04. Spectograms for CV.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Acoustic Anomaly Detection based on Images with Azure AutoML for Images\n",
    "## 04. Spectograms creation\n",
    "\n",
    "<img src='https://github.com/retkowsky/CustomVision_Datasets/blob/master/anomaly.png?raw=true' width = 600>\n",
    "\n",
    "<div class=\"alert alert-info\"> <b> In this notebook We will generate spectograms for all the files</b></div>\n",
    "\n",
    "Serge Retkowsky | serge.retkowsky@microsoft.com | 25-Oct-2022 | V3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "#pip install basic_image_eda"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install --upgrade pillow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import urllib.request\n",
    "import librosa, librosa.display, IPython.display as ipd\n",
    "import numpy as np\n",
    "import wave\n",
    "import os\n",
    "import os.path, time\n",
    "import seaborn\n",
    "import glob\n",
    "import shutil\n",
    "import sys\n",
    "\n",
    "from scipy.io import wavfile as wav\n",
    "from pathlib import Path\n",
    "from datetime import datetime\n",
    "from zipfile import ZipFile\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'3.8.5 (default, Sep  4 2020, 07:30:14) \\n[GCC 7.3.0]'"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sys.version"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2022-10-25 13:05:15\n"
     ]
    }
   ],
   "source": [
    "import datetime\n",
    "\n",
    "print(datetime.datetime.now().strftime(\"%Y-%m-%d %H:%M:%S\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_dir(mydir):\n",
    "    \"\"\"\n",
    "    Creating a directory if needed\n",
    "    \"\"\"\n",
    "    path = Path(mydir)\n",
    "    if path.is_dir():\n",
    "        print(\"Directory\", mydir, 'exists so we will use it.')\n",
    "\n",
    "    else:\n",
    "        os.mkdir(mydir)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generating all the spectograms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "durationsecs = 10\n",
    "hoplength = 512\n",
    "samprate = 44100\n",
    "nfft = 2048"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def spectogram_batch(wavdir, resultdir):\n",
    "    \"\"\"\n",
    "    Spectogram generation\n",
    "    \"\"\"\n",
    "    create_dir(resultdir)\n",
    "\n",
    "    for root, _, files in os.walk(wavdir):\n",
    "        print(\"Directory\", root, \"has\", len(files), \"files.\")\n",
    "\n",
    "    from datetime import datetime\n",
    "    start_time = datetime.now()\n",
    "    print(\"Creating the spectograms for all the wav files from\", wavdir)\n",
    "    print(\"\\nRunning...\\n\")\n",
    "\n",
    "    wavfiles = [file for file in os.listdir(wavdir) if file.endswith(('wav'))]\n",
    "\n",
    "    for wavfile in wavfiles:\n",
    "\n",
    "        plt.figure(figsize=(15, 7))\n",
    "        wavfile = wavdir + wavfile\n",
    "        print(wavfile)\n",
    "\n",
    "        y, sr = librosa.load(wavfile,\n",
    "                             sr=None,\n",
    "                             mono=True,\n",
    "                             offset=0.0,\n",
    "                             duration=durationsecs)\n",
    "        plt.ioff()\n",
    "        fig = plt.figure()\n",
    "\n",
    "        mel_signal = librosa.feature.melspectrogram(y=y,\n",
    "                                                    sr=sr,\n",
    "                                                    hop_length=hoplength,\n",
    "                                                    n_fft=nfft)\n",
    "        spectrogram = np.abs(mel_signal)\n",
    "        power_to_db = librosa.power_to_db(spectrogram, ref=np.max)\n",
    "        librosa.display.specshow(power_to_db,\n",
    "                                 sr=sr,\n",
    "                                 x_axis='time',\n",
