mnist_notebook.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "bd1f546b-52d4-45f9-ba75-73f82316f770",
   "metadata": {},
   "source": [
    "## MNIST Handwritten digits classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "fb529ab3-fdec-4a3f-8b98-2b3d8363ab03",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import the goodies\n",
    "%matplotlib inline\n",
    "\n",
    "import numpy as np\n",
    "import time\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras \n",
    "from tensorflow.keras import layers\n",
    "from tensorflow.keras import models\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "from sklearn.metrics import log_loss, accuracy_score\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "a25df8cd-fa14-4c6c-8608-950a94dd9775",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load the data (Automatically downloads from AWS, so no file manpulation needed)\n",
    "(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "291e40a9-b696-4081-a2c3-6420ca2290e2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reshape the data\n",
    "X = x_train.reshape(60000, 28, 28, 1)\n",
    "y = y_train.reshape(60000, 1)\n",
    "\n",
    "# Process the data (To float)\n",
    "X = X / 255.0\n",
    "\n",
    "# Onehot encode y_train\n",
    "enc = OneHotEncoder().fit(y)\n",
    "y = enc.transform(y).toarray()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "cf8dbee4-b0ae-491b-bf2f-087b171bf63e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Format the test dataset (However, no processing is done on the set)\n",
    "X_ = x_test.reshape(10000, 28, 28, 1)\n",
    "y_ = enc.transform(y_test.reshape((10000,1))).toarray()\n",
    "\n",
    "X_ = X_ / 255.0"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c5cc53a5-e64e-4ebf-909a-2f62a057f971",
   "metadata": {},
   "source": [
    "To find the most ambigious images in the test dataset, we must first build a classifier that gives us the probability of which digit it is. This is done using a [Convolutional neural network](https://towardsdatascience.com/a-comprehensive-guide-to-convolutional-neural-networks-the-eli5-way-3bd2b1164a53) written in Keras/Tensorflow\n",
    "\n",
    "Necessary requirements are in `requirements.txt`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "0ec73c4c-8b4a-4d5d-be28-2d51d62d4183",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d (Conv2D)              (None, 26, 26, 32)        320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 1600)              0         \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 1600)              0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 256)               409856    \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 10)                2570      \n",
      "=================================================================\n",
      "Total params: 431,242\n",
      "Trainable params: 431,242\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-09-09 22:05:29.930651: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:29.938509: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:29.939078: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:29.940109: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2021-09-09 22:05:29.940699: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:29.941240: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:29.941753: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:30.241796: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:30.242169: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:30.242502: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2021-09-09 22:05:30.242810: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 8032 MB memory:  -> device: 0, name: GeForce RTX 3080, pci bus id: 0000:06:00.0, compute capability: 8.6\n"
