladat típusától!\n",
    "A modellt fordítsd le a `compile()` függvény segítségével: a költségfüggvény legyen a szokásos, multi-class klasszifikációhoz használatos (categorical) crossentropy, optimizer algoritmusként pedig alkalmazd az SGD algoritmust megfelelő tanulási rátával! \n",
    "A klasszifikációs neuronhálót tartalmazó keras.models.Sequential() típusú változót nevezd el `cl_model`-nek. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 355,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "30Iwot3xxVXI",
    "outputId": "e1b4996f-640a-49e2-88c5-4e3a82139344"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tester: Classification model architecture OK\n"
     ]
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "cl_model = tf.keras.models.Sequential()\n",
    "cl_model.add(tf.keras.layers.Dense(50, activation=\"tanh\", input_dim=x_train.shape[1]))\n",
    "cl_model.add(tf.keras.layers.Dense(30, activation=\"relu\"))\n",
    "cl_model.add(tf.keras.layers.Dense(cat_count, activation=\"softmax\"))\n",
    "\n",
    "cl_model.compile(optimizer=tf.keras.optimizers.SGD(learning_rate=0.008),\n",
    "                 loss=\"categorical_crossentropy\",\n",
    "                 metrics=[tf.keras.metrics.CategoricalAccuracy()])\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "7uowunGxa2Mq"
   },
   "source": [
    "## **I**: A háló betanítása (multi-class) klasszifikációs feladatra\n",
    "\n",
    "Tanítsd be a neuronhálót a tanítóhalmazon! Használj early stopping-ot a validációs halmazzal! A tanulási rátát, epoch-ok maximális számát, a batch méretet, az early stopping `patience` paraméterét szabadon átállíthatod. Próbálgathatsz különböző kombinációkat, hogy jobb eredményt érj el.\n",
    "\n",
    "- Rajzold ki, hogyan alakult a tanulási és validációs költség a betanítás során! Ehhez használhatod a `matplotlib` könyvtárat, példát találsz a 8. előadás notebookjában. \n",
    "- A crossentropy költség mellett, a helyesen klasszifikált mintaelemek arányának (_accuracy_) tanító- és validációs halmazon történő alakulását is rajzold ki egy másik grafikonon! A grafikonok y tengelyének megjelenített értékhatárait úgy állítsd be, hogy mindegyik idősor látható legyen és könnyen ki lehessen venni a költségek alakulását a tanítás vége felé is!\n",
    "- Számold ki a betanított modell crossentropy költségét és az accuracy metrikát a teszthalamzon, majd add értékül ezeket az értékeket a `test_ce` és `test_acc` változóknak!\n",
    "- Számold ki a multiclass _Fals pozitív arány_ (FPR) és _Fals negatív arány_ (FNR) metrikákat a teszthalmazon a betanított modelledre, majd add értékül ezeket az értékeket a `test_fpr` és `test_fnr` változóknak! Az FPR és FNR kiszámításához az első házi feladat B/1. és B/2. feladataiban található képletek és leírás segíthetnek. A **saját**, 1. házi feladatban beadott FPR vagy FNR metrikát kiszámító kódodat természetesen felhasználhatod itt is."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 356,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "LHWZ8AN2oGoP",
    "outputId": "5a5533f2-502e-4c28-966d-e44ffeddffa2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/600\n",
      "21/21 [==============================] - 1s 11ms/step - loss: 1.1121 - categorical_accuracy: 0.3508 - val_loss: 1.0756 - val_categorical_accuracy: 0.4321\n",
      "Epoch 2/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.0997 - categorical_accuracy: 0.3662 - val_loss: 1.0678 - val_categorical_accuracy: 0.4321\n",
      "Epoch 3/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.0880 - categorical_accuracy: 0.3754 - val_loss: 1.0572 - val_categorical_accuracy: 0.4506\n",
      "Epoch 4/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.0782 - categorical_accuracy: 0.3938 - val_loss: 1.0557 - val_categorical_accuracy: 0.4630\n",
      "Epoch 5/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.0681 - categorical_accuracy: 0.4031 - val_loss: 1.0452 - val_categorical_accuracy: 0.4753\n",
      "Epoch 6/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.0580 - categorical_accuracy: 0.4246 - val_loss: 1.0416 - val_categorical_accuracy: 0.5062\n",
      "Epoch 7/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.0477 - categorical_accuracy: 0.4277 - val_loss: 1.0380 - val_categorical_accuracy: 0.5000\n",
      "Epoch 8/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.0398 - categorical_accuracy: 0.4554 - val_loss: 1.0287 - val_categorical_accuracy: 0.5247\n",
      "Epoch 9/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 1.0294 - categorical_accuracy: 0.4738 - val_loss: 1.0181 - val_categorical_accuracy: 0.5123\n",
      "Epoch 10/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.0197 - categorical_accuracy: 0.4862 - val_loss: 1.0164 - val_categorical_accuracy: 0.5432\n",
      "Epoch 11/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.0107 - categorical_accuracy: 0.4923 - val_loss: 1.0167 - val_categorical_accuracy: 0.5494\n",
      "Epoch 12/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 1.0025 - categorical_accuracy: 0.4923 - val_loss: 1.0089 - val_categorical_accuracy: 0.5617\n",
      "Epoch 13/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9951 - categorical_accuracy: 0.5262 - val_loss: 1.0008 - val_categorical_accuracy: 0.5741\n",
      "Epoch 14/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9852 - categorical_accuracy: 0.5415 - val_loss: 0.9870 - val_categorical_accuracy: 0.5062\n",
      "Epoch 15/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9775 - categorical_accuracy: 0.5108 - val_loss: 0.9850 - val_categorical_accuracy: 0.5741\n",
      "Epoch 16/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9692 - categorical_accuracy: 0.5446 - val_loss: 0.9778 - val_categorical_accuracy: 0.5741\n",
      "Epoch 17/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9617 - categorical_accuracy: 0.5354 - val_loss: 0.9689 - val_categorical_accuracy: 0.5556\n",
      "Epoch 18/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.9539 - categorical_accuracy: 0.5354 - val_loss: 0.9652 - val_categorical_accuracy: 0.5864\n",
      "Epoch 19/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9465 - categorical_accuracy: 0.5354 - val_loss: 0.9616 - val_categorical_accuracy: 0.5679\n",
      "Epoch 20/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.9385 - categorical_accuracy: 0.5323 - val_loss: 0.9603 - val_categorical_accuracy: 0.5679\n",
