QuickDraw_starter.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "QuickDraw_starter.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [
        "Z5sNJtPHRxD2",
        "r8kBfuO9Br_e",
        "1lDKL1KadkvU"
      ]
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "metadata": {
        "id": "CxpafRfNXddC",
        "colab_type": "code",
        "outputId": "827963b8-7221-4722-ffcc-c299f89bec9e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 237
        }
      },
      "cell_type": "code",
      "source": [
        "# fastai 0.7.0\n",
        "!mkdir src -p && cd src && git clone https://github.com/fastai/fastai.git\n",
        "!pip3 install -q bcolz graphviz sklearn_pandas isoweek pandas_summary ipywidgets torch torchvision torchtext\n",
        "import sys\n",
        "sys.path.append(\"/content/src/fastai/old\") # on windows use \\'s instead\n",
        "from fastai.conv_learner import *\n",
        "from fastai.dataset import *\n",
        "import gc\n",
        "print(torch.__version__)\n",
        "torch.backends.cudnn.benchmark=True\n",
        "torch.cuda.is_available(), torch.backends.cudnn.enabled"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Cloning into 'fastai'...\n",
            "remote: Enumerating objects: 32, done.\u001b[K\n",
            "remote: Counting objects:   3% (1/32)   \u001b[K\rremote: Counting objects:   6% (2/32)   \u001b[K\rremote: Counting objects:   9% (3/32)   \u001b[K\rremote: Counting objects:  12% (4/32)   \u001b[K\rremote: Counting objects:  15% (5/32)   \u001b[K\rremote: Counting objects:  18% (6/32)   \u001b[K\rremote: Counting objects:  21% (7/32)   \u001b[K\rremote: Counting objects:  25% (8/32)   \u001b[K\rremote: Counting objects:  28% (9/32)   \u001b[K\rremote: Counting objects:  31% (10/32)   \u001b[K\rremote: Counting objects:  34% (11/32)   \u001b[K\rremote: Counting objects:  37% (12/32)   \u001b[K\rremote: Counting objects:  40% (13/32)   \u001b[K\rremote: Counting objects:  43% (14/32)   \u001b[K\rremote: Counting objects:  46% (15/32)   \u001b[K\rremote: Counting objects:  50% (16/32)   \u001b[K\rremote: Counting objects:  53% (17/32)   \u001b[K\rremote: Counting objects:  56% (18/32)   \u001b[K\rremote: Counting objects:  59% (19/32)   \u001b[K\rremote: Counting objects:  62% (20/32)   \u001b[K\rremote: Counting objects:  65% (21/32)   \u001b[K\rremote: Counting objects:  68% (22/32)   \u001b[K\rremote: Counting objects:  71% (23/32)   \u001b[K\rremote: Counting objects:  75% (24/32)   \u001b[K\rremote: Counting objects:  78% (25/32)   \u001b[K\rremote: Counting objects:  81% (26/32)   \u001b[K\rremote: Counting objects:  84% (27/32)   \u001b[K\rremote: Counting objects:  87% (28/32)   \u001b[K\rremote: Counting objects:  90% (29/32)   \u001b[K\rremote: Counting objects:  93% (30/32)   \u001b[K\rremote: Counting objects:  96% (31/32)   \u001b[K\rremote: Counting objects: 100% (32/32)   \u001b[K\rremote: Counting objects: 100% (32/32), done.\u001b[K\n",
            "remote: Compressing objects: 100% (24/24), done.\u001b[K\n",
            "remote: Total 16102 (delta 17), reused 18 (delta 8), pack-reused 16070\u001b[K\n",
            "Receiving objects: 100% (16102/16102), 298.57 MiB | 27.68 MiB/s, done.\n",
            "Resolving deltas: 100% (10829/10829), done.\n",
            "Checking out files: 100% (695/695), done.\n",
            "tcmalloc: large alloc 1073750016 bytes == 0x58b94000 @  0x7f7e89a022a4 0x591a07 0x5b5d56 0x502e9a 0x506859 0x502209 0x502f3d 0x506859 0x504c28 0x502540 0x502f3d 0x506859 0x504c28 0x502540 0x502f3d 0x506859 0x504c28 0x502540 0x502f3d 0x507641 0x502209 0x502f3d 0x506859 0x504c28 0x502540 0x502f3d 0x507641 0x504c28 0x502540 0x502f3d 0x507641\n",
            "0.4.1\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(True, True)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "metadata": {
        "id": "RUNwlLO6XdgL",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "!git config --global user.email 'nikshuang@163.com'\n",
        "!git config --global user.name 'Alex Huang'\n",
        "!pip install -q kaggle"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "aBr0TxKdYLjl",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "PATH = Path('data/quick_draw')\n",
        "TRN_TXT_PATH = PATH/'txt'\n",
        "TRN_PATH = PATH/'train'\n",
        "TEST_PATH = PATH/'test'\n",
        "DISK_PATH = 'drive/quick_draw'\n",
        "os.makedirs(DISK_PATH, exist_ok=True)\n",
        "os.makedirs(PATH, exist_ok=True)\n",
        "os.makedirs(TRN_PATH, exist_ok=True)\n",
        "os.makedirs(TEST_PATH, exist_ok=True)\n",
        "os.makedirs(TRN_TXT_PATH, exist_ok=True)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "scB_smh8Y4FZ",
        "colab_type": "code",
        "outputId": "fa4282a4-ccaf-4d19-bf5e-b0025e6d23c8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 72
        }
      },
      "cell_type": "code",
      "source": [
        "kaggle competitions download -c human-protein-atlas-image-classification -p {PATH}\n",
        "!cd {PATH}/'tx && unzip -q ../train_simplified.zip"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "test_simplified.csv: Skipping, found more recently modified local copy (use --force to force download)\n",
            "train_simplified.zip: Skipping, found more recently modified local copy (use --force to force download)\n",
            "replace fence.csv? [y]es, [n]o, [A]ll, [N]one, [r]ename: "
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "Z5sNJtPHRxD2",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Prepare data: load from *.png"
      ]
    },
    {
      "metadata": {
        "id": "7k5DEGoRTGsa",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "%%time\n",
        "def get_count(path): return (path.stem, pd.read_csv(path).shape[0])\n",
        "with ThreadPoolExecutor(2) as e: counts = list(e.map(get_count, TRN_PATH.iterdir()))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "UWufBbJUR4Vl",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "counts = sorted(counts, key=lambda x: x[1])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "PSYCT6ebYWIK",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "最小的数据集和最大的数据集相差有20多万个样本，数据分布极度不均"
      ]
    },
    {
      "metadata": {
        "id": "ta7_6oslX999",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "counts[:10], counts[-10:]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "xhSRgTZLZ7sl",