     ]
    }
   ],
   "source": [
    "# Build the network\n",
    "model = models.Sequential()\n",
    "model.add(layers.Conv2D(32, (3,3), activation='relu', input_shape=(28, 28, 1)))  # Filter the data\n",
    "model.add(layers.MaxPooling2D((2,2)))                                            # Make the view larger; find bigger patterns\n",
    "model.add(layers.Conv2D(64, (3,3), activation='relu'))\n",
    "model.add(layers.MaxPooling2D((2,2)))\n",
    "model.add(layers.Flatten())                                                      # Dimension reduce\n",
    "model.add(layers.Dropout(0.5))                                                   # Get rid of conspiracies\n",
    "model.add(layers.Dense(256, activation='relu'))\n",
    "model.add(layers.Dense(10, activation='softmax'))\n",
    "\n",
    "callbacks = [\n",
    "    tf.keras.callbacks.EarlyStopping(patience=5, restore_best_weights=True, monitor=\"val_loss\", verbose=1),  # Stop early if training is only going ok\n",
    "    tf.keras.callbacks.ModelCheckpoint(\"net1.h5\", save_best_only=True, monitor=\"val_loss\"),                      # To save the best weights\n",
    "    tf.keras.callbacks.ProgbarLogger(),\n",
    "]\n",
    "\n",
    "model.compile(loss='binary_crossentropy',\n",
    "              optimizer=keras.optimizers.RMSprop(learning_rate=1e-4),\n",
    "              metrics=['acc'])\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "bbf19c3e-6cfb-42bd-b551-9ca4bf1e1e1b",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2021-09-09 22:05:30.490956: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)\n",
      "2021-09-09 22:05:31.203869: I tensorflow/stream_executor/cuda/cuda_dnn.cc:369] Loaded cuDNN version 8100\n",
      "2021-09-09 22:05:32.053359: I tensorflow/stream_executor/cuda/cuda_blas.cc:1760] TensorFloat-32 will be used for the matrix multiplication. This will only be logged once.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Restoring model weights from the end of the best epoch.\n",
      "Epoch 00081: early stopping\n",
      "47.89 seconds to train\n"
     ]
    }
   ],
   "source": [
    "# Fit the network\n",
    "params = {\n",
    "    'epochs': 100,\n",
    "    'batch_size': 256,\n",
    "    'verbose': 0,\n",
    "    'validation_split': 0.2,\n",
    "    'callbacks': callbacks\n",
    "}\n",
    "\n",
    "tick = time.time()\n",
    "history = model.fit(X, y,  **params)\n",
    "print(f'{(time.time()-tick).__round__(2)} seconds to train')    \n",
    "    \n",
    "model.save('net1.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "1cf7da62-0c12-4cbc-bd54-d62f0cd20d32",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot the training history\n",
    "acc = history.history['acc']\n",
    "val_acc = history.history['val_acc']\n",
    "loss = history.history['loss']\n",
    "val_loss = history.history['val_loss']\n",
    "\n",
    "epochs = range(1, len(acc)+1)\n",
    "\n",
    "plt.plot(epochs, acc, 'r-', label='Training')\n",
    "plt.plot(epochs, val_acc, 'b-', label='Validation')\n",
