ITL_recursive_CNNs.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "colab": {
      "name": "ITL recursive CNNs",
      "provenance": [],
      "collapsed_sections": [
        "uWs7SXc2MI0O",
        "RpuBVnpD_j9I"
      ]
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-gDlTQlA7ETk"
      },
      "source": [
        "# ITL Recursive CNNs - inference + evaluation\n",
        "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/kakainet/TexNet/blob/master/ITL_recursive_CNNs.ipynb)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HobAbLXiMI0H",
        "outputId": "1f65cbd3-541b-4016-f32d-6c8f69bc2071",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "#Run once per session\n",
        "%load_ext autoreload\n",
        "%autoreload 2\n",
        "!pip install -q fastcore==1.2.5\n",
        "!pip install -q fastai==2.0.16"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[K     |████████████████████████████████| 51kB 3.1MB/s \n",
            "\u001b[K     |████████████████████████████████| 194kB 4.8MB/s \n",
            "\u001b[?25h"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uWs7SXc2MI0O"
      },
      "source": [
        "# Dataset loading + initialization"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-hf1lY_FMI0Q"
      },
      "source": [
        "from fastai.vision.all import *"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I1GGI9RBMI0U",
        "outputId": "1ec988d7-8ef9-4a02-c5be-b007cf57ce57",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "VERSION=13\n",
        "ITL_DS_NAME = f'itl_dataset_v{VERSION}.zip'\n",
        "ITL_DS_URL =  f'http://students.mimuw.edu.pl/~mt406390/machine_learning/' + ITL_DS_NAME\n",
        "!wget $ITL_DS_URL\n",
        "!unzip -qq $ITL_DS_NAME -d data/\n",
        "path = Path('data')"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-11-02 17:20:47--  http://students.mimuw.edu.pl/~mt406390/machine_learning/itl_dataset_v13.zip\n",
            "Resolving students.mimuw.edu.pl (students.mimuw.edu.pl)... 193.0.96.129, 2001:6a0:5001:1::3\n",
            "Connecting to students.mimuw.edu.pl (students.mimuw.edu.pl)|193.0.96.129|:80... connected.\n",
            "HTTP request sent, awaiting response... 301 Moved Permanently\n",
            "Location: https://students.mimuw.edu.pl/~mt406390/machine_learning/itl_dataset_v13.zip [following]\n",
            "--2020-11-02 17:20:47--  https://students.mimuw.edu.pl/~mt406390/machine_learning/itl_dataset_v13.zip\n",
            "Connecting to students.mimuw.edu.pl (students.mimuw.edu.pl)|193.0.96.129|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 13406847 (13M) [application/zip]\n",
            "Saving to: ‘itl_dataset_v13.zip’\n",
            "\n",
            "itl_dataset_v13.zip 100%[===================>]  12.79M  6.35MB/s    in 2.0s    \n",
            "\n",
            "2020-11-02 17:20:49 (6.35 MB/s) - ‘itl_dataset_v13.zip’ saved [13406847/13406847]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KKWezwyLMI0Y",
        "outputId": "55f20607-8292-4933-f013-c64f681722c4",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "path.ls()"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(#9) [Path('data/itl_labels.txt'),Path('data/operators.txt'),Path('data/output_black'),Path('data/depths.txt'),Path('data/output_bbox'),Path('data/output_proper'),Path('data/output_color'),Path('data/annotations.json'),Path('data/job.json')]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TXEu4SWkMI0d",
        "outputId": "14e4e44b-064f-4057-9949-a30e79ffa03f",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "with open(path / 'annotations.json') as f:\n",
        "    annots = json.load(f)\n",
        "print(len(annots))"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "800\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_pPlgsF6MI0g",
        "outputId": "257db911-1173-464a-8a9f-fe19caa49750",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 0
        }
      },
      "source": [
        "images = list(x['name'] for x in annots if len(x['exprs']) > 0)\n",
        "print(len(images))\n",
        "images[0]"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "800\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'eq0.png'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eawdh5vfvOXt"
      },
      "source": [
        "ops_cfg = {\n",
        "  \"leaf\": {\n",
        "    \"latex\": \"{0}\",\n",
        "    \"operands\": 0,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"add\": {\n",
        "    \"latex\": \"{0} + {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"times\": {\n",
        "    \"latex\": \"{0} \\\\times {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"star\": {\n",
        "    \"latex\": \"{0} \\\\star {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"div\": {\n",
        "    \"latex\": \"{0} \\\\div {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"otimes\": {\n",
        "    \"latex\": \"{0} \\\\otimes {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"sub\": {\n",
        "    \"latex\": \"{0} - {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"mul\": {\n",
        "    \"latex\": \"{0} \\\\cdot {1}\",\n",