        "colab_type": "code",
        "outputId": "cd388343-892a-49cc-fc55-a02dca4f57af",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 288
        }
      },
      "cell_type": "code",
      "source": [
        "df = pd.read_csv(TRN_PATH/'blueberry.csv')\n",
        "df.head()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>countrycode</th>\n",
              "      <th>drawing</th>\n",
              "      <th>key_id</th>\n",
              "      <th>recognized</th>\n",
              "      <th>timestamp</th>\n",
              "      <th>word</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>TH</td>\n",
              "      <td>[[[47, 39, 22, 12, 0, 0, 2, 7, 17, 42, 50, 62,...</td>\n",
              "      <td>6382699784699904</td>\n",
              "      <td>True</td>\n",
              "      <td>2017-03-16 09:41:33.474350</td>\n",
              "      <td>blueberry</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>IT</td>\n",
              "      <td>[[[222, 207, 196, 190, 191, 194, 208, 242, 252...</td>\n",
              "      <td>6046664550776832</td>\n",
              "      <td>True</td>\n",
              "      <td>2017-01-23 15:11:32.747510</td>\n",
              "      <td>blueberry</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>US</td>\n",
              "      <td>[[[189, 186, 170, 126, 82, 60, 31, 8, 1, 0, 11...</td>\n",
              "      <td>5715196523839488</td>\n",
              "      <td>True</td>\n",
              "      <td>2017-01-25 17:51:09.581080</td>\n",
              "      <td>blueberry</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>US</td>\n",
              "      <td>[[[109, 84, 57, 44, 19, 10, 1, 0, 4, 14, 32, 4...</td>\n",
              "      <td>5147758375731200</td>\n",
              "      <td>True</td>\n",
              "      <td>2017-03-19 13:06:15.610220</td>\n",
              "      <td>blueberry</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>US</td>\n",
              "      <td>[[[77, 57, 26, 14, 2, 0, 2, 14, 33, 62, 86, 12...</td>\n",
              "      <td>6728861146415104</td>\n",
              "      <td>True</td>\n",
              "      <td>2017-03-06 16:01:44.166140</td>\n",
              "      <td>blueberry</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "  countrycode                                            drawing  \\\n",
              "0          TH  [[[47, 39, 22, 12, 0, 0, 2, 7, 17, 42, 50, 62,...   \n",
              "1          IT  [[[222, 207, 196, 190, 191, 194, 208, 242, 252...   \n",
              "2          US  [[[189, 186, 170, 126, 82, 60, 31, 8, 1, 0, 11...   \n",
              "3          US  [[[109, 84, 57, 44, 19, 10, 1, 0, 4, 14, 32, 4...   \n",
              "4          US  [[[77, 57, 26, 14, 2, 0, 2, 14, 33, 62, 86, 12...   \n",
              "\n",
              "             key_id  recognized                   timestamp       word  \n",
              "0  6382699784699904        True  2017-03-16 09:41:33.474350  blueberry  \n",
              "1  6046664550776832        True  2017-01-23 15:11:32.747510  blueberry  \n",
              "2  5715196523839488        True  2017-01-25 17:51:09.581080  blueberry  \n",
              "3  5147758375731200        True  2017-03-19 13:06:15.610220  blueberry  \n",
              "4  6728861146415104        True  2017-03-06 16:01:44.166140  blueberry  "
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 34
        }
      ]
    },
    {
      "metadata": {
        "id": "S-wxpUQXZDmi",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# https://www.kaggle.com/gaborfodor/greyscale-mobilenet-lb-0-892\n",
        "BASE_SIZE = 256\n",
        "\n",
        "def draw_cv2(raw_strokes, size=256, lw=4, time_color=False):\n",
        "    img = np.zeros((BASE_SIZE, BASE_SIZE), np.uint8)\n",
        "    for t, stroke in enumerate(raw_strokes):\n",
        "        for i in range(len(stroke[0]) - 1):\n",
        "            color = 255 - min(t, 10) * 13 if time_color else 255\n",
        "            _ = cv2.line(img, (stroke[0][i], stroke[1][i]),\n",
        "                         (stroke[0][i + 1], stroke[1][i + 1]), color, lw)\n",
        "    img = cv2.copyMakeBorder(img, 4, 4, 4, 4, cv2.BORDER_CONSTANT)\n",
        "    if size != BASE_SIZE:\n",
        "        return cv2.resize(img, (size, size), interpolation=cv2.INTER_LINEAR)\n",
        "    return img"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "sKppm7Ng1n7I",
        "colab_type": "code",
        "outputId": "7769b0dd-8e7b-402d-b931-5aa374a0c6b6",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 367
        }
      },
      "cell_type": "code",
      "source": [
        "plt.imshow(draw_cv2(eval(df.loc[4, 'drawing']), size=256))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f57752ea898>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 36
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAVEAAAFNCAYAAAC5YlyiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f5790d18278>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "67patVMXZvvt",
        "colab_type": "code",
        "outputId": "2e9c5793-b3e4-47ba-9f28-e29d904e3bd3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 367
        }
      },
      "cell_type": "code",
      "source": [
        "plt.imshow(draw_cv2(eval(df.loc[4, 'drawing']), size=128))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f577524f668>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 37
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAVEAAAFNCAYAAAC5YlyiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAElFTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f577528bcf8>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "AYe1kYWncvh-",
        "colab_type": "code",
        "outputId": "b8100e5c-8858-4927-9925-d3ffa94d1c36",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 367
        }
      },
      "cell_type": "code",
      "source": [
        "plt.imshow(draw_cv2(eval(df.loc[4, 'drawing']), size=64))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f5775235518>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 38
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFMCAYAAABCsp4mAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAGEJAAYISwAwQFgCgIH/AzHaZEGHCtzPAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f57752ac240>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "yvMzfM_tdBlg",