    "plt.title('Validation and Training Accuracy')\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Accuracy (%)\")\n",
    "plt.legend()\n",
    "\n",
    "plt.figure()  # Combines the two graphs\n",
    "\n",
    "plt.plot(epochs, loss, 'r-', label='Training')\n",
    "plt.plot(epochs, val_loss, 'b-', label='Validation')\n",
    "plt.title('Validation and Training Loss')\n",
    "plt.xlabel(\"Epoch\")\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "2bbe2f94-a0af-4b5b-ae32-8df48edd0241",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training data\n",
      "Loss 0.01656877341912087\n",
      "Acc 0.99605\n",
      "\n",
      "Test data:\n",
      "Loss 0.028536008968894425\n",
      "Acc 0.9926\n"
     ]
    }
   ],
   "source": [
    "# How well did the model work?\n",
    "train_preds = model.predict(X)\n",
    "print(\"Training data\")\n",
    "print(\"Loss\", log_loss(y, train_preds))\n",
    "print(\"Acc\", accuracy_score(y_train, train_preds.argmax(axis=1)))\n",
    "\n",
    "preds = model.predict(X_)\n",
    "print(\"\\nTest data:\")\n",
    "print(\"Loss\", log_loss(y_, preds))\n",
    "print(\"Acc\", accuracy_score(y_test, preds.argmax(axis=1)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "64c50006-f5df-42be-afd5-bcf518375646",
   "metadata": {},
   "source": [
    "The model ended up with 99.2% accuracy on the test set showing that it fit to the data very well.\n",
    "It can then provide us with valuable insight towards the test dataset of unusual digits in the dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "bd10ffae-cd6a-4251-9b98-6375b7f0295c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWsAAAEWCAYAAACg+rZnAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8rg+JYAAAACXBIWXMAAAsTAAALEwEAmpwYAAAjgklEQVR4nO3deZxcVZn/8c83nQRZQhACSEhYhLgEzA8hRDYRBR02RUZnBGYEHCHGEZBxHOU3M4qKy+DODGiMgIoKiCyawSgoq4BAAAOSABIDSAj7noBAdz/zxz2tN2V1dXWn7q261d83r/vqW3c5z6nq8NTpc+89RxGBmZl1tjHtroCZmQ3NydrMrAKcrM3MKsDJ2sysApyszcwqwMnazKwCnKxHOUn/IOnSJo6bK+kTZdSpJu4HJT0saaWkjUqOvZWkkDQ2vf65pCNGUM4Wqf49ra+ljRbyfdadS9K9wKZAH7AKWAAcGxEr21mvskgaBzwD7BIRt7Yh/lbAPcC4iOgdxnn3AkdFxK8KqpqNQm5Zd763R8R6wI7AzsB/1h4w0PLrQpsCLwMWt6KwLv6cbBRwsq6IiHgA+DmwPUD68/xDku4G7k7bDpS0SNJTkq6TNGPgfElTJV0o6VFJj0s6NW0/UtI1aV2SvibpEUlPS7pN0kC870r6bK68oyUtlfSEpPmSJuf2haQ5ku6W9KSk0ySp3vuStJakr0takZavp22vAu5Khz0l6fI65w50U8xO5z4o6V9z+z8l6XxJP5D0DHCkpImSzkjHPiDpswPdE5J6JH1Z0mOSlgEH1MS7UtJRNZ/BHZKelbRE0o6Svg9sAfxv6vr4WJ3ulMnpM3sifYZH19T5PElnpXIXS5qZ2//xVO9nJd0lae96n6t1oYjw0qELcC+wT1qfStbCPCm9DuCXwIbA2mQt70eANwA9wBHp/LXS61uBrwHrkrVW90jlHAlck9b/BrgZ2AAQ8Fpgs7Tvu8Bn0/pbgMdSzLWA/wGuztU7gItTOVsAjwL7DvIePwNcD2wCbAxcl3uPW6Wyxg5y7sD+c9L7el2KNfCZfQp4CXgnWcNkbeAnwLfS8ZsANwIfSMfPAe5Mn/WGwBX5+MCVZN0bAH8HPED2146AbYEta39v9d4HcBXwjfR72CHVee9cnf8E7J9+b18Ark/7Xg3cD0zOlbtNu/+deilnaXsFvDT45WT/068EngLuS/+Dr532BfCW3LHfHEhyuW13AW8Cdk0J4a+SHqsn67cAvwd2AcbUHPdd/pKszwC+mNu3XkqKW+Xqtkdu/3nACYO8xz8A++de/w1wb1pfLcnVOXdg/2ty274InJHWP8XqXyKbAi8MfIZp26HAFWn9cmBObt/bGiTrS4APN/i91U3WZF8EfcCE3P4vAN/N1flXuX3TgefT+rZkX8j7kPWjt/3fqJfyFneDdL53RsQGEbFlRPxzRDyf23d/bn1L4F9TF8hTkp4iSwyT08/7YoiLZBFxOXAqcBrwsKR5ktavc+hksi+PgfNWAo8Dm+eOeSi3/hxZQq9ntbLS+uRBjh1M/nOoPb/2MxoHPJj7jL5F1sIeqEttWYOZSvZFM1yTgSci4tmaOI0+u5dJGhsRS4HjyRL6I5LOzXc/WXdzsq62/K089wOfS4l9YFknIs5J+7Zo5gJbRPx3ROwEbAe8Cvi3OoetIEt8AEhaF9iIrFtguFYri6zbZMUwy5ja4Pzaz+gFYFLuM1o/IrZL+x+sU9Zg7ge2GWRfo1usVgAbSppQE6epzy4izo6IPcg+swBObuY8qz4n6+7xbWCOpDekC4XrSjogJYUbyRLRf6XtL5O0e20BknZO548ju1XwT2R/stc6G3ifpB0krQV8HrghIu4dQb3PAf5T0saSJgGfBH4wzDI+IWkdSdsB7wN+VO+giHgQuBT4iqT1JY2RtI2kN6VDzgOOkzRF0suBExrEPB34qKSd0ue9raSBL52HgVcOUof7yfrlv5B+DzOA9wM/HOpNSnq1pLekz/xPwPPU//1YF3Ky7hIRcRNwNFk3xpPAUrL+aCKiD3g7WZ/nH4HlwHvqFLM+WdJ/kuxP88eBL9eJdRnwCeACsi+BbYBDRlj1zwI3AbcBvwNuSduG4yqy93sZ8OWIaPSQz+HAeGAJ2fs8H9gs7fs2WV/0rakeFw5WSET8GPgc2RfXs2QXLjdMu79A9gX0lKSP1jn9ULJ+7BXARcCJEfHLod4k2cXc/yK7uPsQWffNvzdxnnUBPxRjlaURPrRiVkVuWZuZVYCTtZlZBbgbxMysAtyyNjOrgI4d2Gbs+M27ssl/3oZvGvqgFvn7J64qLVZZxvWU90/2S5PeWFqsLV8q5w68J3vKG6X12Kd/U1qsp1f+oe7YM8Px0mPLms454ya9co3jDVfHJmszs1L1d/Yt607WZmYA0d/uGjTkZG1mBtDvZG1m1vHCLWszswro6+yHYJ2szczAFxjNzCrB3SBmZhXgC4xmZp1v1F5glPQa4CCy6YqCbOze+RFxR1ExzcxGrMNb1oWMDSLp48C5ZLM+3wgsTOvnSBp09g1JsyXdJOmm/v5VRVTNzKy+vpeaX9qgqJb1+4HtImK1dyXpq8Bistku/kpEzAPmQfeODWJmHWqUdoP089ezVkM2fVJnfyJmNjp1eDdIUcn6eOAySXeTzQIN2QzO2wLHFBTTzGzkRmPLOiJ+IelVwCyyC4wim6R1YZq81cyss4zSljWR3QdzfVHlm5m1UvS358Jhs3yftZkZjN6WtZlZpYzGPmszs8rxQE5mZhXglrWZWQW4z9ry7lqrkCf865qx0dalxLnt8XtKiQMwc8NtS4v1SE95D9HOeKmcge9/Pq68P/VXvfin0mK1hCcfMDOrALeszcw6X6c/r+dkbWYGblmbmVWC7wYxM6sAt6zNzCrAd4OYmVVAh3eDlHfTr5lZJ+vvb34ZgqR9Jd0laWm9qQwl7SXpaUmL0vLJocp0y9rMDFrWZy2pBzgNeCtpHH9J8yNiSc2hv46IA5st1y1rMzPIukGaXRqbBSyNiGUR8SLZ5OEHrWn1Sk/Wkt7XYJ9nNzez9ujrbXrJ56q0zM6VtDl/mc4Qstb15nUi7irpVkk/l7TdUNVrRzfIp4Hv1Nvh2c3NrG2G0Q2Sz1V1qN4pNa9vAbaMiJWS9gd+AkxrFLOQZC3ptsF2AZsWEdPMbI207m6Q5cDU3OspwIrVQkU8k1tfIOkbkiZFxGODFVpUy3pT4G+AJ2u2C7iuoJhmZiPXuodiFgLTJG0NPAAcAhyWP0DSK4CHIyIkzSLrkn68UaFFJeuLgfUiYlHtDklXFhTTzGzkWpSsI6JX0jHAJUAPcGZELJY0J+2fC7wb+KCkXuB54JCIaNj1W0iyjoj3N9h32GD7zMzapnGuHGZRsQBYULNtbm79VODU4ZTp+6zNzAB6/bi5mVnn6/DHzZ2szczAo+6ZmVVCC/usi+BkbWYGblmP1H6veH1psa598q7SYn3iwStKi1XvMaoivHfyLiVFgu+vuL60WDeO+X1psT5x369KifPFnT9SSpxKcrI2M+t80ecJc83MOp9b1mZmFeBb98zMKqDfd4OYmXU+d4OYmVWALzCamVWAW9ZmZhXgPmszswro8LtBCpswV9JrJO0tab2a7fsWFdPMbMT6o/mlDQpJ1pKOA34KHAvcLik/DfvnG5z35xmD71v5xyKqZmZWV/T3N720Q1HdIEcDO6WZe7cCzpe0VUScQoMhK/IzBr99iwM7uwPJzLrLKL0bpCciVgJExL2S9iJL2FtS3vhCZmbN6/ALjEX1WT8kaYeBFylxHwhMAl5XUEwzs5Hr729+aYOiWtaHA6tNaBYRvcDhkr5VUEwzs5Hr8JZ1UbObL2+w79oiYpqZrZEOv3XP91mbmcHobFmbmVVN9I7Ou0HMzKrFLWszswpwn7WZWQW4ZT0yv3jot6XFKvNX9PVN31xarOMfLmcm9TJnHF86fXppsbZdsqS0WB97w4mlxPnoC+uUEgfgktIitUZ0eLIubCAnM7NK6e1rfhmCpH0l3SVpqaQTGhy3s6Q+Se8eqsyObVmbmZWqRS1rST3AacBbgeXAQknzI2JJneNOpsk/QtyyNjODVg6ROgtYGhHLIuJF4FzgoDrHHQtcADzSTPWcrM3MgIhoeskP55yW2bmiNgfuz71enrb9maTNgYOBuc3Wz90gZmYwrG6Q/HDOddQbWbS28K8DH4+IPqm5gUidrM3MoJW37i0HpuZeTwFW1BwzEzg3JepJwP6SeiPiJ4MV6mRtZgZEb8seilkITJO0NfAAcAhw2GqxIrYeWJf0XeDiRokanKzNzDItytUR0SvpGLK7PHqAMyNisaQ5aX/T/dR5hSVrSbOAiIiFkqYD+wJ3RsSComKamY1UKx+KSXluQc22ukk6Io5spsxCkrWkE4H9gLGSfgm8AbgSOEHS6yPic4OcNxuYDTCmZyJjxqxbRPXMzP5ahz/BWFTL+t3ADsBawEPAlIh4RtKXgBuAusk6f4V13PjNO/uTM7Pu0tnjODWXrNM9gVvmj4+Iqxuc0hsRfcBzkv4QEc+kc56X1OEfiZmNRp0+NsiQyVrSycB7gCXAwEPxATRK1i9KWicingN2ypU1kY7//jKz0Sh6K56sgXcCr46IF4ZR7p4Dx0esNkjsOOCIYZRjZlaODm9GNpOsl5El2aaT9WCJPSIeAx5rthwzs7J0+NwDgydrSf9D1t3xHLBI0mXkEnZEHFd89czMSlLVZA3clH7eDMyv2dfZnTtmZsNU2ZZ1RHwPQNKHI+KU/D5JHy66YmZmZYredtegsWaGSK13QfDIFtfDzKytor/5pR0a9VkfSjb4yNaS8t0gE4DHi66YmVmZKtsNAlwHPEg2fN9XctufBW4rslLQvZ3i145Z2e4qVFqZk9i+5uVThz6oRZb2l/Pv4vPjy/tbf9Xvf1parJaI5saVbpdGfdb3AfcBu5ZXHTOz9qhyyxoASc/yl4bueLJ7rldFxPpFVszMrEzRX9GW9YCImJB/LemdZBNCmpl1jf6+zk7Ww54wN81m8JbWV8XMrH0qezfIAEl/m3s5hmzusG69/mdmo1Tlu0GAt+fWe4F7gYMKqY2ZWZtEhzdBGyZrST3AbRHxtZLqY2bWFp3esm7YZ50mEHhHSXUxM2ub/j41vbRDM90g10k6FfgRsGpgY0TcUlitzMxK1ukt60aPm18aEW8DdkubPpPbHQzzjhBJZ0