        "    \"operands\": 2,\n",
        "    \"inline\": true\n",
        "  },\n",
        "  \"frac\": {\n",
        "    \"latex\": \"\\\\frac {{ {0} }} {{ {1} }}\",\n",
        "    \"operands\": 2\n",
        "  },\n",
        "  \"sqrt\": {\n",
        "    \"latex\": \"\\\\sqrt {{ {0} }}\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"int\": {\n",
        "    \"latex\": \"\\\\int \\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"sum\": {\n",
        "    \"latex\": \"\\\\sum \\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"pi\": {\n",
        "    \"latex\": \"\\\\prod \\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"par\": {\n",
        "    \"latex\": \"\\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"fpar\": {\n",
        "    \"latex\": \"f \\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"ffpar\": {\n",
        "    \"latex\": \"F \\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  },\n",
        "  \"gpar\": {\n",
        "    \"latex\": \"G \\\\left( {0} \\\\right)\",\n",
        "    \"operands\": 1\n",
        "  }\n",
        "}"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8XnwBchlMI0m"
      },
      "source": [
        "Next, we want to be able to quickly look up a corresponding image to it's label. We'll use a dictionary"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hOOYWbhxMI0n"
      },
      "source": [
        "img2bbox = dict(zip(images, annots))"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3ZjUaaI8MI0t"
      },
      "source": [
        "Great! Now let's build our `DataBlock`. We'll have two outputs, the bounding box itself and a label, with one input. For our `y`'s, we'll define some `getters`. The first of which will be `noop`, as we don't care about our initial image, and then we'll grab our label and bbox coordinates (we'll also define our `get_items`)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dqMkzWBkMI0t"
      },
      "source": [
        "def bbox_to_fastai(box):\n",
        "    x, y, dx, dy = box\n",
        "    return [x, y, x + dx, y + dy]\n",
        "\n",
        "getters = [\n",
        "    lambda o: path/'output_proper'/o, lambda o: list(map(bbox_to_fastai, img2bbox[o]['exprs'])),\n",
        "    lambda o: len(img2bbox[o]['exprs']) * [img2bbox[o]['op']]\n",
        "]"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jlvhInImMI0w"
      },
      "source": [
        "For our transforms, we'll use some of the ones we defined earlier"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WYOBW1HMMI0x"
      },
      "source": [
        "item_tfms = []\n",
        "batch_tfms = [Zoom(0.6, 1.5), Warp(0.25), Saturation(max_lighting=0.5),\n",
        "              Brightness(), Rotate()]"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2YrjfDxuMI0z"
      },
      "source": [
        "Why do we need a custom `get_images`? Because we want our **images** that came back to us, not the entire folder"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "diaZXpW4MI00"
      },
      "source": [
        "def get_train_imgs(noop):  return images"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0oKEQM7HMI03"
      },
      "source": [
        "We'll now make our `DataBlock`. We want to adjust `n_inp` as we expect two outputs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qD57DuQGMI03"
      },
      "source": [
        "pascal = DataBlock(blocks=(ImageBlock(cls=PILImageBW), BBoxBlock, BBoxLblBlock),\n",
        "                 splitter=RandomSplitter(0.3),\n",
        "                 get_items=get_train_imgs, \n",
        "                 getters=getters,\n",
        "                 item_tfms=item_tfms,\n",
        "                 batch_tfms=batch_tfms,\n",
        "                 n_inp=1)"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OadwEe45MI06"
      },
      "source": [
        "BS = 16\n",
        "dls = pascal.dataloaders('', bs=BS)\n",
        "dls.c = len(ops_cfg)"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jqEEJ0AkMI0_",
        "outputId": "0bece446-e80d-48aa-fab1-cac6109c2e20",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 0
        }
      },
      "source": [
        "dls.show_batch(max_n=25)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x864 with 16 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZB-fSbhqQsnh"
      },
      "source": [
        "# ITL model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HpRX2rdUVtNK"
      },
      "source": [
        "POINTS_PER_BBOX = 4\n",
        "N_BBOXES = 2\n",
        "BBOXES = N_BBOXES * POINTS_PER_BBOX\n",
        "CATS = dls.c\n",
        "OUT = CATS + BBOXES"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dTeL1CgLBiGx"
      },
      "source": [
        "head = nn.Sequential(\n",
        "    nn.Conv2d(512, 128, 3, 2),\n",
        "    nn.BatchNorm2d(128),\n",
        "    nn.Conv2d(128, OUT, 3, 1),\n",
        "    nn.Flatten()\n",
        ")\n",
        "\n",
        "cfg = cnn_config(n_in=1, custom_head=head)"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Lkvpmr5JJ9vP"
      },
      "source": [
        "from torchvision.ops import box_iou\n",
        "\n",
        "def encode_boxes(boxes_points):\n",
