crnn_test.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "TCC - CRNN Test",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3JPNgmUS5m9Q",
        "outputId": "e9196be9-10a4-4034-eafb-275eb601fd68"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-dataset/raw/main/dataset.zip\n",
        "!unzip -qq dataset.zip"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   153    0   153    0     0    326      0 --:--:-- --:--:-- --:--:--   326\n",
            "100 6958k  100 6958k    0     0  4917k      0  0:00:01  0:00:01 --:--:-- 25.0M\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "02bGf13F5oC2"
      },
      "source": [
        "import os\n",
        "import numpy as np\n",
        "import keras"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "426aZk__5pfN"
      },
      "source": [
        "_rotation = 20 \n",
        "_zoom = 0.5\n",
        "_flip = True\n",
        "_batch_size = 32\n",
        "_random_state = 42\n",
        "_input_shape = (150, 150, 1)\n",
        "_dropout = 0.6\n",
        "_alpha = 1e-5\n",
        "_epochs = 10\n",
        "_hidden_unit = 32\n",
        "_kernel_size = 3\n",
        "_trainable_layer = 'block5_conv1'"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3hw_G64E5tpU"
      },
      "source": [
        "# Data preparation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4-Tq73Pi5rYx"
      },
      "source": [
        "from PIL import Image\n",
        "\n",
        "def select_image(filename):\n",
        "    _image = Image.open(filename)\n",
        "    _image = _image.convert('RGB')\n",
        "    _image = _image.resize((150,150))\n",
        "    return np.asarray(_image)"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5Wl1DCZQ5xbV"
      },
      "source": [
        "def load_class(directory, label_class, images, labels):\n",
        "    for filename in os.listdir(directory):\n",
        "        _path = directory + filename\n",
        "        images.append(_path)\n",
        "        labels.append(label_class)\n",
        "    return images, labels"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9m_RDxJh5y3s"
      },
      "source": [
        "def select_dataset(directory):\n",
        "    _images = list()\n",
        "    _labels = list()\n",
        "    for subdir in os.listdir(directory):\n",
        "        _path = directory + subdir + '/'\n",
        "        if not os.path.isdir(_path):\n",
        "            continue\n",
        "        _images, _labels = load_class(_path, subdir, _images, _labels)\n",
        "    return _images, _labels"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sl8P-co751Rx",
        "outputId": "d3be71ca-b514-4ac2-f6aa-6814c798d3f4"
      },
      "source": [
        "_images, _labels = select_dataset('./data/')\n",
        "print(len(_images), len(_labels))"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5220 5220\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VobjzP2S53SA"
      },
      "source": [
        "from sklearn.preprocessing import LabelBinarizer\n",
        "\n",
        "_labelBinarizer = LabelBinarizer()"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dZIQfAI_54br"
      },
      "source": [
        "from keras.utils import to_categorical\n",
        "\n",
        "def normalize(images, labels):\n",
        "    labels = np.array(labels)\n",
        "    labels = _labelBinarizer.fit_transform(labels)\n",
        "    labels = to_categorical(labels)\n",
        "    return images, labels\n",
        "\n",
        "_images, _labels = normalize(_images, _labels)"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AC8TWaGRFnmT"
      },
      "source": [
        "# Download models"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tOUeX6L8FsfC",
        "outputId": "3c92870f-e1cc-4386-ac09-e8cf0dccb8b5"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-classifier/raw/main/models/CRNN_BiGRU.hdf5"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   167    0   167    0     0    479      0 --:--:-- --:--:-- --:--:--   479\n",
            "100 24.2M  100 24.2M    0     0  6857k      0  0:00:03  0:00:03 --:--:-- 11.7M\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YPHJPWVJGLpm",
        "outputId": "83700537-c5a9-476e-aadc-5ef1298204c9"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-classifier/raw/main/models/CRNN_BiLSTM.hdf5"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   168    0   168    0     0    482      0 --:--:-- --:--:-- --:--:--   482\n",
            "100 27.6M  100 27.6M    0     0  8797k      0  0:00:03  0:00:03 --:--:-- 16.6M\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Stjg9anMGmq_",
        "outputId": "319090ea-d890-4bf6-faad-7bf112d040c2"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-classifier/raw/main/models/CRNN_GRU.hdf5"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   165  100   165    0     0    457      0 --:--:-- --:--:-- --:--:--   457\n",
            "100 19.0M  100 19.0M    0     0  7282k      0  0:00:02  0:00:02 --:--:-- 17.0M\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_0Ebmi-JGpjn",
        "outputId": "358e4d8d-f537-4894-e56d-52daa23c3c33"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-classifier/raw/main/models/CRNN_LSTM.hdf5"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   166  100   166    0     0    475      0 --:--:-- --:--:-- --:--:--   474\n",
            "100 20.7M  100 20.7M    0     0  4978k      0  0:00:04  0:00:04 --:--:-- 7825k\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "n3cd7IXnGquZ",