XE4cOvoplZ8aKqTzACGwNExJuHW2jNWCIAAt4saYNUZt2ulfzs5vLs5mZWoio/wTix5ra91UTEhQ3OnUI2Z+PpZK1wkd3y95UG56w2u/lYz25uZiXqr3DLeiJwIFmirRVAo2Q9E/gw8B/Av0XEIknPR8RVI66pmVmBqtwNcl9E/NNICk2T5H5N0o/Tz4eHiGVm1lad/rh5owS6xjWPiOXA30k6AHhmTcszMytKZe8GAd7bqiAR8TPgZ60qz8ys1Tq9z3rQh2Ii4vYyK2Jm1k4RanoZiqR9Jd0laamkE+rsP0jSbZIWSbpJ0h5Dlel+ZDMzWjc2SBqm4zTgrcByYKGk+RGRn+boMmB+RISkGcB5wGsaletkbWZGS7tBZgFLI2IZgKRzyQa/+3Oyjoj8PG7r0sRIpo2eYPxdowIiYsbQdTYzq4b+YVxgzD/Al8xLz4kAbA7cn9u3HHhDnTIOBr4AbAIcMFTMRi3rA9PPD6Wf308//wF4bqiCzcyqZDgt6/wDfHUM9mxKbRkXARdJ2hM4CdinUcyhJsxF0u4RsXtu1wmSrmX1sUJa7vjJexZZ/Gq+vuLq0mL94vHyrts+v+LXpcRZe/IbS4lTtsnjNygt1m2r/lhKnL4Sn6metN3flxbrmVXL1riMFj4UsxyYmns9BVgxeNy4WtI2kiZFxGODHdfMtF7r5q9UStqNrI/FzKxr9IeaXoawEJgmaWtJ44FDgNXGS5K0rSSl9R3JJiN/vFGhzVxgfD9wpqSJ6fVTwIiebDQz61StGowoInolHQNcAvQAZ0bEYklz0v65wLuAwyW9BDwPvCei8f0ozcxufjPw/yStDyginl7D92Jm1nH6+oc9f/igImIBsKBm29zc+snAycMps5kJc9ci+xbYChibWu5ERKF91mZmZerwEVKb6gb5KfA0cDPwQrHVMTNrj1jz4ZAK1UyynhIR+xZeEzOzNurv8BH0m+mkuU7S6wqviZlZG/Wjppd2aKZlvQdwpKR7yLpBBISfYDSzbtIN3SD7FV4LM7M26+uCZL3GPTnpoZpZwO0Rcemalmdm1mqdfjdIM33WPwMuTj8vA5YBP290gqQbc+tHA6cCE4AT643tmjt2dhrb9abbnl3aRNXMzFqjfxhLOwyZrCPidRExI/2cRtZCvmaI08bl1mcDb42ITwNvIxsIarBY8yJiZkTMnDFh2yaqb2bWGoGaXtph2ONZR8QtknYe4rAxkl5O9mWgiHg0nbtKUu8I6mlmVqgOn4KxqScYP5J7OQbYEXh0iNMmkj1EIyAkvSIiHpK0Hi2YiNfMrNXadUtes5ppWU/IrfeS9V1f0OiEiNhqkF39wMFN1czMrER97a7AEJoZyOnTAJImZC9Xm45mWCLiOeCekZ5vZlaUfnV2y3rIC4yStpf0W+B2YLGkmyVtX3zVzMzKE8NY2qGZbpB5wEci4goASXulbbsVVy0zs3J1+n3WzSTrdQcSNUBEXCnJM8WYWVep/N0gwDJJn+AvE+b+I+53NrMu0+mPmzfzBOM/ARsDF6ZlEvC+IitlZla2fjW/tEPDlrWkHuDHEdFwivQiHLvhULdyt04/5c3O/Y2Hrist1jolzTr+pk22KyUOwD/GJqXFOvrhK4Y+qGIOn7xrabHOWvGb0mK1Qqf3WTdsWUdEH/BcbrJcM7Ou1A13g/wJ+J2kXwKrBjZGxHGF1crMrGTdcIHxZ2kxM+tand4N0swTjN8royJmZu3U1+Et60H7rCUdJOlDudc3SFqWlneXUz0zs3JUeTzrjwHzc6/XAnYG9gI+WGCdzMxK1+nJulE3yPiIuD/3+pqIeBx43E8wmlm3adddHs1qlKxfnn8REcfkXm5cTHXMzNqj0+8GadQNckOaP3E1kj4A3FjneDOzyqpyN8i/AD+RdBhwS9q2E1nf9TsbFSrpDcAdEfGMpLWBE8hmmFkCfD4inl7TipuZtVIrJx+QtC9wCtADnB4R/1Wz/x+Aj6eXK4EPRsStjcoctGUdEY9ExG7AScC9aflMROwaEQ8PUdczgefS+ilk03ydnLZ9Z7CT8rObn/3YA0OEMDNrnVaNDZKG6TgN2A+YDhwqaXrNYfcAb4qIGWQ5dt5Q9WvmPuvLgcuHOq7GmIgYmBh3ZkTsmNavkbSoQax5pErft+M+nd7fb2ZdpIXdG7OApRGxDEDSucBBZD0LAEREfpCg64EpQxXazKh7I3G7pIGR+W6VNBNA0quAlwqKaWY2YsMZGyTfC5CW2bmiNgfyd9ItT9sG837g50PVr5nHzUfiKOAUSf8JPAb8RtL9ZG/gqIJimpmNWP8wbt7L9wLUUa+jpG7hkt5Mlqz3GCpmIck6XUA8Mk2y+8oUZ3kTfd1mZm3RwguMy4GpuddTgBW1B0maAZwO7JeeYWmoqJY1ABHxLNDwCqeZWSdoYZ/1QmCapK2BB4BDgMPyB0jagmwyl/dGxO+bKbTQZG1mVhWteigmInolHQNcQnbr3pkRsVjSnLR/LvBJYCPgG5IAeiNiZqNynazNzBhen/VQImIBsKBm29zc+lEM8/qdk7WZGdUeG8TMbNSo/OQDZmajQV+Ht607Nllvc/sdpcXae9MZpcV68oKPlBbrUx+6oZQ4X1lxdSlxAK5icWmxutEVz95dWqyjJ+9eWqxWcMvazKwCWnmBsQhO1mZm+AKjmVkluBvEzKwCfIHRzKwC3GdtZlYBnZ2qnazNzAC3rM3MKqHTLzAWMlOMpOMkTR36SDOzzhDD+K8diprW6yTgBkm/lvTPkjZu5qT8VDn9/asKqpqZ2V/rI5pe2qGoZL2MbHaEk4CdgCWSfiHpiDR7TF0RMS8iZkbEzDFj1i2oamZmf61/GEs7FNVnHRHRD1wKXCppHNm07IcCXwaaammbmZWlP0bnBcbV5lyIiJeA+cB8SWsXFNPMbMQ6O1UXl6zfM9iOiHi+oJhmZiM2Km/da3YCSDOzTtGuuzya5fuszcyAXidrM7PO55a1mVkFdPoTjE7WZmZAjNJb98zMKmVU3g3SCmVOtvntFdeWFuuUf96otFibRE9pscryd5vtXFqsCx++ubRYB2z6+lLizH+wvPc0fsJrS4vVCp58wMysAtyyNjOrAPdZm5lVQKffDVLUqHtmZpXSyvGsJe0r6S5JSyWdUGf/ayT9RtILkj7aTP3csjYzo3V91pJ6gNOAtwLLgYWS5kfEktxhTwDHAe9stly3rM3MgL7ob3oZwixgaUQsi4gXgXOBg/IHRMQjEbEQeKnZ+jlZm5kxvG6Q/KxWaZmdK2pz4P7c6+Vp2xpxN4iZGcObfCAi5gHzBtmtOtvWuI/FydrMjJZOPrAcyE8YPgVYsaaFFpKsJY0HDgFWRMSvJB0G7AbcAcxLM8eYmXWMFj4UsxCYJmlr4AGyXHjYmhZaVMv6O6nsdSQdAawHXAjsTdb5fkS9k1K/z2yAPTfciekTXllQ9czMVteqZB0RvZKOAS4BeoAzI2KxpDlp/1xJrwBuAtYH+iUdD0yPiGcGK7eoZP26iJghaSzZN8vkiOiT9APg1sFOyvcDfXCrv+/sx4nMrKs0cZdH0yJiAbCgZtvc3PpDZN0jTSsqWY9JXSHrAusAE8nuK1wLGFdQTDOzERutkw+cAdxJ9ifAfwA/lrQM2IXsnkMzs44yKscGiYivSfpRWl8h6SxgH+DbEXFjETHNzNbEqB11LyJW5NafAs4vKpaZ2ZoalS1rM7Oq6evwcfecrM3MGN4TjO3gZG1mxui9G8TMrFLcsjYzqwC3rEdowbN3lhZr5qRppcU66dFrSov12olThz6oBS7acM9S4gDcQXkztp/fX94Fp3f0rl9KnHs32rqUOACnrfh1abFOaUEZblmbmVVAKx83L4KTtZkZ7gYxM6uEcMvazKzzjdrHzc3MqsSPm5uZVYBb1mZmFdBX4q2aI+FkbWaG7wYxM6uEUdtnLWkb4GCyKdl7gbuBcyLi6aJimpmNVKf3WY8polBJxwFzgZcBOwNrkyXt30jaq8F5syXdJOmmlX96ooiqmZnVFRFNL+1QVMv6aGCHNKP5V4EFEbGXpG8BPwVeX++k/OzmW240o7O/5sysq4zmC4xjgT6yGc0nAETEHyV5dnMz6zid3g1SVLI+HVgo6XpgT+BkAEkbA+7fMLOOMyovMEbEKZJ+BbwW+GpE3Jm2P0qWvM3MOsqoHSI1IhYDi4sq38yslXyftZlZBYzalrWZWZX0d/gQqYXcZ21mVjWtvM9a0r6S7pK0VNIJdfZL0n+n/bdJ2nGoMp2szcxoXbKW1AOcBuwHTAcOlTS95rD9gGlpmQ18c6j6OVmbmQExjGUIs4ClEbEsIl4EzgUOqjnmIOCsyFwPbCBps8YVHMa3SRUWYHY3xXGsasXqxvfUzbHWpI7ATblldm7fu4HTc6/fC5xac/7FwB6515cBMxvF7MaW9ewui+NY1YrVje+pm2ONSETMi4iZuWVebrfqnVLzupljVtONydrMrJ2Wkw1cN2AKsGIEx6zGydrMrLUWAtMkbS1pPHAIML/mmPnA4emukF2ApyPiwUaFduN91vOGPqRScRyrWrG68T11c6yWi4heSccAlwA9wJkRsVjSnLR/LrAA2B9YCjwHvG+ocpU6t83MrIO5G8TMrAKcrM3MKqBrkvVQj3e2MM6Zkh6RdHtRMXKxpkq6QtIdkhZL+nBBcV4m6UZJt6Y4ny4iTk3MHkm/lXRxwXHulfQ7SYsk3VRwrA0knS/pzvQ727WgOK9O72dgeUbS8QXF+pf0b+J2SedIelkRcVKsD6c4i4t6P5XW7pvLW3SDeg/wB+CVwHjgVmB6QbH2BHYEbi/hfW0G7JjWJwC/L+J9kd3zuV5aHwfcAOxS8Hv7CHA2cHHBce4FJhX9u0qxvgccldbHAxuUELMHeAjYsoCyNwfuAdZOr88DjizofWwP3A6sQ3bjw6+AaWX83qqydEvLupnHO1siIq6mpNluIuLBiLglrT8L3EH2P1Cr40RErEwvx6WlsCvPkqYAB5DNKNQVJK1P9kV+BkBEvBgRT5UQem/gDxFxX0HljwXWljSWLJE2vBd4DbwWuD4inouIXuAq4OCCYlVStyTrzYH7c6+XU0BSaydJW5FNNHxDQeX3SFoEPAL8MiIKiZN8HfgYUMaYlAFcKulmSUU+GfdK4FHgO6l753RJ6xYYb8AhwDlFFBwRDwBfBv4IPEh2L/ClRcQia1XvKWkjSeuQ3dY2dYhzRpVuSdbDfnSzSiStB1wAHB8RzxQRIyL6ImIHsiepZknavog4kg4EHomIm4sov47dI2JHslHOPiSpqGnlxpJ1j30zIl4PrAIKu3YCkB64eAfw44LKfznZX6hbA5OBdSX9YxGxIuIOsrlafwn8gqwrs7eIWFXVLcl62I9uVkWaDf4C4IcRcWHR8dKf7lcC+xYUYnfgHZLuJeuueoukHxQUi4hYkX4+AlxE1mVWhOXA8txfJOeTJe8i7QfcEhEPF1T+PsA9EfFoRLwEXAjsVlAsIuKMiNgxIvYk62q8u6hYVdQtybqZxzsrR5LI+kDviIivFhhnY0kbpPW1yf4nvbOIWBHx/yNiSkRsRfZ7ujwiCmmtSVpX0oSBdeBtZH9ut1xEPATcL+nVadPewJIiYuUcSkFdIMkfgV0krZP+Le5Ndt2kEJI2ST+3AP6WYt9b5XTF4+YxyOOdRcSSdA6wFzBJ0nLgxIg4o4hYZK3Q9wK/S/3JAP8eEQtaHGcz4Htp0PQxwHkRUegtdSXZFLgoyzOMBc6OiF8UGO9Y4IepwbCMJh4hHqnUr/tW4ANFxYiIGySdD9xC1iXxW4p9FPwCSRsBLwEfiognC4xVOX7c3MysArqlG8TMrKs5WZuZVYCTtZlZBThZm5lVgJO1mVkFOFkbkvrS6G23S/pxui1spGV9V9K70/rpkqY3OHYvSbvlXs+RdPhIY7eSpB0k7Z97vVpdzcrmZG0Az0fEDhGxPfAiMCe/M91/PWwRcVRENHowZC9yT8RFxNyIOGsksQqwA9n4FAP2YphP76XBj8xawsnaav0a2Da1JK+QdDbZQzk9kr4kaaGk2yR9ALKnLCWdKmmJpJ8BmwwUJOlKSTPT+r6SbkljZl+WBqaaA/xLatW/UdKnJH00Hb+DpOtTrIvSOBUDZZ6sbPzt30t6Y+0bkLSZpKtzfy28sV4d0rZZkq5Lgy9dp2ys6PHAZ4D3pDI+XqeuG0u6IH0eCyXtnsr7lKR5ki4FOuWLx7qAv/ntz1JLcD+ygXQgG0dj+4i4R9mIdU9HxM6S1gKuTQnp9cCrgdeRPTG4BDizptyNgW8De6ayNoyIJyTNBVZGxJfTcXvnTjsLODYirpL0GeBE4Pi0b2xEzErdFCeSPR6fdxhwSUR8Lv1VsE69OqRj70zbeiXtA3w+It4l6ZPAzIg4JtVt7Zq6ng18LSKuSY9HX0I2zCfATsAeEfF8s5+92VCcrA2y8YoXpfVfk41HshtwY0Tck7a/DZgx0B8NTASmkY3hfE5E9AErJF1ep/xdgKsHyoqIhuOBS5pINnD/VWnT91h9ZLmBAa1uBraqU8RC4Exlg2D9JCIWSdprkDpMJHvUfhrZSI3jGtUtZx9genqUHWB9pXFIgPlO1NZqTtYGqc86vyEloVX5TWQt3UtqjtufoYejVRPHDMcL6Wcfdf4NR8TVyoZCPQD4vqQvAU8NUoeTgCsi4uDUNXNlk3UYA+xam5TrfG5mLeE+a2vWJcAHU2sVSa9SNpLd1cAhqU97M+DNdc79DfAmSVuncwe6IJ4lm65sNRHxNPBkrj/6vWQzhzRF0pZkY2Z/m+yvhB0b1GEi8EBaPzJXTG3dal9fChyTi7lDs/UzGwkna2vW6WT90bcomyz4W2St2ovIxh3+HfBN6iTViHgUmA1cKOlW4Edp1/8CBw9ctKs57QjgS5JuI7sz4zPDqOtewCJJvwXeBZzSoA5fBL4g6VqyERsHXEHWzbFI0nvq1PU4YGa6ALqEmjtozFrNo+6ZmVWAW9ZmZhXgZG1mVgFO1mZmFeBkbWZWAU7WZmYV4GRtZlYBTtZmZhXwfy3t91GW6sJiAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Find what the model's predictions actually were\n",