        "    boxes_points = (boxes_points + 1) / 2\n",
        "    x1, y1, x2, y2 = boxes_points[..., 0], boxes_points[..., 1], boxes_points[..., 2], boxes_points[..., 3]\n",
        "    return torch.stack((x1, y1, torch.sqrt(x2 - x1), torch.sqrt(y2 - y1)), dim=-1)\n",
        "    \n",
        "def decode_boxes(boxes_lens):\n",
        "    x, y, dx, dy = boxes_lens[..., 0], boxes_lens[..., 1], torch.square(boxes_lens[..., 2]), torch.square(boxes_lens[..., 3])\n",
        "    boxes_points = torch.stack((x, y, x + dx, y + dy), dim=-1)\n",
        "    return 2 * boxes_points - 1\n",
        "\n",
        "class MseAndCrossEntropy(Module):\n",
        "    def forward(self, pred, target_bboxes, target_cat):\n",
        "        bboxes_preds = torch.sigmoid(pred[..., CATS:]).view((-1, N_BBOXES, POINTS_PER_BBOX))\n",
        "        mask = (target_cat > 0)\n",
        "\n",
        "        box_loss = F.mse_loss(bboxes_preds[mask], target_bboxes[mask])\n",
        "        ce_loss = F.cross_entropy(pred[..., :CATS], target_cat[..., 0] - 1)\n",
        "        return ce_loss + 50 * torch.sqrt(box_loss)\n",
        "\n",
        "def ops_accuracy(pred, target_bboxes, target_cat):\n",
        "    return accuracy(pred[..., :CATS], target_cat[..., 0] - 1)\n",
        "    \n",
        "def boxes_iou(pred, target_bboxes, target_cat):\n",
        "    bboxes_preds = torch.sigmoid(pred[..., CATS:]).view((-1, N_BBOXES, POINTS_PER_BBOX))\n",
        "    bboxes_decoded = decode_boxes(bboxes_preds)\n",
        "\n",
        "    mask = (target_cat > 0)\n",
        "    x, y = bboxes_decoded[mask], target_bboxes[mask]\n",
        "    ious = torch.diagonal(box_iou(x, y))\n",
        "    return ious.mean()\n",
        "\n",
        "\n",
        "@log_args\n",
        "class ITLLoss(BaseLoss):\n",
        "    def __init__(self, *args, axis=-1, **kwargs):\n",
        "        super().__init__(MseAndCrossEntropy, *args, axis=axis, **kwargs)\n",
        "\n",
        "    def __call__(self, inp, targ, labels, **kwargs):\n",
        "        targ_encoded = encode_boxes(targ)\n",
        "\n",
        "        return self.func.__call__(inp, targ_encoded, labels)\n",
        "\n",
        "    def decodes(self, x):\n",
        "        categories, bboxes = x[...,:CATS], x[...,CATS:]\n",
        "        bboxes = torch.sigmoid(bboxes)\n",
        "\n",
        "        # Scaling back\n",
        "        bboxes = decode_boxes(bboxes.view(-1, 2, 4))\n",
        "        \n",
        "\n",
        "        cats = (1 + categories.argmax(dim=-1)).unsqueeze(dim=-1)\n",
        "\n",
        "        cats_stacked = torch.cat((cats, torch.zeros(cats.shape, dtype=torch.int64)), axis=-1)\n",
        "        n = bboxes.shape[0]\n",
        "        for i in range(n):\n",
        "            current_cat = cats[i, 0]\n",
        "\n",
        "            cat_lbl = dls.vocab[current_cat]\n",
        "\n",
        "            if ops_cfg[cat_lbl]['operands'] == 2:\n",
        "                cats_stacked[i, 1] = cats_stacked[i, 0]\n",
        "            else:\n",
        "                bboxes[i, 1] = 0\n",
        "\n",
        "        return bboxes, TensorMultiCategory(cats_stacked)\n"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uokx32ABlOkk"
      },
      "source": [
        "learn = cnn_learner(dls, xse_resnext18, loss_func=ITLLoss(), pretrained=False, config=cfg,\n",
        "                    metrics=[ops_accuracy, boxes_iou])"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x9fVIUeWZ-mA",
        "outputId": "85b60c7b-3a39-42c1-fcf1-ca52c70e6be1",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!wget https://students.mimuw.edu.pl/~mt406390/machine_learning/infer30_16.pth\n",
        "!mkdir models\n",
        "!mv infer30_16.pth models/"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-11-02 17:20:54--  https://students.mimuw.edu.pl/~mt406390/machine_learning/infer30_16.pth\n",
            "Resolving students.mimuw.edu.pl (students.mimuw.edu.pl)... 193.0.96.129, 2001:6a0:5001:1::3\n",
            "Connecting to students.mimuw.edu.pl (students.mimuw.edu.pl)|193.0.96.129|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 55125825 (53M)\n",
            "Saving to: ‘infer30_16.pth’\n",
            "\n",
            "infer30_16.pth      100%[===================>]  52.57M  4.90MB/s    in 11s     \n",
            "\n",
            "2020-11-02 17:21:06 (4.62 MB/s) - ‘infer30_16.pth’ saved [55125825/55125825]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hpfEXDyeXCq-",
        "outputId": "cd4e0a58-687d-4763-be75-5ac567e764b4",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "learn.load('infer30_16')"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<fastai.learner.Learner at 0x7f9a31ef6908>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tdBRssmlyCHa"
      },
      "source": [
        "# Example CNN predictions"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kBAFPaq6CUNM",
        "outputId": "8aac2398-4011-4843-d403-97054e03d177",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 683
        }
      },
      "source": [