    "                                 y_axis='mel',\n",
    "                                 cmap='magma',\n",
    "                                 hop_length=hoplength)\n",
    "\n",
    "        plt.axis('off')\n",
    "        outputfilename = wavfile.split('.')\n",
    "        outputfilename = outputfilename[0].split('/')\n",
    "\n",
    "        savefilename = resultdir + '/' + str(outputfilename[2]) + '.jpg'\n",
    "        plt.savefig(savefilename, dpi=300)\n",
    "        plt.close('all')\n",
    "\n",
    "    print('\\nDone.')\n",
    "    end_time = datetime.now()\n",
    "    print(\"Execution time:\", (end_time - start_time))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "create_dir(\"spectograms\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "create_dir(\"spectograms/OK\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "create_dir(\"spectograms/KO\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Running the spectograms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Directory spectograms/KO/ exists so we will use it.\n",
      "Directory audio_samples/KO/ has 61 files.\n",
      "Creating the spectograms for all the wav files from audio_samples/KO/\n",
      "\n",
      "Running...\n",
      "\n",
      "audio_samples/KO/machine_KO_1.wav\n",
      "audio_samples/KO/machine_KO_10.wav\n",
      "audio_samples/KO/machine_KO_11.wav\n",
      "audio_samples/KO/machine_KO_12.wav\n",
      "audio_samples/KO/machine_KO_13.wav\n",
      "audio_samples/KO/machine_KO_14.wav\n",
      "audio_samples/KO/machine_KO_15.wav\n",
      "audio_samples/KO/machine_KO_16.wav\n",
      "audio_samples/KO/machine_KO_17.wav\n",
      "audio_samples/KO/machine_KO_18.wav\n",
      "audio_samples/KO/machine_KO_19.wav\n",
      "audio_samples/KO/machine_KO_2.wav\n",
      "audio_samples/KO/machine_KO_20.wav\n",
      "audio_samples/KO/machine_KO_21.wav\n",
      "audio_samples/KO/machine_KO_22.wav\n",
      "audio_samples/KO/machine_KO_23.wav\n",
      "audio_samples/KO/machine_KO_24.wav\n",
      "audio_samples/KO/machine_KO_25.wav\n",
      "audio_samples/KO/machine_KO_26.wav\n",
      "audio_samples/KO/machine_KO_27.wav\n",
      "audio_samples/KO/machine_KO_28.wav\n",
      "audio_samples/KO/machine_KO_29.wav\n",
      "audio_samples/KO/machine_KO_3.wav\n",
      "audio_samples/KO/machine_KO_30.wav\n",
      "audio_samples/KO/machine_KO_31.wav\n",
      "audio_samples/KO/machine_KO_32.wav\n",
      "audio_samples/KO/machine_KO_33.wav\n",
      "audio_samples/KO/machine_KO_34.wav\n",
      "audio_samples/KO/machine_KO_35.wav\n",
      "audio_samples/KO/machine_KO_36.wav\n",
      "audio_samples/KO/machine_KO_37.wav\n",
      "audio_samples/KO/machine_KO_38.wav\n",
      "audio_samples/KO/machine_KO_39.wav\n",
      "audio_samples/KO/machine_KO_4.wav\n",
      "audio_samples/KO/machine_KO_40.wav\n",
      "audio_samples/KO/machine_KO_41.wav\n",
      "audio_samples/KO/machine_KO_42.wav\n",
      "audio_samples/KO/machine_KO_43.wav\n",
      "audio_samples/KO/machine_KO_44.wav\n",
      "audio_samples/KO/machine_KO_45.wav\n",
      "audio_samples/KO/machine_KO_46.wav\n",
      "audio_samples/KO/machine_KO_47.wav\n",
      "audio_samples/KO/machine_KO_48.wav\n",
      "audio_samples/KO/machine_KO_49.wav\n",
      "audio_samples/KO/machine_KO_5.wav\n",
      "audio_samples/KO/machine_KO_50.wav\n",
      "audio_samples/KO/machine_KO_51.wav\n",
      "audio_samples/KO/machine_KO_52.wav\n",
      "audio_samples/KO/machine_KO_53.wav\n",
      "audio_samples/KO/machine_KO_54.wav\n",
      "audio_samples/KO/machine_KO_55.wav\n",
      "audio_samples/KO/machine_KO_56.wav\n",
      "audio_samples/KO/machine_KO_57.wav\n",
      "audio_samples/KO/machine_KO_58.wav\n",
      "audio_samples/KO/machine_KO_59.wav\n",
      "audio_samples/KO/machine_KO_6.wav\n",
      "audio_samples/KO/machine_KO_60.wav\n",
      "audio_samples/KO/machine_KO_61.wav\n",