        "colab_type": "code",
        "outputId": "04e60d92-e282-40c7-c418-ee9b1e91670f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 365
        }
      },
      "cell_type": "code",
      "source": [
        "plt.imshow(draw_cv2(eval(df.loc[4, 'drawing']), size=32))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f5775199400>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 39
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFKCAYAAACU6307AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f577520b470>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "y5_OTxYL0k55",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "k = 250\n",
        "sz = 128\n",
        "mn = f'mobilenet_{sz}sz_{k}'\n",
        "dn = f'quickdraw_data_{sz}sz_{k}.tar.bz2'\n",
        "FN_PKL = f'fnames_{sz}sz_{k}.pkl'\n",
        "VAL_PKL = f'val_idxs_{sz}sz_{k}.pkl'"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "AlUJSG_-g1N3",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def df_to_ims(path):\n",
        "  df = pd.read_csv(path)\n",
        "  selected = df[df.recognized == True].sample(k)\n",
        "  for row in selected.iterrows():\n",
        "    kid, drawing, label = row[1].key_id, eval(row[1].drawing), '_'.join(row[1].word.split())\n",
        "    ary = draw_cv2(drawing, size=sz)\n",
        "    rgb_ary = np.repeat(ary[:, :, None], 3, -1)\n",
        "    plt.imsave(f'{PATH}/train_{sz}/{label}-{kid}.png', rgb_ary)\n",
        "#     PIL.Image.fromarray(rgb_ary).save(f'{PATH}/train_{sz}/{label}-{kid}.png')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "_ZDAjTIXmQTS",
        "colab_type": "code",
        "outputId": "b4460163-9f7d-4760-dcb4-e15319703ddd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "cell_type": "code",
      "source": [
        "%%time\n",
        "shutil.rmtree(PATH/f'train_{sz}', ignore_errors=True)\n",
        "os.makedirs(PATH/f'train_{sz}', exist_ok=True)\n",
        "with ThreadPoolExecutor(2) as e: e.map(df_to_ims, TRN_PATH.iterdir())"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CPU times: user 11min 24s, sys: 44.1 s, total: 12min 8s\n",
            "Wall time: 9min 39s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "3Q5nu91kgeHx",
        "colab_type": "code",
        "outputId": "6d4f2925-fa9e-4e07-fd89-97d86a60ee10",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        }
      },
      "cell_type": "code",
      "source": [
        "!ls {PATH}/train_{sz} -l | wc -l\n",
        "!ls {PATH}/train_{sz} -lhS | head -5"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "85001\n",
            "total 353M\n",
            "-rw-r--r-- 1 root root 8.4K Nov 10 01:32 camouflage-5968440256888832.png\n",
            "-rw-r--r-- 1 root root 7.9K Nov 10 01:29 squiggle-5598703249260544.png\n",
            "-rw-r--r-- 1 root root 7.7K Nov 10 01:30 blackberry-4885994497114112.png\n",
            "-rw-r--r-- 1 root root 7.6K Nov 10 01:30 blackberry-5734184150630400.png\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "NtPU7zaHX-Jr",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# PIL.Image.open(PATH/f'train_{sz}/blackberry-4927079281852416.png')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "uVjxFvAWX-HR",
        "colab_type": "code",
        "outputId": "15d25204-e25f-4202-ac1e-36a1473b595f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 203
        }
      },
      "cell_type": "code",
      "source": [
        "df = pd.read_csv(PATH/'test_simplified.csv')\n",
        "df.head()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>key_id</th>\n",
              "      <th>countrycode</th>\n",
              "      <th>drawing</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>9000003627287624</td>\n",
              "      <td>DE</td>\n",
              "      <td>[[[17, 18, 20, 25, 137, 174, 242, 249, 251, 25...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>9000010688666847</td>\n",
              "      <td>UA</td>\n",
              "      <td>[[[174, 145, 106, 38, 11, 4, 4, 15, 29, 78, 16...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>9000023642890129</td>\n",
              "      <td>BG</td>\n",
              "      <td>[[[0, 12, 14, 17, 16, 24, 55, 57, 60, 79, 82, ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>9000038588854897</td>\n",
              "      <td>US</td>\n",
              "      <td>[[[0, 9, 23, 40, 54, 60, 81, 105, 123, 167, 20...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>9000052667981386</td>\n",
              "      <td>AR</td>\n",
              "      <td>[[[87, 82, 71, 63, 66, 92, 96, 95], [220, 218,...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "             key_id countrycode  \\\n",
              "0  9000003627287624          DE   \n",
              "1  9000010688666847          UA   \n",
              "2  9000023642890129          BG   \n",
              "3  9000038588854897          US   \n",
              "4  9000052667981386          AR   \n",
              "\n",
              "                                             drawing  \n",
              "0  [[[17, 18, 20, 25, 137, 174, 242, 249, 251, 25...  \n",
              "1  [[[174, 145, 106, 38, 11, 4, 4, 15, 29, 78, 16...  \n",
              "2  [[[0, 12, 14, 17, 16, 24, 55, 57, 60, 79, 82, ...  \n",
              "3  [[[0, 9, 23, 40, 54, 60, 81, 105, 123, 167, 20...  \n",
              "4  [[[87, 82, 71, 63, 66, 92, 96, 95], [220, 218,...  "
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 47
        }
      ]
    },
    {
      "metadata": {
        "id": "NGlcXF7G_yBU",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def test_df_to_ims(path):\n",
        "  df = pd.read_csv(path)\n",
        "  for row in df.iterrows():\n",
        "    kid, drawing = row[1].key_id, eval(row[1].drawing)\n",
        "    ary = draw_cv2(drawing, size=sz)\n",
        "    rgb_ary = np.repeat(ary[:, :, None], 3, -1)\n",
        "    plt.imsave(f'{PATH}/test_{sz}/{kid}.png', rgb_ary)\n",
        "#     PIL.Image.fromarray(rgb_ary).save(f'{PATH}/test_{sz}/{kid}.png')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "ByNbalk5AiiK",
        "colab_type": "code",
        "outputId": "0138a316-0619-48a4-eb54-c6f63e27efab",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "cell_type": "code",
      "source": [
        "%%time\n",
        "shutil.rmtree(PATH/f'test_{sz}', ignore_errors=True)\n",