        "learn.show_results(max_n=20)"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x864 with 16 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "C1d4INwBx5aB"
      },
      "source": [
        "# ITL model inference"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YKU6ClxA28v1"
      },
      "source": [
        "class Predictor():\n",
        "    def __init__(self, learn: Learner):\n",
        "        self.learn = learn\n",
        "\n",
        "        self.labeler, self.point_scaler, self.to_tensor = learn.dls.valid.after_item\n",
        "        self.scale_pixels = learn.dls.valid.after_batch[0]\n",
        "\n",
        "        self.tfms = [self.labeler, self.point_scaler, self.to_tensor, self.scale_pixels]\n",
        "\n",
        "        self.loss = ITLLoss()\n",
        "    \n",
        "    def _transform_img(self, img: PILImageBW) -> TensorImageBW:\n",
        "        return self.scale_pixels(self.to_tensor(self.labeler(self.point_scaler(img))))[None]\n",
        "\n",
        "    def _decode_after_fastai(self, boxes, cats):\n",
        "        boxes = ((1 + boxes) / 2) * img.shape[1]\n",
        "        \n",
        "        x1, y1, x2, y2 = (boxes[0, :, i] for i in range(POINTS_PER_BBOX))\n",
        "        boxes_coco = torch.stack((x1, y1, x2 - x1, y2 - y1), dim=-1)\n",
        "\n",
        "        cat = self.learn.dls.vocab[cats[0, 0]]\n",
        "        return boxes_coco, cat\n",
        "\n",
        "    def _decode_preds(self, x):\n",
        "        boxes, cats = self.loss.decodes(x)\n",
        "        return self._decode_after_fastai(boxes, cats)\n",
        "\n",
        "\n",
        "    def predict(self, img: PILImageBW):\n",
        "        with torch.no_grad():\n",
        "            preds = self.learn.model(to_device(self._transform_img(img), default_device()))\n",
        "\n",
        "        return self._decode_preds(preds.cpu())"
      ],
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "a__KYhXLfDEX"
      },
      "source": [
        "import abc\n",
        "from dataclasses import dataclass\n",
        "from typing import List\n",
        "\n",
        "@dataclass\n",
        "class Box:\n",
        "    x: int\n",
        "    y: int\n",
        "    dx: int\n",
        "    dy: int\n",
        "\n",
        "@dataclass\n",
        "class Operator:\n",
        "    op_id: str\n",
        "\n",
        "@dataclass\n",
        "class Prediction:\n",
        "    op: Operator\n",
        "    bboxes: List[Box]\n",
        "\n",
        "class Model(abc.ABC):\n",
        "    @abc.abstractmethod\n",
        "    def predict(self, input) -> Prediction:\n",
        "        raise NotImplementedError\n",
        "\n",
        "class ModelExample(Model):\n",
        "    def predict(self, input) -> Prediction:\n",
        "        return 'add', [Box(25, 25, 17, 17), Box(34, 34, 17, 17)]\n",
        "\n",
        "class FastaiModelAdapter(Model):\n",
        "    def __init__(self, predictor: Predictor):\n",
        "        self._pred = predictor\n",
        "\n",
        "    def _box_from_tensor(self, box_tensor):\n",
        "        return Box(*map(lambda x: int(x.round().item()), box_tensor))\n",
        "\n",
        "    def _make_pred(self, boxes_tensor, cat):\n",
        "        boxes = [self._box_from_tensor(boxes_tensor[i])\n",
        "            for i in range(boxes_tensor.shape[0])]\n",
        "\n",
        "        return Prediction(\n",
        "            op=Operator(cat),\n",
        "            bboxes=boxes\n",
        "        )\n",
        "\n",
        "    def predict(self, input: PILImageBW) -> Prediction:\n",
        "        return self._make_pred(*self._pred.predict(input))\n"
      ],
      "execution_count": 29,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "A3F75fSl1F-_"
      },
      "source": [
        "pred = Predictor(learn)\n",
        "fastai_model = FastaiModelAdapter(pred)"
      ],
      "execution_count": 30,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hrZ3I2660XNm"
      },
      "source": [
        "import cv2\n",
        "from google.colab.patches import cv2_imshow\n",
        "\n",
        "def resize_and_center_pred(full_img: PILImageBW, pred_bbox: Box) -> PILImageBW:\n",
        "    def crop_img(image, bbox):\n",
        "        proc = 0.03\n",
        "        x_margin, y_margin = int(proc * bbox.dx), int(proc * bbox.dy) \n",
        "        return image[bbox.y:bbox.y + bbox.dy + y_margin, bbox.x:bbox.x + bbox.dx + x_margin]\n",
        "\n",
        "    img_arr = np.array(full_img)\n",
        "    img_cropped = crop_img(img_arr, pred_bbox)\n",
        "\n",
        "    DIM, WHITE = 224, 255\n",
        "    res = np.full((DIM, DIM), WHITE, dtype=np.uint8)\n",
        "\n",
        "    y_shift = (DIM - img_cropped.shape[0]) // 2\n",
        "    x_shift = (DIM - img_cropped.shape[1]) // 2\n",
        "    \n",
        "    res[y_shift:y_shift + img_cropped.shape[0], x_shift:x_shift + img_cropped.shape[1]] = img_cropped\n",
        "\n",
        "    return PILImageBW.create(res)\n"
      ],
      "execution_count": 31,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RpuBVnpD_j9I"
      },
      "source": [
        "# seq2seq model (for leaves)\n",