      "audio_samples/KO/machine_KO_7.wav\n",
      "audio_samples/KO/machine_KO_8.wav\n",
      "audio_samples/KO/machine_KO_9.wav\n",
      "\n",
      "Done.\n",
      "Execution time: 0:00:30.591706\n"
     ]
    }
   ],
   "source": [
    "spectogram_batch(\"audio_samples/KO/\", \"spectograms/KO/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Directory spectograms/OK/ exists so we will use it.\n",
      "Directory audio_samples/OK/ has 61 files.\n",
      "Creating the spectograms for all the wav files from audio_samples/OK/\n",
      "\n",
      "Running...\n",
      "\n",
      "audio_samples/OK/machine_OK_1.wav\n",
      "audio_samples/OK/machine_OK_10.wav\n",
      "audio_samples/OK/machine_OK_11.wav\n",
      "audio_samples/OK/machine_OK_12.wav\n",
      "audio_samples/OK/machine_OK_13.wav\n",
      "audio_samples/OK/machine_OK_14.wav\n",
      "audio_samples/OK/machine_OK_15.wav\n",
      "audio_samples/OK/machine_OK_16.wav\n",
      "audio_samples/OK/machine_OK_17.wav\n",
      "audio_samples/OK/machine_OK_18.wav\n",
      "audio_samples/OK/machine_OK_19.wav\n",
      "audio_samples/OK/machine_OK_2.wav\n",
      "audio_samples/OK/machine_OK_20.wav\n",
      "audio_samples/OK/machine_OK_21.wav\n",
      "audio_samples/OK/machine_OK_22.wav\n",
      "audio_samples/OK/machine_OK_23.wav\n",
      "audio_samples/OK/machine_OK_24.wav\n",
      "audio_samples/OK/machine_OK_25.wav\n",
      "audio_samples/OK/machine_OK_26.wav\n",
      "audio_samples/OK/machine_OK_27.wav\n",
      "audio_samples/OK/machine_OK_28.wav\n",
      "audio_samples/OK/machine_OK_29.wav\n",
      "audio_samples/OK/machine_OK_3.wav\n",
      "audio_samples/OK/machine_OK_30.wav\n",
      "audio_samples/OK/machine_OK_31.wav\n",
      "audio_samples/OK/machine_OK_32.wav\n",
      "audio_samples/OK/machine_OK_33.wav\n",
      "audio_samples/OK/machine_OK_34.wav\n",
      "audio_samples/OK/machine_OK_35.wav\n",
      "audio_samples/OK/machine_OK_36.wav\n",
      "audio_samples/OK/machine_OK_37.wav\n",
      "audio_samples/OK/machine_OK_38.wav\n",
      "audio_samples/OK/machine_OK_39.wav\n",
      "audio_samples/OK/machine_OK_4.wav\n",
      "audio_samples/OK/machine_OK_40.wav\n",
      "audio_samples/OK/machine_OK_41.wav\n",
      "audio_samples/OK/machine_OK_42.wav\n",
      "audio_samples/OK/machine_OK_43.wav\n",
      "audio_samples/OK/machine_OK_44.wav\n",
      "audio_samples/OK/machine_OK_45.wav\n",
      "audio_samples/OK/machine_OK_46.wav\n",
      "audio_samples/OK/machine_OK_47.wav\n",
      "audio_samples/OK/machine_OK_48.wav\n",
      "audio_samples/OK/machine_OK_49.wav\n",
      "audio_samples/OK/machine_OK_5.wav\n",
      "audio_samples/OK/machine_OK_50.wav\n",
      "audio_samples/OK/machine_OK_51.wav\n",
      "audio_samples/OK/machine_OK_52.wav\n",
      "audio_samples/OK/machine_OK_53.wav\n",
      "audio_samples/OK/machine_OK_54.wav\n",
      "audio_samples/OK/machine_OK_55.wav\n",
      "audio_samples/OK/machine_OK_56.wav\n",
      "audio_samples/OK/machine_OK_57.wav\n",
      "audio_samples/OK/machine_OK_58.wav\n",
      "audio_samples/OK/machine_OK_59.wav\n",
      "audio_samples/OK/machine_OK_6.wav\n",
      "audio_samples/OK/machine_OK_60.wav\n",
      "audio_samples/OK/machine_OK_61.wav\n",
      "audio_samples/OK/machine_OK_7.wav\n",
      "audio_samples/OK/machine_OK_8.wav\n",
      "audio_samples/OK/machine_OK_9.wav\n",
      "\n",
      "Done.\n",
      "Execution time: 0:00:29.402427\n"
     ]
    }
   ],
   "source": [
    "spectogram_batch(\"audio_samples/OK/\", \"spectograms/OK/\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Checking the number of generated files"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Directory spectograms/OK/ has 61 spectogram jpg files.\n"
     ]
    }
   ],
   "source": [
    "for root, _, files in os.walk('spectograms/OK/'):\n",