      "Epoch 21/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.9316 - categorical_accuracy: 0.5600 - val_loss: 0.9478 - val_categorical_accuracy: 0.5741\n",
      "Epoch 22/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.9249 - categorical_accuracy: 0.5508 - val_loss: 0.9424 - val_categorical_accuracy: 0.5556\n",
      "Epoch 23/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.9158 - categorical_accuracy: 0.5754 - val_loss: 0.9373 - val_categorical_accuracy: 0.5926\n",
      "Epoch 24/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.9094 - categorical_accuracy: 0.5692 - val_loss: 0.9319 - val_categorical_accuracy: 0.5679\n",
      "Epoch 25/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.9029 - categorical_accuracy: 0.5815 - val_loss: 0.9245 - val_categorical_accuracy: 0.5864\n",
      "Epoch 26/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8953 - categorical_accuracy: 0.5846 - val_loss: 0.9175 - val_categorical_accuracy: 0.5926\n",
      "Epoch 27/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8887 - categorical_accuracy: 0.5815 - val_loss: 0.9203 - val_categorical_accuracy: 0.5370\n",
      "Epoch 28/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8828 - categorical_accuracy: 0.5938 - val_loss: 0.9058 - val_categorical_accuracy: 0.6049\n",
      "Epoch 29/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.8758 - categorical_accuracy: 0.5877 - val_loss: 0.9031 - val_categorical_accuracy: 0.5802\n",
      "Epoch 30/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.8689 - categorical_accuracy: 0.6123 - val_loss: 0.8982 - val_categorical_accuracy: 0.5802\n",
      "Epoch 31/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8644 - categorical_accuracy: 0.5969 - val_loss: 0.8966 - val_categorical_accuracy: 0.5494\n",
      "Epoch 32/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8577 - categorical_accuracy: 0.6154 - val_loss: 0.8895 - val_categorical_accuracy: 0.5926\n",
      "Epoch 33/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.8512 - categorical_accuracy: 0.6092 - val_loss: 0.8836 - val_categorical_accuracy: 0.6111\n",
      "Epoch 34/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8469 - categorical_accuracy: 0.5969 - val_loss: 0.8820 - val_categorical_accuracy: 0.5926\n",
      "Epoch 35/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8405 - categorical_accuracy: 0.6185 - val_loss: 0.8831 - val_categorical_accuracy: 0.5988\n",
      "Epoch 36/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8354 - categorical_accuracy: 0.6154 - val_loss: 0.8787 - val_categorical_accuracy: 0.5617\n",
      "Epoch 37/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.8309 - categorical_accuracy: 0.6246 - val_loss: 0.8778 - val_categorical_accuracy: 0.5494\n",
      "Epoch 38/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8254 - categorical_accuracy: 0.6246 - val_loss: 0.8817 - val_categorical_accuracy: 0.5802\n",
      "Epoch 39/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8218 - categorical_accuracy: 0.6277 - val_loss: 0.8721 - val_categorical_accuracy: 0.5679\n",
      "Epoch 40/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8154 - categorical_accuracy: 0.6092 - val_loss: 0.8697 - val_categorical_accuracy: 0.6049\n",
      "Epoch 41/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.8121 - categorical_accuracy: 0.6308 - val_loss: 0.8617 - val_categorical_accuracy: 0.5741\n",
      "Epoch 42/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.8055 - categorical_accuracy: 0.6338 - val_loss: 0.8544 - val_categorical_accuracy: 0.5864\n",
      "Epoch 43/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.8018 - categorical_accuracy: 0.6246 - val_loss: 0.8526 - val_categorical_accuracy: 0.5617\n",
      "Epoch 44/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7956 - categorical_accuracy: 0.6369 - val_loss: 0.8473 - val_categorical_accuracy: 0.5864\n",
      "Epoch 45/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7917 - categorical_accuracy: 0.6215 - val_loss: 0.8419 - val_categorical_accuracy: 0.6111\n",
      "Epoch 46/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7877 - categorical_accuracy: 0.6277 - val_loss: 0.8381 - val_categorical_accuracy: 0.5988\n",
      "Epoch 47/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7813 - categorical_accuracy: 0.6308 - val_loss: 0.8461 - val_categorical_accuracy: 0.6173\n",
      "Epoch 48/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7786 - categorical_accuracy: 0.6308 - val_loss: 0.8298 - val_categorical_accuracy: 0.6049\n",
      "Epoch 49/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7746 - categorical_accuracy: 0.6431 - val_loss: 0.8275 - val_categorical_accuracy: 0.6173\n",
      "Epoch 50/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7689 - categorical_accuracy: 0.6523 - val_loss: 0.8219 - val_categorical_accuracy: 0.6049\n",
      "Epoch 51/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7640 - categorical_accuracy: 0.6492 - val_loss: 0.8202 - val_categorical_accuracy: 0.6111\n",
      "Epoch 52/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7606 - categorical_accuracy: 0.6338 - val_loss: 0.8185 - val_categorical_accuracy: 0.6111\n",
      "Epoch 53/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7562 - categorical_accuracy: 0.6431 - val_loss: 0.8210 - val_categorical_accuracy: 0.5741\n",
      "Epoch 54/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7531 - categorical_accuracy: 0.6308 - val_loss: 0.8243 - val_categorical_accuracy: 0.6481\n",
      "Epoch 55/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7479 - categorical_accuracy: 0.6554 - val_loss: 0.8195 - val_categorical_accuracy: 0.6543\n",
      "Epoch 56/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7436 - categorical_accuracy: 0.6400 - val_loss: 0.8141 - val_categorical_accuracy: 0.6235\n",
      "Epoch 57/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7393 - categorical_accuracy: 0.6646 - val_loss: 0.8209 - val_categorical_accuracy: 0.6358\n",
      "Epoch 58/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7348 - categorical_accuracy: 0.6523 - val_loss: 0.8141 - val_categorical_accuracy: 0.6481\n",