        "\n",
        "PyTorch implementation of https://arxiv.org/abs/1609.04938v2\n",
        "forked from [im2latex](https://github.com/ArminKaramzade/im2latex)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rgIfh_Y03u0U"
      },
      "source": [
        "import os\n",
        "import pickle\n",
        "from collections import defaultdict\n",
        "import torch\n",
        "import glob\n",
        "import sys\n",
        "import cv2\n",
        "import numpy as np\n",
        "import datetime, pytz\n",
        "from torchvision import transforms\n",
        "from PIL import Image\n",
        "\n",
        "class utils:\n",
        "    class Vocab:\n",
        "        def __init__(self, config):\n",
        "            '''\n",
        "                config: pathes, unk_threshold=2\n",
        "            '''\n",
        "            self.pathes = config['pathes']\n",
        "            self.unk_threshold = config.get('unk_threshold', 2)\n",
        "            self.token2idx = {}\n",
        "            self.idx2token = {}\n",
        "            self.load()\n",
        "\n",
        "        def build(self):\n",
        "            self.start_token = 0\n",
        "            self.end_token = 1\n",
        "            self.pad_token = 2\n",
        "            self.unk_token = 3\n",
        "            self.frequency = defaultdict(int)\n",
        "            self.total = 0\n",
        "            for path in self.pathes:\n",
        "                formulas = open(path, 'r')\n",
        "                lines = formulas.readlines()\n",
        "                for line in lines:\n",
        "                    tokens = line.rstrip('\\n').strip(' ').split()\n",
        "                    for token in tokens:\n",
        "                        self.frequency[token] += 1\n",
        "                        self.total += 1\n",
        "            self.token2idx = {'<f>' : 0, '</f>' : 1, '<pad>' : 2, '<unk>' : 3}\n",
        "            self.idx2token = {0 : '<f>', 1 : '</f>', 2 : '<pad>', 3 : '<unk>'}\n",
        "            idx = 4\n",
        "            for path in self.pathes:\n",
        "                formulas = open(path, 'r')\n",
        "                lines = formulas.readlines()\n",
        "                for line in lines:\n",
        "                    tokens = line.rstrip('\\n').strip(' ').split()\n",
        "                    for token in tokens:\n",
        "                        if self.is_eligible(token) and token not in self.token2idx:\n",
        "                            self.token2idx[token] = idx\n",
        "                            self.idx2token[idx] = token\n",
        "                            idx += 1\n",
        "            # save vocab\n",
        "            if not os.path.isdir('vocab'):\n",
        "                os.mkdir('vocab')\n",
        "            f = open(os.path.join('vocab', 'vocab.pkl'), 'wb')\n",
        "            pickle.dump(self, f)\n",
        "            f.close()\n",
        "\n",
        "        def is_eligible(self, token):\n",
        "            if self.frequency[token] >= self.unk_threshold:\n",
        "                return True\n",
        "            return False\n",
        "\n",
        "        def load(self):\n",
        "            try:\n",
        "                with open(os.path.join('vocab', 'vocab.pkl'), 'rb') as f:\n",
        "                    vocab = pickle.load(f)\n",
        "                    self.token2idx = vocab.token2idx\n",
        "                    self.idx2token = vocab.idx2token\n",
        "                    self.start_token = vocab.start_token\n",
        "                    self.unk_token = vocab.unk_token\n",
        "                    self.pad_token = vocab.pad_token\n",
        "                    self.end_token = vocab.end_token\n",
        "                    self.frequency = vocab.frequency\n",
        "                    self.total = vocab.total\n",
        "            except:\n",
        "                self.build()\n",
        "\n",
        "        def formulas2tensor(self, formulas, max_len):\n",
        "            sz = max_len + 2\n",
        "            tensor = np.zeros((len(formulas), sz))\n",
        "            for i in range(tensor.shape[0]):\n",
        "                tensor[i, 0] = self.start_token\n",
        "                for j in range(len(formulas[i])):\n",
        "                    tensor[i, j+1] = self.token2idx.get(formulas[i][j], self.unk_token)\n",
        "                for j in range(len(formulas[i])+1, sz-1):\n",
        "                    tensor[i, j] = self.pad_token\n",
        "                tensor[i, sz-1] = self.end_token\n",
        "            return tensor\n",
        "\n",
        "        def tensor2formula(self, tensor, pretty=False, tags=True):\n",
        "            if not pretty:\n",
        "                if tags:\n",
        "                    return ' '.join(self.idx2token[tensor[i]] for i in range(tensor.shape[0]))\n",
        "                else:\n",
        "                    return ' '.join(self.idx2token[tensor[i]] for i in range(tensor.shape[0])\n",