    "    print(\"Directory\", root, \"has\", len(files), \"spectogram jpg files.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Directory spectograms/KO/ has 61 spectogram jpg files.\n"
     ]
    }
   ],
   "source": [
    "for root, _, files in os.walk('spectograms/KO/'):\n",
    "    print(\"Directory\", root, \"has\", len(files), \"spectogram jpg files.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## EDA On the images to check the shape of each spectograms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def image_eda(data_dir, option):\n",
    "    \"\"\"\n",
    "    Images EDA\n",
    "    \"\"\"\n",
    "    from basic_image_eda import BasicImageEDA\n",
    "\n",
    "    if option == 'basic':\n",
    "        BasicImageEDA.explore(data_dir)  # Use this for basic Image EDA\n",
    "\n",
    "    if option == 'advanced':\n",
    "        extensions = ['jpg']\n",
    "        threads = 0\n",
    "        dimension_plot = True\n",
    "        channel_hist = True\n",
    "        nonzero = False\n",
    "        hw_division_factor = 1.0\n",
    "\n",
    "        BasicImageEDA.explore(data_dir, extensions, threads, dimension_plot,\n",
    "                              channel_hist, nonzero, hw_division_factor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "found 61 images.\n",
      "Using 8 threads. (max:8)\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 61/61 [00:02<00:00, 29.52it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "*--------------------------------------------------------------------------------------*\n",
      "number of images                         |  61\n",
      "\n",
      "dtype                                    |  uint8\n",
      "channels                                 |  [3]\n",
      "extensions                               |  ['jpg']\n",
      "\n",
      "min height                               |  1200\n",
      "max height                               |  1200\n",
      "mean height                              |  1200.0\n",
      "median height                            |  1200\n",
      "\n",
      "min width                                |  1800\n",
      "max width                                |  1800\n",
      "mean width                               |  1800.0\n",
      "median width                             |  1800\n",
      "\n",
      "mean height/width ratio                  |  0.6666666666666666\n",
      "median height/width ratio                |  0.6666666666666666\n",
      "recommended input size(by mean)          |  [1200 1800] (h x w, multiples of 8)\n",
      "recommended input size(by mean)          |  [1200 1792] (h x w, multiples of 16)\n",
      "recommended input size(by mean)          |  [1216 1792] (h x w, multiples of 32)\n",
      "\n",
      "channel mean(0~1)                        |  [0.8595011  0.59775245 0.65449506]\n",
      "channel std(0~1)                         |  [0.23223472 0.3655816  0.30226487]\n",
      "*--------------------------------------------------------------------------------------*\n",
      "eda ended in 00 hours 00 minutes 02 seconds\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "image_eda(\"./spectograms/OK\",  \"basic\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "found 61 images.\n",
      "Using 8 threads. (max:8)\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 61/61 [00:01<00:00, 32.31it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "*--------------------------------------------------------------------------------------*\n",
      "number of images                         |  61\n",
      "\n",
      "dtype                                    |  uint8\n",
      "channels                                 |  [3]\n",
      "extensions                               |  ['jpg']\n",
      "\n",
      "min height                               |  1200\n",
      "max height                               |  1200\n",
      "mean height                              |  1200.0\n",