        "os.makedirs(PATH/f'test_{sz}', exist_ok=True)\n",
        "test_df_to_ims(PATH/'test_simplified.csv')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CPU times: user 5min 33s, sys: 8.72 s, total: 5min 42s\n",
            "Wall time: 5min 42s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "6fU--WMlAidl",
        "colab_type": "code",
        "outputId": "104c0625-4b6e-4040-8239-8d4e8dbb91ff",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        }
      },
      "cell_type": "code",
      "source": [
        "!ls {PATH}/test_{sz} -l | wc -l\n",
        "!ls -lhS {PATH}/test_{sz} | head -5"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "112200\n",
            "total 501M\n",
            "-rw-r--r-- 1 root root 7.7K Nov 10 01:41 9831494525348231.png\n",
            "-rw-r--r-- 1 root root 7.6K Nov 10 01:39 9525461686320036.png\n",
            "-rw-r--r-- 1 root root 7.5K Nov 10 01:40 9679715569692673.png\n",
            "-rw-r--r-- 1 root root 7.5K Nov 10 01:39 9534292565812127.png\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "_PvcmGwhAiZR",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# PIL.Image.open(f'{PATH}/test_{sz}/9985742567546855.png')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "OyOn5-raCvHA",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "%%time\n",
        "!cd {PATH} && tar jcf {dn} train_{sz} test_{sz}\n",
        "shutil.copy(f'{PATH}/{dn}', DISK_PATH)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "6R9nJHUl92vW",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "image_exts = ['.jpg', '.jpeg', '.png']\n",
        "def get_image_files(path, check_ext=True, recurse=False):\n",
        "  return [o for o in Path(path).glob('**/*' if recurse else '*')\n",
        "          if not o.name.startswith('.') and not o.is_dir()\n",
        "          and (not check_ext or o.suffix in image_exts)]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "4QIZjgASOwSq",
        "colab_type": "code",
        "outputId": "e3458e8f-bfcd-4b70-bb3c-d05ed1e4e97a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        }
      },
      "cell_type": "code",
      "source": [
        "fnames = get_image_files(PATH/f'train_{sz}')\n",
        "val_idxs = get_cv_idxs(len(fnames))\n",
        "len(val_idxs), val_idxs[:2], len(fnames), fnames[:3]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(17000,\n",
              " array([20569,  6186]),\n",
              " 85000,\n",
              " [PosixPath('data/quick_draw/train_128/animal_migration-6466977700773888.png'),\n",
              "  PosixPath('data/quick_draw/train_128/bandage-6028172833325056.png'),\n",
              "  PosixPath('data/quick_draw/train_128/fan-5967909526437888.png')])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 52
        }
      ]
    },
    {
      "metadata": {
        "id": "-HZvjNNrav6a",
        "colab_type": "code",
        "outputId": "5410724f-5f45-4a7a-b56d-fcd7b63f11c8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "with open(FN_PKL, 'wb') as fp:\n",
        "  pickle.dump(fnames, fp)\n",
        "with open(VAL_PKL, 'wb') as fp:\n",
        "  pickle.dump(val_idxs, fp)\n",
        "shutil.copy(f'{FN_PKL}', DISK_PATH)\n",
        "shutil.copy(f'{VAL_PKL}', DISK_PATH)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'drive/quick_draw/val_idxs_128sz_250.pkl'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 53
        }
      ]
    },
    {
      "metadata": {
        "colab_type": "text",
        "id": "r8kBfuO9Br_e"
      },
      "cell_type": "markdown",
      "source": [
        "### Prepare data: load from *.txt"
      ]
    },
    {
      "metadata": {
        "colab_type": "code",
        "id": "bbiGvNGWBsAM",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "NUM_SAMPLES_PER_CLASS = 200\n",
        "\n",
        "def create_train_txts_from_df(path):\n",
        "    df = pd.read_csv(path, nrows=NUM_SAMPLES_PER_CLASS)\n",
        "    klass = '_'.join(path.stem.split())\n",
        "    drawings = df.drawing.values\n",
        "    for i, drawing in enumerate(drawings):\n",
        "        fpth = TRN_PATH/f'{klass}-{i}.txt'\n",
        "        if fpth.exists(): continue\n",
        "        with open(fpth, mode='w') as f: f.write(drawing)\n",
        "\n",
        "def create_test_txts_from_df(path):\n",
        "    df = pd.read_csv(path)\n",
        "    kids = df.key_id.values\n",
        "    drawings = df.drawing.values\n",
        "    for kid, drawing in zip(kids, drawings):\n",
        "        with open(TEST_PATH/f'{kid}.txt', mode='w') as f: f.write(drawing)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "colab_type": "code",
        "outputId": "dc2483c4-5b64-48f9-b7c7-c427609e8f8c",
        "id": "KlE1E4fvBsAS",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "cell_type": "code",
      "source": [
        "%%time\n",
        "with ThreadPoolExecutor(2) as e: e.map(create_train_txts_from_df, TRN_TXT_PATH.iterdir())"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CPU times: user 12.4 s, sys: 11.2 s, total: 23.6 s\n",
            "Wall time: 15.4 s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "colab_type": "code",
        "id": "oj_p2o57BsAb",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "create_test_txts_from_df(PATH/'test_simplified.csv')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "colab_type": "code",
        "outputId": "53a375e9-4e6b-419a-fc47-3c36f115dfdf",
        "id": "Dge-D6OzBsAf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 108
        }
      },
      "cell_type": "code",
      "source": [
        "drawing_exts = ['.txt']\n",
        "def get_image_files(path, check_ext=True, recurse=False):\n",
        "  return [o for o in Path(path).glob('**/*' if recurse else '*')\n",
        "          if not o.name.startswith('.') and not o.is_dir()\n",
        "          and (not check_ext or o.suffix in drawing_exts)]\n",
        "\n",
        "fnames = get_image_files(TRN_PATH)\n",
        "test_fnames = get_image_files(TEST_PATH)\n",
        "val_idxs = get_cv_idxs(len(fnames))\n",
        "len(val_idxs), val_idxs[:2], len(fnames), fnames[:2]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(13600,\n",
              " array([23742, 59794]),\n",
              " 68000,\n",
              " [PosixPath('data/quick_draw/train/pliers-101.txt'),\n",