        "                                    if self.idx2token[tensor[i]] not in ['<f>', '</f>', '<pad>'])\n",
        "            else:\n",
        "                s = ' '.join(self.idx2token[tensor[i]] for i in range(tensor.shape[0]))\n",
        "                end = s.find('</f>')\n",
        "                if end != -1 : end = end - 1\n",
        "                s = s[4:end]\n",
        "                s = s.replace('<pad>', '')\n",
        "                s = s.replace('<unk>', '')\n",
        "                return s\n",
        "\n",
        "    class data_loader:\n",
        "        def __init__(self, vocab, config, train=False):\n",
        "            '''\n",
        "                config: vocab, batch_size, images_path, formulas_path=None, sort_by_formulas_len=False, shuffle=False):\n",
        "            '''\n",
        "            self.vocab = vocab\n",
        "            self.batch_size = config['batch_size']\n",
        "            self.images_path = config['images_path']\n",
        "            self.formulas_path = config.get('formulas_path', None)\n",
        "            self.sort_by_formulas_len = config.get('sort_by_formulas_len', False)\n",
        "            self.shuffle = config.get('shuffle', False)\n",
        "            self.cursor = 0\n",
        "            self.images = glob.glob(self.images_path + '/*.png')\n",
        "            self.images.sort()\n",
        "            self.images.sort(key=len)\n",
        "            self.has_label = (self.formulas_path is not None)\n",
        "            if self.has_label:\n",
        "                with open(self.formulas_path, 'r') as f:\n",
        "                    self.formulas = [line.rstrip('\\n').strip(' ').split()\n",
        "                                    for line in f.readlines()]\n",
        "                assert(len(self.images) == len(self.formulas))\n",
        "                assert(not (self.shuffle and self.sort_by_formulas_len))\n",
        "                self.max_len = 0\n",
        "                for formula in self.formulas: self.max_len = max(self.max_len, len(formula))\n",
        "                if self.sort_by_formulas_len:\n",
        "                    zipped = sorted(enumerate(self.formulas), key=lambda x: len(x[1]), reverse=True)\n",
        "                    idx, self.formulas = zip(*zipped)\n",
        "                    self.images = [self.images[i] for i in idx]\n",
        "                if self.shuffle:\n",
        "                    idx = np.random.permutation(len(self.formulas))\n",
        "                    self.formulas = [self.formulas[i] for i in idx]\n",
        "                    self.images = [self.images[i] for i in idx]\n",
        "            if train:\n",
        "                self.data_transforms = transforms.Compose([\n",
        "                    transforms.Grayscale(num_output_channels=1),\n",
        "                    transforms.RandomAffine(\n",
        "                        degrees=5,\n",
        "                        translate=(0.1, 0.1),\n",
        "                        scale=(0.15, 1.1),\n",
        "                        fillcolor=255\n",
        "                    ),\n",
        "                    transforms.ColorJitter(brightness=0.05, saturation=0.05),\n",
        "                    transforms.ToTensor()\n",
        "                ])\n",
        "            else:\n",
        "                self.data_transforms = transforms.Compose([\n",
        "                    transforms.Grayscale(num_output_channels=1),\n",
        "                    transforms.ToTensor()                            \n",
        "                ])\n",
        "\n",
        "        def get_next_batch(self):\n",
        "            current_batch_size = min(self.batch_size, len(self.images)-self.cursor)\n",
        "            if current_batch_size == 0: end_of_epoch = True\n",
        "            if self.has_label:\n",
        "                if self.sort_by_formulas_len:\n",
        "                    max_batch_len = len(self.formulas[self.cursor])\n",
        "                else:\n",
        "                    max_batch_len = -1\n",
        "                    for i in range(self.cursor, self.cursor+current_batch_size):\n",
        "                        max_batch_len = max(max_batch_len, len(self.formulas[i]))\n",
        "                batch_formulas_tensor = self.vocab.formulas2tensor(self.formulas[self.cursor:self.cursor+current_batch_size], max_batch_len)\n",
        "            batch_imgs = []\n",
        "            for i in range(current_batch_size):\n",
        "                img = Image.open(self.images[self.cursor])\n",
        "                img = self.data_transforms(img)\n",
        "                batch_imgs.append(img)\n",
        "                end_of_epoch = self.move_cursor()\n",
        "            if self.has_label:\n",
        "                return torch.stack(batch_imgs), batch_formulas_tensor, end_of_epoch\n",
        "            else:\n",
        "                return torch.stack(batch_imgs), end_of_epoch\n",
        "\n",
        "        def move_cursor(self):\n",