      "median height                            |  1200\n",
      "\n",
      "min width                                |  1800\n",
      "max width                                |  1800\n",
      "mean width                               |  1800.0\n",
      "median width                             |  1800\n",
      "\n",
      "mean height/width ratio                  |  0.6666666666666666\n",
      "median height/width ratio                |  0.6666666666666666\n",
      "recommended input size(by mean)          |  [1200 1800] (h x w, multiples of 8)\n",
      "recommended input size(by mean)          |  [1200 1792] (h x w, multiples of 16)\n",
      "recommended input size(by mean)          |  [1216 1792] (h x w, multiples of 32)\n",
      "\n",
      "channel mean(0~1)                        |  [0.84513074 0.58485675 0.65627944]\n",
      "channel std(0~1)                         |  [0.24067363 0.3731367  0.30098653]\n",
      "*--------------------------------------------------------------------------------------*\n",
      "eda ended in 00 hours 00 minutes 02 seconds\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "image_eda(\"./spectograms/KO\", \"basic\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'Number of Spectograms')"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x504 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "number_classes = {\n",
    "    'OK': len(os.listdir('spectograms/OK')),\n",
    "    'KO': len(os.listdir('spectograms/KO'))\n",
    "}\n",
    "\n",
    "plt.figure(figsize=(15, 7))\n",
    "plt.bar(number_classes.keys(), number_classes.values(), width=.5)\n",
    "plt.title(\"Number of MEL spectograms by Class\")\n",
    "plt.xlabel('Class')\n",
    "plt.ylabel('Number of Spectograms')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Zipping the files"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "create_dir(\"archive\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/mnt/batch/tasks/shared/LS_root/mounts/clusters/standardd13v2/code/Users/seretkow/Sound/MachineSoundAnomaly/archive/spectograms.zip'"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "shutil.make_archive(\"archive/spectograms\", 'zip', \"spectograms/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-rwxrwxrwx 1 root root 52M Oct 25 13:08 archive/spectograms.zip\r\n"
     ]
    }
   ],
   "source": [
    "!ls archive/spectograms.zip -lh"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Click to download the file:\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<a href='archive/spectograms.zip' target='_blank'>archive/spectograms.zip</a><br>"
      ],
      "text/plain": [
       "/mnt/batch/tasks/shared/LS_root/mounts/clusters/standardd13v2/code/Users/seretkow/Sound/MachineSoundAnomaly/archive/spectograms.zip"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import FileLink\n",
    "\n",
    "LINK = FileLink(path='archive/spectograms.zip')\n",
    "print(\"Click to download the file:\")\n",
    "LINK"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now you can use Azure Custom Vision to create an image classification model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "> End. You can go to the next notebook"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note: All these Python notebooks were made for demo purposes. They were not designed for production usage. Some feature can be currently in public preview. This preview version is provided without a service-level agreement. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews. https://azure.microsoft.com/en-us/support/legal/preview-supplemental-terms/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernel_info": {
   "name": "python3-azureml"
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  },
  "nteract": {
   "version": "nteract-front-end@1.0.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}