      "Epoch 59/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7311 - categorical_accuracy: 0.6431 - val_loss: 0.8042 - val_categorical_accuracy: 0.6173\n",
      "Epoch 60/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7279 - categorical_accuracy: 0.6400 - val_loss: 0.7985 - val_categorical_accuracy: 0.6049\n",
      "Epoch 61/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7247 - categorical_accuracy: 0.6492 - val_loss: 0.7989 - val_categorical_accuracy: 0.6173\n",
      "Epoch 62/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7218 - categorical_accuracy: 0.6646 - val_loss: 0.8011 - val_categorical_accuracy: 0.6358\n",
      "Epoch 63/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7134 - categorical_accuracy: 0.6615 - val_loss: 0.7960 - val_categorical_accuracy: 0.6605\n",
      "Epoch 64/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.7103 - categorical_accuracy: 0.6677 - val_loss: 0.7934 - val_categorical_accuracy: 0.6605\n",
      "Epoch 65/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7053 - categorical_accuracy: 0.6708 - val_loss: 0.7839 - val_categorical_accuracy: 0.6358\n",
      "Epoch 66/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.7026 - categorical_accuracy: 0.6585 - val_loss: 0.7905 - val_categorical_accuracy: 0.6667\n",
      "Epoch 67/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6988 - categorical_accuracy: 0.6738 - val_loss: 0.7904 - val_categorical_accuracy: 0.6543\n",
      "Epoch 68/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6932 - categorical_accuracy: 0.6892 - val_loss: 0.7786 - val_categorical_accuracy: 0.6296\n",
      "Epoch 69/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6909 - categorical_accuracy: 0.6769 - val_loss: 0.7662 - val_categorical_accuracy: 0.6481\n",
      "Epoch 70/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6876 - categorical_accuracy: 0.6738 - val_loss: 0.7795 - val_categorical_accuracy: 0.6728\n",
      "Epoch 71/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6817 - categorical_accuracy: 0.6892 - val_loss: 0.7617 - val_categorical_accuracy: 0.6173\n",
      "Epoch 72/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6798 - categorical_accuracy: 0.6800 - val_loss: 0.7742 - val_categorical_accuracy: 0.6728\n",
      "Epoch 73/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6729 - categorical_accuracy: 0.6800 - val_loss: 0.7568 - val_categorical_accuracy: 0.6420\n",
      "Epoch 74/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6687 - categorical_accuracy: 0.6985 - val_loss: 0.7628 - val_categorical_accuracy: 0.6543\n",
      "Epoch 75/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6645 - categorical_accuracy: 0.6892 - val_loss: 0.7643 - val_categorical_accuracy: 0.6852\n",
      "Epoch 76/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6614 - categorical_accuracy: 0.6892 - val_loss: 0.7551 - val_categorical_accuracy: 0.6728\n",
      "Epoch 77/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6577 - categorical_accuracy: 0.6954 - val_loss: 0.7551 - val_categorical_accuracy: 0.6728\n",
      "Epoch 78/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6540 - categorical_accuracy: 0.7015 - val_loss: 0.7519 - val_categorical_accuracy: 0.6296\n",
      "Epoch 79/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6495 - categorical_accuracy: 0.7046 - val_loss: 0.7461 - val_categorical_accuracy: 0.6728\n",
      "Epoch 80/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6436 - categorical_accuracy: 0.7108 - val_loss: 0.7442 - val_categorical_accuracy: 0.6481\n",
      "Epoch 81/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6420 - categorical_accuracy: 0.7108 - val_loss: 0.7455 - val_categorical_accuracy: 0.6667\n",
      "Epoch 82/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6362 - categorical_accuracy: 0.7200 - val_loss: 0.7522 - val_categorical_accuracy: 0.6481\n",
      "Epoch 83/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6371 - categorical_accuracy: 0.7077 - val_loss: 0.7374 - val_categorical_accuracy: 0.6543\n",
      "Epoch 84/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6309 - categorical_accuracy: 0.6954 - val_loss: 0.7659 - val_categorical_accuracy: 0.6481\n",
      "Epoch 85/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6291 - categorical_accuracy: 0.7077 - val_loss: 0.7449 - val_categorical_accuracy: 0.6790\n",
      "Epoch 86/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6245 - categorical_accuracy: 0.7446 - val_loss: 0.7412 - val_categorical_accuracy: 0.6605\n",
      "Epoch 87/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6210 - categorical_accuracy: 0.7108 - val_loss: 0.7365 - val_categorical_accuracy: 0.6667\n",
      "Epoch 88/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6151 - categorical_accuracy: 0.7323 - val_loss: 0.7371 - val_categorical_accuracy: 0.6790\n",
      "Epoch 89/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6118 - categorical_accuracy: 0.7323 - val_loss: 0.7702 - val_categorical_accuracy: 0.6481\n",
      "Epoch 90/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.6114 - categorical_accuracy: 0.7262 - val_loss: 0.7376 - val_categorical_accuracy: 0.6852\n",
      "Epoch 91/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.6042 - categorical_accuracy: 0.7262 - val_loss: 0.7276 - val_categorical_accuracy: 0.6667\n",
      "Epoch 92/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5981 - categorical_accuracy: 0.7292 - val_loss: 0.7416 - val_categorical_accuracy: 0.6605\n",
      "Epoch 93/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5976 - categorical_accuracy: 0.7292 - val_loss: 0.7405 - val_categorical_accuracy: 0.6667\n",
      "Epoch 94/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5922 - categorical_accuracy: 0.7354 - val_loss: 0.7193 - val_categorical_accuracy: 0.6728\n",
      "Epoch 95/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5902 - categorical_accuracy: 0.7354 - val_loss: 0.7309 - val_categorical_accuracy: 0.6667\n",
      "Epoch 96/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5867 - categorical_accuracy: 0.7323 - val_loss: 0.7188 - val_categorical_accuracy: 0.6728\n",
      "Epoch 97/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5824 - categorical_accuracy: 0.7415 - val_loss: 0.7152 - val_categorical_accuracy: 0.6605\n",
      "Epoch 98/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5787 - categorical_accuracy: 0.7415 - val_loss: 0.7115 - val_categorical_accuracy: 0.6790\n",