        "outputId": "77f7b3fe-d1cb-440d-a0b1-93c2124827cc"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-classifier/raw/main/models/CRNN_GRU_GRU.hdf5"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   169    0   169    0     0    485      0 --:--:-- --:--:-- --:--:--   484\n",
            "100 24.3M  100 24.3M    0     0  6696k      0  0:00:03  0:00:03 --:--:-- 11.2M\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "UB2U-hiSGs-1",
        "outputId": "e9027113-c072-48f2-c3d8-e0687175d92f"
      },
      "source": [
        "!curl -LO https://github.com/chsponciano/pocket-parkinson-classifier/raw/main/models/CRNN_LSTM_LSTM.hdf5"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "100   171  100   171    0     0    502      0 --:--:-- --:--:-- --:--:--   501\n",
            "100 27.8M  100 27.8M    0     0  8474k      0  0:00:03  0:00:03 --:--:-- 15.3M\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w9dxOEWkG2Z3"
      },
      "source": [
        "# Testing the models"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0vF6W-NIPi27"
      },
      "source": [
        "from IPython.display import HTML, display\n",
        "\n",
        "def progress_bar(title, subtitle, hits, errors, value, max):\n",
        "    return HTML(\"\"\"\n",
        "        <h3>{title}</h3>\n",
        "        <p>{subtitle}</p>\n",
        "        <p>Hits: {hits} - Errors: {errors}</p>\n",
        "        <progress value='{value}' max='{max}', style='width: 90%'>{value}</progress>\"\"\"\n",
        "    .format(title=title, subtitle=subtitle, hits=hits, errors=errors, value=value, max=max))"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t6HVfwX-Gucf"
      },
      "source": [
        "_models = ['CRNN_GRU', 'CRNN_GRU_GRU', 'CRNN_BiGRU', 'CRNN_LSTM', 'CRNN_LSTM_LSTM', 'CRNN_BiLSTM']"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GZ5ZK7FKH0bs"
      },
      "source": [
        "_results = {}"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 654
        },
        "id": "8IM0SMLFH6dM",
        "outputId": "36eab019-f39a-4562-f8e7-99922dc170fd"
      },
      "source": [
        "from keras.models import load_model\n",
        "\n",
        "_size = len(_images)\n",
        "\n",
        "for model_name in _models:\n",
        "  _panel = display(progress_bar, display_id=True)\n",
        "  _panel.update(progress_bar(model_name, f'Testing the model {model_name}', 0, 0, 0, _size))\n",
        "  _results[model_name] = { 'hits': 0, 'errors': 0  }\n",
        "  _model = load_model(f'/content/{model_name}.hdf5')\n",
        "\n",
        "  for i in range(0, _size):\n",
        "    _image = np.array([select_image(_images[i])]) / 255.0\n",
        "    _predict = _model.predict(_image)[0]\n",
        "    _predict = 0 if _predict[0] > _predict[1] else 1\n",
        "    _results[model_name]['hits' if _labels[i][_predict] == 1.0 else 'errors'] += 1\n",
        "    _panel.update(progress_bar(model_name, f'Testing the model {model_name}', _results[model_name]['hits'], _results[model_name]['errors'], i, _size))"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "        <h3>CRNN_GRU</h3>\n",
              "        <p>Testing the model CRNN_GRU</p>\n",
              "        <p>Hits: 5198 - Errors: 22</p>\n",
              "        <progress value='5219' max='5220', style='width: 90%'>5219</progress>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "        <h3>CRNN_GRU_GRU</h3>\n",
              "        <p>Testing the model CRNN_GRU_GRU</p>\n",
              "        <p>Hits: 5216 - Errors: 4</p>\n",
              "        <progress value='5219' max='5220', style='width: 90%'>5219</progress>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "        <h3>CRNN_BiGRU</h3>\n",
              "        <p>Testing the model CRNN_BiGRU</p>\n",
              "        <p>Hits: 5194 - Errors: 26</p>\n",
              "        <progress value='5219' max='5220', style='width: 90%'>5219</progress>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "        <h3>CRNN_LSTM</h3>\n",
              "        <p>Testing the model CRNN_LSTM</p>\n",
              "        <p>Hits: 5073 - Errors: 147</p>\n",
              "        <progress value='5219' max='5220', style='width: 90%'>5219</progress>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "        <h3>CRNN_LSTM_LSTM</h3>\n",
              "        <p>Testing the model CRNN_LSTM_LSTM</p>\n",
              "        <p>Hits: 5200 - Errors: 20</p>\n",
              "        <progress value='5219' max='5220', style='width: 90%'>5219</progress>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "        <h3>CRNN_BiLSTM</h3>\n",
              "        <p>Testing the model CRNN_BiLSTM</p>\n",
              "        <p>Hits: 5202 - Errors: 18</p>\n",
              "        <progress value='5219' max='5220', style='width: 90%'>5219</progress>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 472
        },
        "id": "L4S1eGJ4szN6",
        "outputId": "c0d08a33-a634-4125-c8a4-6ab03f03da78"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "plt.rcdefaults()\n",
        "\n",
        "_y_pos = np.arange(len(_models))\n",
        "plt.barh(_y_pos, [_results[m]['errors'] for m in _models], align='center', alpha=0.5)\n",
        "plt.yticks(_y_pos, _models)\n",
        "plt.xlabel('Errors')\n",
        "plt.title('Errors the models')\n",
        "plt.show()"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}