        "            self.cursor += 1\n",
        "            if len(self.images) <= self.cursor:\n",
        "                self.reset_cursor()\n",
        "                return True\n",
        "            return False\n",
        "\n",
        "        # def normalize(self, image):\n",
        "        #     return (image / 255.) * 2 - 1\n",
        "\n",
        "        def reset_cursor(self):\n",
        "            self.cursor = 0\n",
        "\n",
        "    class Logger:\n",
        "        def __init__(self, config):\n",
        "            '''\n",
        "                config: owner, log_dir, timezone\n",
        "            '''\n",
        "            self.owner = config['owner']\n",
        "            self.log_dir = config['log_dir']\n",
        "            if not os.path.isdir(os.path.dirname(self.log_dir)):\n",
        "                os.mkdir(os.path.dirname(self.log_dir))\n",
        "            self.timezone = config['timezone']\n",
        "            self.log = open(self.log_dir, 'a')\n",
        "\n",
        "        def __call__(self, message):\n",
        "            tz = pytz.timezone(self.timezone)\n",
        "            ts = pytz.utc.localize(datetime.datetime.utcnow()).astimezone(tz).strftime(\"%Y-%m-%d %H:%M:%S\")\n",
        "            print('[{}], {}: {}'.format(self.owner, ts, message))\n",
        "            print('[{}], {}: {}'.format(self.owner, ts, message), file=self.log)"
      ],
      "execution_count": 32,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ys3aRVBp3zQk"
      },
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "\n",
        "class CNN(nn.Module):\n",
        "    def __init__(self, params):\n",
        "        super(CNN, self).__init__()\n",
        "\n",
        "        self.conv1 = nn.Conv2d(in_channels=1, out_channels=params['conv1_c'],\n",
        "                               kernel_size=params['conv1_k'], stride=params['conv1_s'],\n",
        "                               padding=params['conv1_p'])\n",
        "\n",
        "        self.pool1 = nn.MaxPool2d(kernel_size=params['pool1_k'], stride=params['pool1_s'],\n",
        "                                  padding=params['pool1_p'])\n",
        "\n",
        "        self.conv2 = nn.Conv2d(in_channels=params['conv1_c'], out_channels=params['conv2_c'],\n",
        "                               kernel_size=params['conv2_k'], stride=params['conv2_s'],\n",
        "                               padding=params['conv2_p'])\n",
        "\n",
        "        self.pool2 = nn.MaxPool2d(kernel_size=params['pool2_k'], stride=params['pool2_s'],\n",
        "                                  padding=params['pool2_p'])\n",
        "\n",
        "        self.conv3 = nn.Conv2d(in_channels=params['conv2_c'], out_channels=params['conv3_c'],\n",
        "                               kernel_size=params['conv3_k'], stride=params['conv3_s'],\n",
        "                               padding=params['conv3_p'])\n",
        "\n",
        "        self.conv3_bn = nn.BatchNorm2d(params['conv3_c'])\n",
        "\n",
        "        self.conv4 = nn.Conv2d(in_channels=params['conv3_c'], out_channels=params['conv4_c'],\n",
        "                               kernel_size=params['conv4_k'], stride=params['conv4_s'],\n",
        "                               padding=params['conv4_p'])\n",
        "\n",
        "        self.pool3 = nn.MaxPool2d(kernel_size=params['pool3_k'], stride=params['pool3_s'],\n",
        "                                  padding=params['pool3_p'])\n",
        "\n",
        "        self.conv5 = nn.Conv2d(in_channels=params['conv4_c'], out_channels=params['conv5_c'],\n",
        "                               kernel_size=params['conv5_k'], stride=params['conv5_s'],\n",
        "                               padding=params['conv5_p'])\n",
        "\n",
        "        self.conv5_bn = nn.BatchNorm2d(params['conv5_c'])\n",
        "\n",
        "        self.pool4 = nn.MaxPool2d(kernel_size=params['pool4_k'], stride=params['pool4_s'],\n",
        "                                  padding=params['pool4_p'])\n",
        "\n",
        "        self.conv6 = nn.Conv2d(in_channels=params['conv5_c'], out_channels=params['conv6_c'],\n",
        "                               kernel_size=params['conv6_k'], stride=params['conv6_s'],\n",
        "                               padding=params['conv6_p'])\n",
        "\n",
        "        self.conv6_bn = nn.BatchNorm2d(params['conv6_c'])\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = F.relu(self.conv1(x))\n",
        "        x = self.pool1(x)\n",
        "        x = F.relu(self.conv2(x))\n",
        "        x = self.pool2(x)\n",
        "        x = F.relu(self.conv3_bn(self.conv3(x)))\n",
        "        x = F.relu(self.conv4(x))\n",
        "        x = self.pool3(x)\n",
        "        x = F.relu(self.conv5_bn(self.conv5(x)))\n",
        "        x = self.pool4(x)\n",
        "        x = F.relu(self.conv6_bn(self.conv6(x)))\n",
        "        return x\n"
      ],
      "execution_count": 33,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zWMGRi-B34Nq"
      },
      "source": [
        "class Encoder(nn.Module):\n",