      "Epoch 99/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5742 - categorical_accuracy: 0.7446 - val_loss: 0.7095 - val_categorical_accuracy: 0.6852\n",
      "Epoch 100/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5699 - categorical_accuracy: 0.7631 - val_loss: 0.7052 - val_categorical_accuracy: 0.6790\n",
      "Epoch 101/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5680 - categorical_accuracy: 0.7477 - val_loss: 0.7046 - val_categorical_accuracy: 0.6790\n",
      "Epoch 102/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5635 - categorical_accuracy: 0.7600 - val_loss: 0.7020 - val_categorical_accuracy: 0.6790\n",
      "Epoch 103/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5619 - categorical_accuracy: 0.7415 - val_loss: 0.7080 - val_categorical_accuracy: 0.6852\n",
      "Epoch 104/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5581 - categorical_accuracy: 0.7662 - val_loss: 0.7141 - val_categorical_accuracy: 0.6728\n",
      "Epoch 105/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5549 - categorical_accuracy: 0.7508 - val_loss: 0.7001 - val_categorical_accuracy: 0.6790\n",
      "Epoch 106/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5500 - categorical_accuracy: 0.7569 - val_loss: 0.7007 - val_categorical_accuracy: 0.6914\n",
      "Epoch 107/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5484 - categorical_accuracy: 0.7600 - val_loss: 0.6942 - val_categorical_accuracy: 0.6914\n",
      "Epoch 108/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5450 - categorical_accuracy: 0.7877 - val_loss: 0.6914 - val_categorical_accuracy: 0.6914\n",
      "Epoch 109/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5415 - categorical_accuracy: 0.7754 - val_loss: 0.6807 - val_categorical_accuracy: 0.7037\n",
      "Epoch 110/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5414 - categorical_accuracy: 0.7908 - val_loss: 0.6954 - val_categorical_accuracy: 0.6852\n",
      "Epoch 111/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5320 - categorical_accuracy: 0.7877 - val_loss: 0.7166 - val_categorical_accuracy: 0.6914\n",
      "Epoch 112/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5299 - categorical_accuracy: 0.7754 - val_loss: 0.6999 - val_categorical_accuracy: 0.6914\n",
      "Epoch 113/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5305 - categorical_accuracy: 0.7815 - val_loss: 0.7097 - val_categorical_accuracy: 0.6728\n",
      "Epoch 114/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5264 - categorical_accuracy: 0.8000 - val_loss: 0.7127 - val_categorical_accuracy: 0.6852\n",
      "Epoch 115/600\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.5224 - categorical_accuracy: 0.8000 - val_loss: 0.6922 - val_categorical_accuracy: 0.6852\n",
      "Epoch 116/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5176 - categorical_accuracy: 0.7908 - val_loss: 0.6894 - val_categorical_accuracy: 0.6852\n",
      "Epoch 117/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5145 - categorical_accuracy: 0.8000 - val_loss: 0.6908 - val_categorical_accuracy: 0.6914\n",
      "Epoch 118/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5126 - categorical_accuracy: 0.7938 - val_loss: 0.7070 - val_categorical_accuracy: 0.6852\n",
      "Epoch 119/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5093 - categorical_accuracy: 0.8062 - val_loss: 0.6920 - val_categorical_accuracy: 0.6667\n",
      "Epoch 120/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5085 - categorical_accuracy: 0.7938 - val_loss: 0.6865 - val_categorical_accuracy: 0.6914\n",
      "Epoch 121/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.5068 - categorical_accuracy: 0.7938 - val_loss: 0.6826 - val_categorical_accuracy: 0.7037\n",
      "Epoch 122/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.5042 - categorical_accuracy: 0.8123 - val_loss: 0.6810 - val_categorical_accuracy: 0.7037\n",
      "Epoch 123/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4990 - categorical_accuracy: 0.8185 - val_loss: 0.7095 - val_categorical_accuracy: 0.6667\n",
      "Epoch 124/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4944 - categorical_accuracy: 0.8246 - val_loss: 0.6976 - val_categorical_accuracy: 0.6914\n",
      "Epoch 125/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4975 - categorical_accuracy: 0.8123 - val_loss: 0.6971 - val_categorical_accuracy: 0.6790\n",
      "Epoch 126/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4920 - categorical_accuracy: 0.8062 - val_loss: 0.6935 - val_categorical_accuracy: 0.6852\n",
      "Epoch 127/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4893 - categorical_accuracy: 0.8123 - val_loss: 0.6814 - val_categorical_accuracy: 0.6728\n",
      "Epoch 128/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4842 - categorical_accuracy: 0.8246 - val_loss: 0.6831 - val_categorical_accuracy: 0.6852\n",
      "Epoch 129/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4824 - categorical_accuracy: 0.8246 - val_loss: 0.6759 - val_categorical_accuracy: 0.6914\n",
      "Epoch 130/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4782 - categorical_accuracy: 0.8185 - val_loss: 0.6772 - val_categorical_accuracy: 0.6728\n",
      "Epoch 131/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4767 - categorical_accuracy: 0.8154 - val_loss: 0.6780 - val_categorical_accuracy: 0.6852\n",
      "Epoch 132/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4761 - categorical_accuracy: 0.8308 - val_loss: 0.6948 - val_categorical_accuracy: 0.6852\n",
      "Epoch 133/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4731 - categorical_accuracy: 0.8400 - val_loss: 0.6821 - val_categorical_accuracy: 0.6975\n",
      "Epoch 134/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4709 - categorical_accuracy: 0.8400 - val_loss: 0.6759 - val_categorical_accuracy: 0.6914\n",
      "Epoch 135/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4672 - categorical_accuracy: 0.8246 - val_loss: 0.6778 - val_categorical_accuracy: 0.6852\n",
      "Epoch 136/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4679 - categorical_accuracy: 0.8277 - val_loss: 0.6722 - val_categorical_accuracy: 0.6790\n",