              "  PosixPath('data/quick_draw/train/grapes-137.txt')])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 79
        }
      ]
    },
    {
      "metadata": {
        "colab_type": "code",
        "outputId": "eda2356c-ea73-4ab1-a2d9-720c0eea2ce1",
        "id": "oELSvOgMBsAi",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "cell_type": "code",
      "source": [
        "labels = np.array([o.name.split('-')[0] for o in fnames])\n",
        "len(labels), labels[:5]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(68000,\n",
              " array(['pliers', 'grapes', 'spreadsheet', 'calendar', 'pool'], dtype='<U23'))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 80
        }
      ]
    },
    {
      "metadata": {
        "id": "lFZx5j6jFfLT",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "classes = np.unique(labels)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "colab_type": "code",
        "outputId": "6a2235fa-6a40-4309-a851-e141266997e1",
        "id": "3o3SzV8bBsAq",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "cell_type": "code",
      "source": [
        "x_names = np.array(['train/' + o.name for o in fnames])\n",
        "test_x_names = np.array(['test/' + o.name for o in test_fnames])\n",
        "x_names[:2], test_x_names[:2]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(array(['train/pliers-101.txt', 'train/grapes-137.txt'], dtype='<U37'),\n",
              " array(['test/9294575425747848.txt', 'test/9854825525933368.txt'], dtype='<U25'))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 82
        }
      ]
    },
    {
      "metadata": {
        "colab_type": "code",
        "outputId": "a3bdf43d-6951-442a-db09-5aadab226621",
        "id": "yBQ3_jf4BsAs",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "cell_type": "code",
      "source": [
        "((val_x, trn_x), (val_y, trn_y)) = split_by_idx(val_idxs, x_names, labels)\n",
        "len(trn_x), len(trn_y), len(val_x), len(val_y)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(54400, 54400, 13600, 13600)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 83
        }
      ]
    },
    {
      "metadata": {
        "id": "zc_SpGrjI5o3",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "colors = [(255, 0, 0) , (255, 255, 0),  (128, 255, 0),  (0, 255, 0), (0, 255, 128), (0, 255, 255), \n",
        "          (0, 128, 255), (0, 0, 255), (128, 0, 255), (255, 0, 255)]\n",
        "BASE_SIZE = 256\n",
        "\n",
        "def draw_cv2(raw_strokes, size=256, lw=7, time_color=True):\n",
        "    img = np.zeros((BASE_SIZE, BASE_SIZE, 3), np.uint8)\n",
        "    for t, stroke in enumerate(raw_strokes):\n",
        "        for i in range(len(stroke[0]) - 1):\n",
        "            color = colors[min(t, len(colors)-1)]\n",
        "            _ = cv2.line(img, (stroke[0][i], stroke[1][i]),\n",
        "                         (stroke[0][i + 1], stroke[1][i + 1]), color, lw, lineType=cv2.LINE_AA)\n",
        "    if size != BASE_SIZE:\n",
        "        return cv2.resize(img, (size, size), interpolation=cv2.INTER_LINEAR)\n",
        "    else:\n",
        "        return img\n",
        "\n",
        "def gen_img(path):\n",
        "    with open(path) as f: drawing = eval(f.read())\n",
        "    return draw_cv2(drawing, size=sz)\n",
        "\n",
        "def open_image(fn):\n",
        "    if not os.path.exists(fn) and not str(fn).startswith(\"http\"):\n",
        "        raise OSError('No such file or directory: {}'.format(fn))\n",
        "    elif os.path.isdir(fn) and not str(fn).startswith(\"http\"):\n",
        "        raise OSError('Is a directory: {}'.format(fn))\n",
        "    else:\n",
        "        try:\n",
        "            return gen_img(str(fn)).astype(np.float32) / 255\n",
        "        except Exception as e:\n",
        "            raise OSError('Error handling image at: {}'.format(fn)) from e\n",
        "\n",
        "class DrawingDataset(FilesDataset):\n",
        "    def __init__(self, fnames, y, tfm, path, classes):\n",
        "        assert isinstance(classes, (list, np.ndarray)), 'classes must be label list.'\n",
        "        self.y, self.c = y, len(classes)\n",
        "        self.cls2id = {o:i for i, o in enumerate(classes)}\n",
        "        super().__init__(fnames, tfm, path)\n",
        "        \n",
        "    def get_y(self, i): return self.cls2id[self.y[i]]\n",
        "    def get_x(self, i): return open_image(os.path.join(self.path, self.fnames[i]))\n",
        "    def get_c(self): return self.c"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "5da9YFGpI5uy",
        "colab_type": "code",
        "outputId": "115db6d6-c6bf-44cf-a6d2-f21f40525a3b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "cell_type": "code",
      "source": [
        "arch = resnet34\n",
        "bs = 640\n",
        "sz = 128\n",
        "\n",
        "aug_tfms = [\n",
        "    RandomFlip(),\n",
        "#     RandomRotate(5, p=0.5),\n",
        "    RandomLighting(0.05, 0.05)\n",
        "]\n",
        "\n",
        "tfms = tfms_from_model(arch, sz, aug_tfms=aug_tfms, max_zoom=1.05)\n",
        "ds = ImageData.get_ds(DrawingDataset, (trn_x, trn_y), (val_x, val_y), tfms, test=test_x_names, path=PATH, classes=classes)\n",
        "md = ImageData(PATH, ds, bs, num_workers=8, classes=classes)\n",
        "md.c, md.classes[:3]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(340,\n",
              " array(['The_Eiffel_Tower', 'The_Great_Wall_of_China', 'The_Mona_Lisa'], dtype='<U23'))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 85
        }
      ]
    },
    {
      "metadata": {
        "id": "-OS4gaynI50h",
        "colab_type": "code",
        "outputId": "c33e1017-a740-483e-d4ce-486015ae4153",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "cell_type": "code",
      "source": [
        "%time x, y = next(iter(md.trn_dl))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CPU times: user 25.4 s, sys: 7.21 s, total: 32.6 s\n",
            "Wall time: 19.6 s\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "VEvKZMpbI5x6",
        "colab_type": "code",
        "outputId": "16917840-116f-40e7-84d8-2eb62aa105a5",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 437
        }
      },
      "cell_type": "code",
      "source": [
        "x = md.trn_ds.denorm(x)\n",
        "fig, axes = plt.subplots(3, 3, figsize=(9, 6))\n",