    "summary = np.zeros((10,10))\n",
    "\n",
    "for p, truth in zip(preds.copy(), y_test):\n",
    "    p[truth] = 0  # Remove the correct result\n",
    "    summary[truth] += p\n",
    "\n",
    "row_sums = summary.sum(axis=1)\n",
    "summary = summary / row_sums[:, np.newaxis]  \n",
    "\n",
    "# Add a seaborn plot to show the scatter of the 1-10 classes\n",
    "ax = plt.axes()\n",
    "sns.heatmap(summary, ax=ax)\n",
    "ax.set_title(f'Precision of predictions')\n",
    "plt.ylabel(\"Ground Truth\")\n",
    "plt.xlabel(\"Prediction scatter\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "39d85487-0d4f-467a-aa70-fa1c3c5e5833",
   "metadata": {},
   "source": [
    "This plot shows the scatter present in the predictions of the model when not considering the correct answer. Lighter squares show where the model commonly confused different digits, however this gives only general trends"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "3609be3c-eeed-40da-91be-314a4b7174ee",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[4.76100000e+03, 9.00000000e+00, 8.00000000e+00, 4.25442070e-01],\n",
       "       [1.03900000e+03, 7.00000000e+00, 7.00000000e+00, 4.80446577e-01],\n",
       "       [4.24800000e+03, 2.00000000e+00, 2.00000000e+00, 4.84062403e-01],\n",
       "       ...,\n",
       "       [4.10500000e+03, 1.00000000e+00, 1.00000000e+00, 1.00000000e+00],\n",
       "       [4.13000000e+03, 7.00000000e+00, 7.00000000e+00, 1.00000000e+00],\n",
       "       [9.99900000e+03, 6.00000000e+00, 6.00000000e+00, 1.00000000e+00]])"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Create a useful table of results to analyse\n",
    "arr = np.vstack((np.arange(0,10000), y_test, preds.argmax(axis=1), np.max(preds, axis=1))).T\n",
    "# Arr by columns: Index, Truth, Guess, Confidence\n",
    "\n",
    "# Find the 10 most worst classified datapoints (by its own estimation: sort by 4th column)\n",
    "arr = arr[arr[:,3].argsort()]\n",
    "arr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "d6added4-e4ac-413f-bdc6-b100e47a6e2c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Show worst digits classified (Least confident)\n",
    "n = 100\n",
    "heat = np.zeros((10,10))\n",
    "\n",
    "for i, truth, guess, con in arr[:n]:\n",
    "    heat[int(truth), int(guess)] += 1\n",
    "\n",
    "row_sums = heat.sum(axis=1)\n",
    "heat = heat / row_sums[:, np.newaxis] \n",
    "\n",
    "ax = plt.axes()\n",
    "sns.heatmap(heat, ax=ax)\n",
    "ax.set_title(f'The {n} least confident classifications')\n",
    "plt.ylabel(\"Ground Truth\")\n",
    "plt.xlabel(\"Prediction\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2ce36c6f-2a4a-496d-9ea1-e949269fbb4f",
   "metadata": {},
   "source": [
    "This shows the digits that the model was unsure about (Lower confidence spread over multiple classes). However, the diagonal is still the most common prediction, showing the model is accurate."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "fe767318-1b95-4cc4-af44-a760de35b5a9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Least confident digits (Right and wrong)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 12 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Show 12 of these digits\n",