      "Epoch 137/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4617 - categorical_accuracy: 0.8338 - val_loss: 0.6784 - val_categorical_accuracy: 0.6852\n",
      "Epoch 138/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4640 - categorical_accuracy: 0.8246 - val_loss: 0.6789 - val_categorical_accuracy: 0.6790\n",
      "Epoch 139/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4588 - categorical_accuracy: 0.8431 - val_loss: 0.7121 - val_categorical_accuracy: 0.6420\n",
      "Epoch 140/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4557 - categorical_accuracy: 0.8492 - val_loss: 0.6788 - val_categorical_accuracy: 0.6914\n",
      "Epoch 141/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4551 - categorical_accuracy: 0.8308 - val_loss: 0.6676 - val_categorical_accuracy: 0.6667\n",
      "Epoch 142/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4507 - categorical_accuracy: 0.8308 - val_loss: 0.6925 - val_categorical_accuracy: 0.6790\n",
      "Epoch 143/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4501 - categorical_accuracy: 0.8338 - val_loss: 0.6771 - val_categorical_accuracy: 0.6975\n",
      "Epoch 144/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4534 - categorical_accuracy: 0.8277 - val_loss: 0.6882 - val_categorical_accuracy: 0.6914\n",
      "Epoch 145/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4442 - categorical_accuracy: 0.8431 - val_loss: 0.6677 - val_categorical_accuracy: 0.6790\n",
      "Epoch 146/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4475 - categorical_accuracy: 0.8308 - val_loss: 0.6710 - val_categorical_accuracy: 0.7037\n",
      "Epoch 147/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4417 - categorical_accuracy: 0.8431 - val_loss: 0.7113 - val_categorical_accuracy: 0.6358\n",
      "Epoch 148/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4397 - categorical_accuracy: 0.8431 - val_loss: 0.6669 - val_categorical_accuracy: 0.6728\n",
      "Epoch 149/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4427 - categorical_accuracy: 0.8462 - val_loss: 0.6845 - val_categorical_accuracy: 0.6975\n",
      "Epoch 150/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4356 - categorical_accuracy: 0.8492 - val_loss: 0.6922 - val_categorical_accuracy: 0.7099\n",
      "Epoch 151/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4333 - categorical_accuracy: 0.8400 - val_loss: 0.6926 - val_categorical_accuracy: 0.6852\n",
      "Epoch 152/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4337 - categorical_accuracy: 0.8462 - val_loss: 0.6865 - val_categorical_accuracy: 0.6852\n",
      "Epoch 153/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4304 - categorical_accuracy: 0.8400 - val_loss: 0.6916 - val_categorical_accuracy: 0.6790\n",
      "Epoch 154/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4278 - categorical_accuracy: 0.8523 - val_loss: 0.6869 - val_categorical_accuracy: 0.6975\n",
      "Epoch 155/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4244 - categorical_accuracy: 0.8431 - val_loss: 0.7436 - val_categorical_accuracy: 0.6605\n",
      "Epoch 156/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4304 - categorical_accuracy: 0.8400 - val_loss: 0.6994 - val_categorical_accuracy: 0.6481\n",
      "Epoch 157/600\n",
      "21/21 [==============================] - 0s 6ms/step - loss: 0.4280 - categorical_accuracy: 0.8492 - val_loss: 0.6611 - val_categorical_accuracy: 0.6914\n",
      "Epoch 158/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4281 - categorical_accuracy: 0.8369 - val_loss: 0.6669 - val_categorical_accuracy: 0.7099\n",
      "Epoch 159/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4237 - categorical_accuracy: 0.8369 - val_loss: 0.6723 - val_categorical_accuracy: 0.6914\n",
      "Epoch 160/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4200 - categorical_accuracy: 0.8400 - val_loss: 0.6743 - val_categorical_accuracy: 0.6852\n",
      "Epoch 161/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4213 - categorical_accuracy: 0.8338 - val_loss: 0.6682 - val_categorical_accuracy: 0.6790\n",
      "Epoch 162/600\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.4148 - categorical_accuracy: 0.8462 - val_loss: 0.6818 - val_categorical_accuracy: 0.6914\n",
      "Epoch 163/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4131 - categorical_accuracy: 0.8585 - val_loss: 0.7205 - val_categorical_accuracy: 0.6543\n",
      "Epoch 164/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4185 - categorical_accuracy: 0.8400 - val_loss: 0.6680 - val_categorical_accuracy: 0.6914\n",
      "Epoch 165/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4121 - categorical_accuracy: 0.8431 - val_loss: 0.6846 - val_categorical_accuracy: 0.6852\n",
      "Epoch 166/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4068 - categorical_accuracy: 0.8492 - val_loss: 0.6717 - val_categorical_accuracy: 0.7037\n",
      "Epoch 167/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4098 - categorical_accuracy: 0.8523 - val_loss: 0.6746 - val_categorical_accuracy: 0.7037\n",
      "Epoch 168/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.4099 - categorical_accuracy: 0.8492 - val_loss: 0.7239 - val_categorical_accuracy: 0.6605\n",
      "Epoch 169/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4060 - categorical_accuracy: 0.8523 - val_loss: 0.6619 - val_categorical_accuracy: 0.6975\n",
      "Epoch 170/600\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.4064 - categorical_accuracy: 0.8431 - val_loss: 0.6724 - val_categorical_accuracy: 0.7037\n",
      "Epoch 171/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3988 - categorical_accuracy: 0.8585 - val_loss: 0.6793 - val_categorical_accuracy: 0.7037\n",
      "Epoch 172/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.4026 - categorical_accuracy: 0.8431 - val_loss: 0.6823 - val_categorical_accuracy: 0.7037\n",
      "Epoch 173/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3994 - categorical_accuracy: 0.8554 - val_loss: 0.6776 - val_categorical_accuracy: 0.7037\n",
      "Epoch 174/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3962 - categorical_accuracy: 0.8585 - val_loss: 0.6729 - val_categorical_accuracy: 0.6914\n",
      "Epoch 175/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3919 - categorical_accuracy: 0.8523 - val_loss: 0.6954 - val_categorical_accuracy: 0.6852\n",