        "for i, ax in enumerate(axes.flat):\n",
        "  ax.imshow(x[i])\n",
        "  ax.set_title(md.classes[y[i]])\n",
        "plt.tight_layout()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhkAAAGkCAYAAACcpje9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f8bda6404e0>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "GmuhEP6wKWnF",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### MobileNet"
      ]
    },
    {
      "metadata": {
        "id": "ZqMhdMmxxEdY",
        "colab_type": "code",
        "outputId": "1c9d1e4c-c935-4789-e02b-43e214156de0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "mn = f'mobilenet_{sz}sz'\n",
        "mn"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'mobilenet_128sz'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "metadata": {
        "id": "gnbfF4BRKWDO",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def std_conv(nin, nf, ks=3, stride=1):\n",
        "  return nn.Sequential(\n",
        "      nn.Conv2d(nin, nf, ks, stride, padding=ks//2, bias=False),\n",
        "      nn.BatchNorm2d(nf),\n",
        "      nn.ReLU(inplace=True)\n",
        "#       nn.LeakyReLU(inplace=True)\n",
        "  )\n",
        "\n",
        "def conv_dw(nin, nf, stride=2):\n",
        "  return nn.Sequential(\n",
        "      nn.Conv2d(nin, nin, 3, stride, 1, groups=nin, bias=False),\n",
        "      nn.BatchNorm2d(nin),\n",
        "      nn.ReLU(inplace=True),\n",
        "#       nn.LeakyReLU(inplace=True),\n",
        "      nn.Conv2d(nin, nf, 1, 1, bias=False),\n",
        "      nn.BatchNorm2d(nf),\n",
        "      nn.ReLU(inplace=True)\n",
        "#       nn.LeakyReLU(inplace=True)\n",
        "  )\n",
        "\n",
        "class MobileNet(nn.Module):\n",
        "  def __init__(self, num_classes, nf=512, ps=[0.25, 0.5]):\n",
        "    super().__init__()\n",
        "    self.layers = nn.Sequential(\n",
        "        std_conv(3, 32, 2),\n",
        "        conv_dw(32, 64, 1),\n",
        "        conv_dw(64, 128),\n",
        "        conv_dw(128, 128, 1),\n",
        "        conv_dw(128, 256),\n",
        "        conv_dw(256, 256, 1),\n",
        "        conv_dw(256, 512),\n",
        "        conv_dw(512, 512, 1),\n",
        "        conv_dw(512, 512, 1),\n",
        "        conv_dw(512, 512, 1),\n",
        "        conv_dw(512, 512, 1),\n",
        "        conv_dw(512, 512, 1),\n",
        "        conv_dw(512, 1024, 2),\n",
        "        conv_dw(1024, 1024, 1),\n",
        "        nn.AdaptiveAvgPool2d(1),\n",
        "        Flatten(),\n",
        "        nn.BatchNorm1d(1024),\n",
        "        nn.Dropout(p=ps[0]),\n",
        "        nn.Linear(1024, nf),\n",
        "        nn.ReLU(),\n",
        "        nn.BatchNorm1d(nf),\n",
        "        nn.Dropout(p=ps[1]),\n",
        "        nn.Linear(nf, num_classes),      \n",
        "        nn.LogSoftmax()\n",
        "    )\n",
        "\n",
        "  def forward(self, x):\n",
        "    return self.layers(x)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "JpWK-R1pQAc6",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def map_top3(preds, targs):\n",
        "  top5 = torch.sort(preds, dim=1, descending=True)[0][:, :3]\n",
        "  return torch.mean(torch.sum(torch.exp(top5), 1))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "CXojUSNxx8L9",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "ps = [0.25, 0.5]\n",
        "nf = 512\n",
        "\n",
        "m = to_gpu(MobileNet(md.c, nf, ps))\n",
        "learn = ConvLearner.from_model_data(m, md)\n",
        "# learn.cirt = nn.CrossEntropyLoss()\n",
        "learn.metrics = [accuracy, map_top3]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "7AMEzlWKn2pr",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# load_model_weights(subdir='20181105')\n",
        "# learn.load('mobilenet_64sz_5k')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "XHmBkolm5F0Y",
        "colab_type": "code",
        "outputId": "e894cadf-534b-4f8a-96fd-0bd5ebe38463",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 402
        }
      },
      "cell_type": "code",
      "source": [
        "learn.lr_find()\n",
        "learn.sched.plot()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "b4cec0e470774810aaf12f042bc58d26",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(IntProgress(value=0, description='Epoch', max=1, style=ProgressStyle(description_width='initial…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            " 87%|████████▋ | 743/850 [08:42<01:17,  1.37it/s, loss=24.3]"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAe0AAAFfCAYAAABjtpSFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nb8AlAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f8b650bea90>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "hYzokIxrxRTc",
        "colab_type": "code",
        "outputId": "a31a71fd-0ccf-4eda-dca3-0173d8717256",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 218
        }
      },
      "cell_type": "code",
      "source": [
        "lr = 8e-2\n",
        "learn.fit(lr, 1, cycle_len=10, use_clr=(20, 10))\n",
        "learn.save(mn)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "df305c65a86f403a99a4b8dfb388b3c7",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(IntProgress(value=0, description='Epoch', max=10, style=ProgressStyle(description_width='initia…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "epoch      trn_loss   val_loss   accuracy   map_top3   \n",
            "    0      4.221134   3.929499   0.174265   0.373     \n",
            "    1      3.122172   2.83353    0.352059   0.528392  \n",
            "    2      2.647427   2.299856   0.460735   0.640049  \n",
            "    3      2.421247   2.322631   0.454706   0.634115  \n",
            "    4      2.163204   1.90709    0.545147   0.721836  \n",
            "    5      1.975682   1.765632   0.575662   0.747461  \n",
            "    6      1.821589   1.665574   0.603162   0.768702  \n",
            "    7      1.698267   1.60571    0.617794   0.786565  \n",
            "    8      1.58901    1.53894    0.634118   0.807489  \n",
            "    9      1.459088   1.491365   0.644044   0.82133   \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "hDFt1Ut_T7dV",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Resnet34"
      ]
    },
    {