    "cols = 4\n",
    "rows = 3\n",
    "to_plot = arr[:, 0][:rows*cols].astype('int')\n",
    "fig, axs = plt.subplots(rows, cols, figsize=(8, 8))\n",
    "\n",
    "cnt = 0\n",
    "for i in range(rows):\n",
    "    for j in range(cols):\n",
    "        axs[i,j].matshow(x_test[to_plot[cnt]])\n",
    "        axs[i, j].set_title(f'{to_plot[cnt]} ({int(arr[j+i*cols, 2])}->{int(arr[j+i*cols, 1])})')\n",
    "        axs[i, j].xaxis.set_ticklabels([])\n",
    "        axs[i, j].yaxis.set_ticklabels([])\n",
    "        cnt += 1\n",
    "print(\"Least confident digits (Right and wrong)\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "28a81113-8ba1-4716-9350-c323a4bb9a13",
   "metadata": {},
   "source": [
    "Legend: `ID (GUESS -> TRUTH)`\n",
    "\n",
    "These digits are very vague and the model struggles to get them correct as a result. (As it correctly gave a low confidence)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "352ffcad-2f36-4624-aba4-5d2d618c1d94",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Most confident incorrect digits\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 12 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Looking at all the digits that the classifier got wrong, which ones was it most confident about?\n",
    "wrong = arr[arr[:,1] != arr[:,2]]\n",
    "wrong = wrong[wrong[:,3].argsort()[::-1]]  # Set into descending order (Most confident)\n",
    "cols = 4\n",
    "rows = 3\n",
    "\n",
    "wrong_indx = wrong[:, 0].astype('int')\n",
    "fig, axs = plt.subplots(rows, cols, figsize=(8, 8))\n",
    "\n",
    "#fig.tight_layout()\n",
    "cnt = 0\n",
    "for i in range(rows):\n",
    "    for j in range(cols):\n",
    "        axs[i,j].matshow(x_test[wrong_indx[cnt]])\n",
    "        axs[i, j].set_title(f'{wrong_indx[cnt]} ({int(wrong[j+i*cols, 2])}->{int(wrong[j+i*cols, 1])})')\n",
    "        axs[i, j].xaxis.set_ticklabels([])\n",
    "        axs[i, j].yaxis.set_ticklabels([])\n",
    "        cnt += 1\n",
    "print(\"Most confident incorrect digits\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1976e222-efe4-41d8-91a2-04f553b788c6",
   "metadata": {},
   "source": [
    "These are the digits that the model was most confident about, but was wrong. Most of them are very ambigious to me too. Overall, I think the model has done very well to find the most ambiguous digits in the dataset.\n",
    "\n",
    "Examples\n",
    "- 4176 (Looks like a scribble)\n",
    "- 9729 (I would have guessed a 6 too)\n",
    "- 3520 (Looks like neither a 4 nor 6)\n",
    "- 2654 (Appears to be a 1 not a 6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "fefd3d63-2210-4c69-a66d-702ad97aeef3",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Save the 10 most ambigious images to disk (Filename is the index of the image in the test dataset)\n",
    "img_folder = \"imgs/\"\n",
    "for i in range(10):\n",
    "    plt.imsave(img_folder + str(wrong_indx[i]) + \".jpg\", x_test[wrong_indx[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e48acd47-124e-4ead-a1d7-319a7a3dd43a",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:kaggle_base]",
   "language": "python",
   "name": "conda-env-kaggle_base-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}