      "Epoch 176/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3935 - categorical_accuracy: 0.8554 - val_loss: 0.6940 - val_categorical_accuracy: 0.6790\n",
      "Epoch 177/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3941 - categorical_accuracy: 0.8585 - val_loss: 0.6845 - val_categorical_accuracy: 0.6975\n",
      "Epoch 178/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3935 - categorical_accuracy: 0.8554 - val_loss: 0.6769 - val_categorical_accuracy: 0.6975\n",
      "Epoch 179/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3945 - categorical_accuracy: 0.8554 - val_loss: 0.6676 - val_categorical_accuracy: 0.7037\n",
      "Epoch 180/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3895 - categorical_accuracy: 0.8523 - val_loss: 0.7120 - val_categorical_accuracy: 0.6667\n",
      "Epoch 181/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3865 - categorical_accuracy: 0.8554 - val_loss: 0.6691 - val_categorical_accuracy: 0.7099\n",
      "Epoch 182/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3888 - categorical_accuracy: 0.8554 - val_loss: 0.6852 - val_categorical_accuracy: 0.7099\n",
      "Epoch 183/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3867 - categorical_accuracy: 0.8554 - val_loss: 0.6809 - val_categorical_accuracy: 0.7037\n",
      "Epoch 184/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3791 - categorical_accuracy: 0.8615 - val_loss: 0.6921 - val_categorical_accuracy: 0.7037\n",
      "Epoch 185/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3869 - categorical_accuracy: 0.8523 - val_loss: 0.6842 - val_categorical_accuracy: 0.6914\n",
      "Epoch 186/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3796 - categorical_accuracy: 0.8523 - val_loss: 0.7276 - val_categorical_accuracy: 0.6667\n",
      "Epoch 187/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3841 - categorical_accuracy: 0.8554 - val_loss: 0.7081 - val_categorical_accuracy: 0.6667\n",
      "Epoch 188/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3773 - categorical_accuracy: 0.8677 - val_loss: 0.7227 - val_categorical_accuracy: 0.6543\n",
      "Epoch 189/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3782 - categorical_accuracy: 0.8523 - val_loss: 0.7546 - val_categorical_accuracy: 0.6605\n",
      "Epoch 190/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3814 - categorical_accuracy: 0.8646 - val_loss: 0.6682 - val_categorical_accuracy: 0.7037\n",
      "Epoch 191/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3765 - categorical_accuracy: 0.8492 - val_loss: 0.6889 - val_categorical_accuracy: 0.6914\n",
      "Epoch 192/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3727 - categorical_accuracy: 0.8677 - val_loss: 0.6765 - val_categorical_accuracy: 0.7222\n",
      "Epoch 193/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3736 - categorical_accuracy: 0.8615 - val_loss: 0.6600 - val_categorical_accuracy: 0.7160\n",
      "Epoch 194/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3749 - categorical_accuracy: 0.8615 - val_loss: 0.6824 - val_categorical_accuracy: 0.7160\n",
      "Epoch 195/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3727 - categorical_accuracy: 0.8523 - val_loss: 0.6868 - val_categorical_accuracy: 0.6975\n",
      "Epoch 196/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3697 - categorical_accuracy: 0.8677 - val_loss: 0.6799 - val_categorical_accuracy: 0.7099\n",
      "Epoch 197/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3709 - categorical_accuracy: 0.8615 - val_loss: 0.7110 - val_categorical_accuracy: 0.6728\n",
      "Epoch 198/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3657 - categorical_accuracy: 0.8677 - val_loss: 0.6927 - val_categorical_accuracy: 0.6975\n",
      "Epoch 199/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3660 - categorical_accuracy: 0.8646 - val_loss: 0.6854 - val_categorical_accuracy: 0.7160\n",
      "Epoch 200/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3638 - categorical_accuracy: 0.8585 - val_loss: 0.6783 - val_categorical_accuracy: 0.7160\n",
      "Epoch 201/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3665 - categorical_accuracy: 0.8646 - val_loss: 0.6878 - val_categorical_accuracy: 0.7222\n",
      "Epoch 202/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3696 - categorical_accuracy: 0.8492 - val_loss: 0.6860 - val_categorical_accuracy: 0.6975\n",
      "Epoch 203/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3628 - categorical_accuracy: 0.8800 - val_loss: 0.6959 - val_categorical_accuracy: 0.6975\n",
      "Epoch 204/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3594 - categorical_accuracy: 0.8677 - val_loss: 0.6716 - val_categorical_accuracy: 0.7222\n",
      "Epoch 205/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3580 - categorical_accuracy: 0.8646 - val_loss: 0.6934 - val_categorical_accuracy: 0.7037\n",
      "Epoch 206/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3599 - categorical_accuracy: 0.8615 - val_loss: 0.6912 - val_categorical_accuracy: 0.7037\n",
      "Epoch 207/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3596 - categorical_accuracy: 0.8615 - val_loss: 0.6717 - val_categorical_accuracy: 0.7222\n",
      "Epoch 208/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3534 - categorical_accuracy: 0.8708 - val_loss: 0.7134 - val_categorical_accuracy: 0.6914\n",
      "Epoch 209/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3572 - categorical_accuracy: 0.8615 - val_loss: 0.6914 - val_categorical_accuracy: 0.7099\n",
      "Epoch 210/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3510 - categorical_accuracy: 0.8677 - val_loss: 0.6862 - val_categorical_accuracy: 0.7099\n",
      "Epoch 211/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3510 - categorical_accuracy: 0.8646 - val_loss: 0.7189 - val_categorical_accuracy: 0.6975\n",
      "Epoch 212/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3501 - categorical_accuracy: 0.8677 - val_loss: 0.6927 - val_categorical_accuracy: 0.7037\n",
      "Epoch 213/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3527 - categorical_accuracy: 0.8769 - val_loss: 0.6785 - val_categorical_accuracy: 0.7346\n",
      "Epoch 214/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3513 - categorical_accuracy: 0.8738 - val_loss: 0.6880 - val_categorical_accuracy: 0.7099\n",