      "metadata": {
        "colab_type": "code",
        "id": "nWEqA455D66-",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "models = ConvnetBuilder(arch, len(classes), False, False)\n",
        "learn = ConvLearner(md, models)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "I3n9cfKRDdnX",
        "colab_type": "code",
        "outputId": "19ae469b-4a61-4705-acf9-e3ea35f6d1b8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 438
        }
      },
      "cell_type": "code",
      "source": [
        "learn.lr_find()\n",
        "learn.sched.plot(0)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "4f8994f6abc24ae48078a885d6ab70e8",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(IntProgress(value=0, description='Epoch', max=1, style=ProgressStyle(description_width='initial…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "epoch      trn_loss   val_loss   accuracy   \n",
            "    0      5.329136   5.645055   0.206324  \n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAe8AAAFcCAYAAADh1zYWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIiI3w/ImIiJyM/8fOrBBVeY1spgAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f8bd64d2fd0>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "6gEI0hnwT-Vb",
        "colab_type": "code",
        "outputId": "b7cf3845-d5f9-48df-dcbd-da2b0eecf6db",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 145
        }
      },
      "cell_type": "code",
      "source": [
        "lr = 4e-1\n",
        "learn.fit(lr, 2, cycle_len=1, cycle_mult=2)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "8854c90683ad4ff3819994e2f1b27224",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(IntProgress(value=0, description='Epoch', max=3, style=ProgressStyle(description_width='initial…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "epoch      trn_loss   val_loss   accuracy   \n",
            "    0      3.799521   3.013281   0.344265  \n",
            "    1      3.202584   2.542      0.418088  \n",
            "    2      2.804773   2.404532   0.445515  \n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[2.4045316331526814, 0.4455147070043227]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 92
        }
      ]
    },
    {
      "metadata": {
        "id": "qxsadDCBV1LX",
        "colab_type": "code",
        "outputId": "685c1016-f333-4070-bc26-71dd40e8a606",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 438
        }
      },
      "cell_type": "code",
      "source": [
        "learn.unfreeze()\n",
        "learn.lr_find(lr / 1000)\n",
        "learn.sched.plot()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "22bfcedf882a4a77bb09166ebc2f3a9d",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(IntProgress(value=0, description='Epoch', max=1, style=ProgressStyle(description_width='initial…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "epoch      trn_loss   val_loss   accuracy   \n",
            "    0      3.380707   10646.408054 0.003162  \n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAe8AAAFcCAYAAADh1zYWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\njHgDQGe+AN0NRJh5Ev04AAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f8bd7c8c550>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "AO_xu2Kbce6L",
        "colab_type": "code",
        "outputId": "fe43a7d9-4b92-4039-c484-71cb5eec2142",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 302
        }
      },
      "cell_type": "code",
      "source": [
        "lr = 8e-4\n",
        "lrs = np.array([lr / 9, lr / 3, lr])\n",
        "learn.fit(lrs, 3, cycle_len=2, cycle_mult=2)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "582ea40bfe004ddf8cb1204f1c02bc82",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(IntProgress(value=0, description='Epoch', max=14, style=ProgressStyle(description_width='initia…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "epoch      trn_loss   val_loss   accuracy   \n",
            "    0      2.496831   2.085892   0.510588  \n",
            "    1      2.305883   2.030547   0.520294  \n",
            "    2      2.123826   1.874482   0.554559  \n",
            "    3      2.017922   1.802687   0.5675    \n",
            "    4      1.924573   1.772526   0.577059  \n",
            "    5      1.901727   1.768791   0.578382  \n",
            "    6      1.890097   1.712933   0.589559  \n",
            "    7      1.868803   1.668383   0.597059  \n",
            "    8      1.666557   1.637589   0.605     \n",
            "    9      1.597122   1.614383   0.608088  \n",
            "    10     1.609161   1.599738   0.610882  \n",
            "    11     1.522107   1.590971   0.613529  \n",
            "    12     1.544126   1.593027   0.612059  \n",
            "    13     1.543969   1.589572   0.613603  \n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[1.5895717581580666, 0.6136029411764706]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 63
        }
      ]
    },
    {
      "metadata": {
        "id": "VaSdrsotKHTI",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "pa = f'resnet34_{sz}sz'\n",
        "learn.save(pa)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "ZL6CU938LF4V",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "learn.metrics = [accuracy, map_top3]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Y5OtrrxZOcez",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "save_model_weights()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Rh2uGjT9f8qy",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "learn.fit(lrs, 1, cycle_len=10, use_clr=(20, 10))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "1lDKL1KadkvU",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Submission"
      ]
    },
    {
      "metadata": {
        "id": "80HUPmVcQ3Ds",
        "colab_type": "code",
        "outputId": "ae3b130d-f5d2-4112-a171-8b16c757587f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        }
      },
      "cell_type": "code",
      "source": [
        "preds = learn.TTA(is_test=True)[0]"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            ""
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "wXpnsQvfg54A",
        "colab_type": "code",
        "outputId": "d0e6f0e0-5862-4c40-9c0c-6f62e7b671ad",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "preds.shape"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(5, 112199, 340)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 78
        }
      ]
    },
    {
      "metadata": {
        "id": "_5EXNaAPh3U6",
        "colab_type": "code",
        "outputId": "5cf93d6f-f154-47de-e0b8-2b645fa02397",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "pr = np.mean(np.exp(preds), 0)\n",
        "pr.shape"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(112199, 340)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 79
        }
      ]
    },
    {
      "metadata": {
        "id": "ejCyEXulidR7",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "top_3 = np.argsort(pr)[:, ::-1][:, :3]\n",
        "cat = np.array([' '.join(str(md.classes[i]) for i in o) for o in top_3])\n",
        "image = md.test_ds.fnames"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Vck1nG75jO8W",
        "colab_type": "code",
        "outputId": "7bc7ad6b-0d46-4922-8658-e809413f632d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 203
        }
      },
      "cell_type": "code",
      "source": [
        "submission = pd.DataFrame({'key_id': image, 'word': cat})\n",
        "submission.head()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>key_id</th>\n",
              "      <th>word</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>test_128/9028014602137518.png</td>\n",
              "      <td>banana elbow matches</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>test_128/9325698402591604.png</td>\n",
              "      <td>paint_can hot_tub cake</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>test_128/9791244567109262.png</td>\n",
              "      <td>string_bean feather dolphin</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>test_128/9396277400422292.png</td>\n",
              "      <td>bulldozer tractor car</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>test_128/9037773561277176.png</td>\n",
              "      <td>flamingo giraffe saxophone</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                          key_id                         word\n",
              "0  test_128/9028014602137518.png         banana elbow matches\n",
              "1  test_128/9325698402591604.png       paint_can hot_tub cake\n",
              "2  test_128/9791244567109262.png  string_bean feather dolphin\n",
              "3  test_128/9396277400422292.png        bulldozer tractor car\n",
              "4  test_128/9037773561277176.png   flamingo giraffe saxophone"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 81
        }
      ]
    },
    {
      "metadata": {
        "id": "DBuk-R0pkxVE",
        "colab_type": "code",
        "outputId": "3ca19c2e-cf67-49da-9624-f159df3f60f0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 203
        }
      },
      "cell_type": "code",
      "source": [
        "submission.key_id = submission.key_id.str.replace('^test_128/', '')\n",
        "# submission.key_id = submission.key_id.str.replace('^test_64/', '')\n",
        "# submission.key_id = submission.key_id.str.replace('^test_32/', '')\n",
        "submission.key_id = submission.key_id.str.replace('.png$', '')\n",
        "submission.head()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>key_id</th>\n",
              "      <th>word</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>9028014602137518</td>\n",
              "      <td>banana elbow matches</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>9325698402591604</td>\n",
              "      <td>paint_can hot_tub cake</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>9791244567109262</td>\n",
              "      <td>string_bean feather dolphin</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>9396277400422292</td>\n",
              "      <td>bulldozer tractor car</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>9037773561277176</td>\n",
              "      <td>flamingo giraffe saxophone</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "             key_id                         word\n",
              "0  9028014602137518         banana elbow matches\n",
              "1  9325698402591604       paint_can hot_tub cake\n",
              "2  9791244567109262  string_bean feather dolphin\n",
              "3  9396277400422292        bulldozer tractor car\n",
              "4  9037773561277176   flamingo giraffe saxophone"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 83
        }
      ]
    },
    {
      "metadata": {
        "id": "8KDhjhQRmsZX",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "submission.to_csv('submission.csv.gz', index=False, compression='gzip')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "izVvIj7jm6HL",
        "colab_type": "code",
        "outputId": "9ca8bcd1-096a-46a0-f579-83c66ff50c11",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "cell_type": "code",
      "source": [
        "!kaggle competitions submit -c quickdraw-doodle-recognition -f submission.csv.gz -m \"\""
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100% 1.57M/1.57M [00:04<00:00, 348kB/s]\n",
            "Successfully submitted to Quick, Draw! Doodle Recognition Challenge"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "G97vJuTjkK05",
        "colab_type": "code",
        "outputId": "103b1aff-5a27-4f87-b94b-a8cf8799d776",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 140
        }
      },
      "cell_type": "code",
      "source": [
        "!kaggle competitions submissions -c quickdraw-doodle-recognition"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "fileName           date                 description  status    publicScore  privateScore  \n",
            "-----------------  -------------------  -----------  --------  -----------  ------------  \n",
            "submission.csv.gz  2018-11-10 07:05:27               complete  0.788        None          \n",
            "submission.csv.gz  2018-11-06 08:23:56               complete  0.833        None          \n",
            "submission.csv.gz  2018-11-06 08:22:05               error     None         None          \n",
            "submission.csv.gz  2018-11-01 09:31:01               complete  0.740        None          \n",
            "submission.csv     2018-11-01 09:26:52               error     None         None          \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "0ISOsP1okK8G",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}