      "Epoch 215/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3464 - categorical_accuracy: 0.8800 - val_loss: 0.6878 - val_categorical_accuracy: 0.6975\n",
      "Epoch 216/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3495 - categorical_accuracy: 0.8738 - val_loss: 0.7068 - val_categorical_accuracy: 0.6914\n",
      "Epoch 217/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3519 - categorical_accuracy: 0.8800 - val_loss: 0.6884 - val_categorical_accuracy: 0.7037\n",
      "Epoch 218/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3445 - categorical_accuracy: 0.8831 - val_loss: 0.7705 - val_categorical_accuracy: 0.6667\n",
      "Epoch 219/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3483 - categorical_accuracy: 0.8800 - val_loss: 0.7560 - val_categorical_accuracy: 0.6728\n",
      "Epoch 220/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3455 - categorical_accuracy: 0.8677 - val_loss: 0.6868 - val_categorical_accuracy: 0.7222\n",
      "Epoch 221/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3470 - categorical_accuracy: 0.8677 - val_loss: 0.7806 - val_categorical_accuracy: 0.6667\n",
      "Epoch 222/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3490 - categorical_accuracy: 0.8554 - val_loss: 0.7001 - val_categorical_accuracy: 0.7160\n",
      "Epoch 223/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3454 - categorical_accuracy: 0.8646 - val_loss: 0.6691 - val_categorical_accuracy: 0.7222\n",
      "Epoch 224/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3420 - categorical_accuracy: 0.8800 - val_loss: 0.7153 - val_categorical_accuracy: 0.6975\n",
      "Epoch 225/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3412 - categorical_accuracy: 0.8646 - val_loss: 0.6861 - val_categorical_accuracy: 0.7222\n",
      "Epoch 226/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3402 - categorical_accuracy: 0.8769 - val_loss: 0.6942 - val_categorical_accuracy: 0.7099\n",
      "Epoch 227/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3417 - categorical_accuracy: 0.8769 - val_loss: 0.6816 - val_categorical_accuracy: 0.7222\n",
      "Epoch 228/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3360 - categorical_accuracy: 0.8800 - val_loss: 0.7088 - val_categorical_accuracy: 0.7037\n",
      "Epoch 229/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3328 - categorical_accuracy: 0.8708 - val_loss: 0.6896 - val_categorical_accuracy: 0.7346\n",
      "Epoch 230/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3355 - categorical_accuracy: 0.8800 - val_loss: 0.6981 - val_categorical_accuracy: 0.7099\n",
      "Epoch 231/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3323 - categorical_accuracy: 0.8769 - val_loss: 0.6883 - val_categorical_accuracy: 0.7346\n",
      "Epoch 232/600\n",
      "21/21 [==============================] - 0s 3ms/step - loss: 0.3356 - categorical_accuracy: 0.8862 - val_loss: 0.6910 - val_categorical_accuracy: 0.7160\n",
      "Epoch 233/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3333 - categorical_accuracy: 0.8800 - val_loss: 0.6991 - val_categorical_accuracy: 0.7160\n",
      "Epoch 234/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3319 - categorical_accuracy: 0.8738 - val_loss: 0.7032 - val_categorical_accuracy: 0.7222\n",
      "Epoch 235/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3329 - categorical_accuracy: 0.8800 - val_loss: 0.7299 - val_categorical_accuracy: 0.6975\n",
      "Epoch 236/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3395 - categorical_accuracy: 0.8738 - val_loss: 0.6808 - val_categorical_accuracy: 0.7346\n",
      "Epoch 237/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3276 - categorical_accuracy: 0.8862 - val_loss: 0.6850 - val_categorical_accuracy: 0.7407\n",
      "Epoch 238/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3333 - categorical_accuracy: 0.8738 - val_loss: 0.6921 - val_categorical_accuracy: 0.7222\n",
      "Epoch 239/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3258 - categorical_accuracy: 0.8738 - val_loss: 0.6982 - val_categorical_accuracy: 0.7160\n",
      "Epoch 240/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3273 - categorical_accuracy: 0.8738 - val_loss: 0.6870 - val_categorical_accuracy: 0.7222\n",
      "Epoch 241/600\n",
      "21/21 [==============================] - 0s 5ms/step - loss: 0.3306 - categorical_accuracy: 0.8738 - val_loss: 0.6897 - val_categorical_accuracy: 0.7160\n",
      "Epoch 242/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3259 - categorical_accuracy: 0.8769 - val_loss: 0.7265 - val_categorical_accuracy: 0.6975\n",
      "Epoch 243/600\n",
      "21/21 [==============================] - 0s 4ms/step - loss: 0.3252 - categorical_accuracy: 0.8954 - val_loss: 0.6820 - val_categorical_accuracy: 0.7407\n",
      "6/6 [==============================] - 0s 3ms/step - loss: 0.6463 - categorical_accuracy: 0.7778\n",
      "0.11222370568164959\n",
      "0.21984051036682617\n",
      "0.646265983581543\n",
      "0.7777777910232544\n"
     ]
    },
    {
     "ename": "AssertionError",
     "evalue": "ignored",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mAssertionError\u001b[0m                            Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-356-ec3347e89509>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     79\u001b[0m \u001b[0;31m# implement your solution ABOVE\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     80\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 81\u001b[0;31m \u001b[0mtester\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtest\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'cl_model_learning'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtest_ce\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtest_acc\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtest_fpr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtest_fnr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     82\u001b[0m \u001b[0mtester\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mprint_all_tests_successful\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/content/annbsc22_p1_hw2_tester.py\u001b[0m in \u001b[0;36mtest\u001b[0;34m(self, test_name, *args)\u001b[0m\n\u001b[1;32m    154\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtest_results\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtest_name\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mFalse\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    155\u001b[0m         \u001b[0mtest_func\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTESTS_DICT\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtest_name\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 156\u001b[0;31m         \u001b[0mtest_func\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    157\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtest_results\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtest_name\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mTrue\u001b[0m    \u001b[0;31m# only executed if no assert happened during test\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    158\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/content/annbsc22_p1_hw2_tester.py\u001b[0m in \u001b[0;36m__test_cl_model_learning\u001b[0;34m(self, *args)\u001b[0m\n\u001b[1;32m    622\u001b[0m                                               \u001b[0;34m\"FPR between \"\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmin_test_fpr\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m\" and \"\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmax_test_fpr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    623\u001b[0m         \u001b[0;32massert\u001b[0m \u001b[0mtest_fpr\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0mmin_test_fnr\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mtest_fpr\u001b[0m \u001b[0;34m<\u001b[0m \u001b[0mmax_test_fnr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"Tester: A well-trained model should produce an \"\u001b[0m\u001b[0;31m \u001b[0m\u001b[0;31m\\\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 624\u001b[0;31m                                               \u001b[0;34m\"FNR between \"\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmin_test_fnr\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m\" and \"\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmax_test_fnr\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    625\u001b[0m         \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"Tester: Classification model learning OK\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mAssertionError\u001b[0m: Tester: A well-trained model should produce an FNR between 0.15 and 0.55"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2304x1152 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# implement your solution BELOW\n",
    "\n",
    "def multiclass_fpr_fnr(y_onehot_pred, y_onehot_true):\n",
    "    pred_classes = np.argmax(y_onehot_pred, axis=1)\n",
    "    class_cnt = y_onehot_pred.shape[1]\n",
    "    y_true = np.argmax(y_onehot_test, axis=1)  # reverse one-hot\n",
    "\n",
    "    # https://www.mdpi.com/2227-7080/9/4/81/htm\n",
    "    confusion_mx = np.bincount(y_true * class_cnt + pred_classes,\n",
    "                               minlength=class_cnt*class_cnt)\\\n",
    "                     .reshape((class_cnt, class_cnt))\n",
    "\n",
    "    per_class_true_pos = np.diagonal(confusion_mx)\n",
    "\n",
    "    per_class_true_pos_plus_false_pos = np.sum(confusion_mx, axis=0)\n",
    "    per_class_false_pos = per_class_true_pos_plus_false_pos - per_class_true_pos\n",
    "\n",
    "    per_class_true_pos_plus_false_neg = np.sum(confusion_mx, axis=1)\n",
    "    per_class_false_neg = per_class_true_pos_plus_false_neg - per_class_true_pos\n",
    "\n",
    "    per_class_true_neg = np.sum(confusion_mx) - \\\n",
    "                         (per_class_true_pos_plus_false_pos + per_class_false_neg)\n",
    "\n",
    "    # to avoid warning when dividing with 0 in case of 0/0\n",
    "    summed_fpr = np.sum(np.divide(per_class_false_pos,\n",
    "                                  per_class_false_pos + per_class_true_neg,\n",
    "                                  out=np.zeros(per_class_false_pos.shape),\n",
    "                                  where=per_class_true_pos_plus_false_pos!=0))\n",
    "        \n",
    "    summed_fnr = np.sum(np.divide(per_class_false_neg,\n",
    "                                  per_class_true_pos_plus_false_neg,\n",
    "                                  out=np.zeros(per_class_false_neg.shape),\n",
    "                                  where=per_class_true_pos_plus_false_neg!=0))\n",
    "\n",
    "    return (summed_fpr / class_cnt), (summed_fnr / class_cnt)\n",
    "\n",
    "# training\n",
    "earlystopping_callback = tf.keras.callbacks.EarlyStopping(monitor=\"val_loss\", patience=50)\n",
    "\n",
    "history = cl_model.fit(x_train, y_onehot_train, validation_data=(x_val, y_onehot_val),\n",
    "                       batch_size=16, epochs=600, verbose=1, callbacks=[earlystopping_callback])\n",
    "\n",
    "# metrics visualization\n",
    "tr_mse_losses = history.history['loss']\n",
    "val_mse_losses = history.history['val_loss']\n",
    "tr_acc = history.history['categorical_accuracy']\n",
    "val_acc = history.history['val_categorical_accuracy']\n",
    "\n",
    "# plots\n",
    "plt.figure(figsize=(32, 16))\n",
    "\n",
    "plt.subplot(3, 4, 1)\n",
    "plt.plot(tr_mse_losses, label=\"J_train\")\n",
    "plt.plot(val_mse_losses, label=\"J_val\")\n",
    "plt.ylim((0,1.5))\n",
    "plt.xlabel(\"Number of epochs\")\n",
    "plt.ylabel(\"Cost (J)\")\n",
    "plt.legend()\n",
    "plt.title(\"Categorical crossentropy during training\")\n",
    "\n",
    "plt.subplot(3, 4, 2)\n",
    "plt.plot(tr_acc, label=\"A_train\")\n",
    "plt.plot(val_acc, label=\"A_val\")\n",
    "plt.ylim((0,1))\n",
    "plt.xlabel(\"Number of epochs\")\n",
    "plt.ylabel(\"Accuracy (%)\")\n",
    "plt.legend()\n",
    "plt.title(\"Accuracy during training\")\n",
    "\n",
    "# metrics\n",
    "test_ce, test_acc = cl_model.evaluate(x_test, y_onehot_test)\n",
    "\n",
    "test_fpr, test_fnr = multiclass_fpr_fnr(cl_model.predict(x_test), y_onehot_test)\n",
    "\n",
    "print(test_fpr)\n",
    "print(test_fnr)\n",
    "print(test_ce)\n",
    "print(test_acc)\n",
    "\n",
    "# implement your solution ABOVE"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "nvp190_handin_2.ipynb",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}