        "    def __init__(self, input_size, hidden_size, bidirectional, device):\n",
        "        super(Encoder, self).__init__()\n",
        "        self.hidden_size = hidden_size\n",
        "        self.device = device\n",
        "        self.bidirectional = bidirectional\n",
        "        self.lstm = nn.LSTM(input_size, hidden_size,\n",
        "                            batch_first=True, bidirectional=bidirectional)\n",
        "\n",
        "    def forward(self, x):\n",
        "        # x -> (batch, time, encoder_input_size)\n",
        "        outputs, (hn, cn) = self.lstm(x, self.init_hidden(x.shape[0]))\n",
        "        # outputs: ecnoder hiddens -> (batch, time, directions x encoder_hidden_size)\n",
        "        # hn, cn -> (#layers, batch, directions x encoder_hidden_size)\n",
        "        return outputs, (self.cat_directions(hn), self.cat_directions(cn))\n",
        "\n",
        "    def init_hidden(self, batch_size):\n",
        "        if self.bidirectional:\n",
        "            return (torch.zeros(2, batch_size, self.hidden_size, device=self.device),\n",
        "                    torch.zeros(2, batch_size, self.hidden_size, device=self.device))\n",
        "        else:\n",
        "            return (torch.zeros(1, batch_size, self.hidden_size, device=self.device),\n",
        "                    torch.zeros(1, batch_size, self.hidden_size, device=self.device))\n",
        "\n",
        "    def cat_directions(self, h):\n",
        "        if self.bidirectional:\n",
        "            h = torch.cat([h[0:h.size(0):2], h[1:h.size(0):2]], 2)\n",
        "        return h\n",
        "\n",
        "class Embedding(nn.Module):\n",
        "    def __init__(self, input_size, embedded_size, padding_idx):\n",
        "        super(Embedding, self).__init__()\n",
        "        self.embedding = nn.Embedding(input_size, embedded_size, padding_idx=padding_idx)\n",
        "\n",
        "    def forward(self, x):\n",
        "        # x: symbols id -> (batch, time)\n",
        "        embedded = self.embedding(x)\n",
        "        # embedded -> (batch, time, embedded_size)\n",
        "        return embedded\n",
        "\n",
        "class BahdanauAttention(nn.Module):\n",
        "    def __init__(self, query_size, memory_size, hidden_size):\n",
        "        '''\n",
        "            query_size = decoder_hidden_size\n",
        "            memory_size = encoder_hidden_size\n",
        "        '''\n",
        "        super(BahdanauAttention, self).__init__()\n",
        "        self.query_layer = nn.Linear(query_size, hidden_size, bias=False)\n",
        "        self.memory_layer = nn.Linear(memory_size, hidden_size, bias=False)\n",
        "        self.alignment_layer = nn.Linear(hidden_size, 1, bias=False)\n",
        "\n",
        "    def score(self, query, keys):\n",
        "        query = self.query_layer(query) # (batch, 1, hidden_size)\n",
        "        keys = self.memory_layer(keys) # (batch, time, hidden_size)\n",
        "        alignments = self.alignment_layer(torch.tanh(query + keys)) # (batch, time, 1)\n",
        "        return alignments\n",
        "\n",
        "    def forward(self, query, keys):\n",
        "        '''\n",
        "            query: last hidden of top layer decoder -> (batch, 1, query_size)\n",
        "            keys: encoder outputs -> (batch, time, memory_size)\n",
        "            return context -> (batch, 1, memory_size),\n",
        "                   weights -> (batch, 1, time)\n",
        "        '''\n",
        "        scores = self.score(query, keys) # (batch, time, 1)\n",
        "        weights = F.softmax(scores, dim=1).transpose(1, 2) # (batch, 1, time)\n",
        "        context = torch.bmm(weights, keys) # (batch, 1, memory_size)\n",
        "        return context, weights\n",
        "\n",
        "class AttnDecoder(nn.Module):\n",
        "    def __init__(self, input_size, hidden_size, encoder_hidden_size, output_size, device):\n",
        "        super(AttnDecoder, self).__init__()\n",
        "        self.hidden_size = hidden_size\n",
        "        self.output_size = output_size\n",
        "        self.device = device\n",
        "        self.lstm = nn.LSTM(input_size+hidden_size, hidden_size, batch_first=True)\n",
        "        self.attn = BahdanauAttention(hidden_size, encoder_hidden_size, hidden_size)\n",
        "        self.o = nn.Linear(2*hidden_size, hidden_size, bias=False)\n",
        "        self.out = nn.Linear(hidden_size, output_size, bias=False)\n",
        "\n",
        "    def forward(self, x, hp, cp, op, encoder_outputs):\n",
        "        # x -> (batch, 1, input_size)\n",
        "        # op -> (batch, hidden_size)\n",
        "        # hp, cp -> (#layers, batch, hidden_size)\n",
        "        # encoder_outputs -> (batch, time, #directions*encoder_hidden_size)\n",
        "        lstm_input = torch.cat([x, op.unsqueeze(dim=1)], dim=2) # (batch, 1, input_size + hidden_size)\n",
        "        hidden, (hn, cn) = self.lstm(lstm_input, (hp, cp))\n",
        "        # hidden -> (batch, 1, hidden_size)\n",
        "        # (hn, cn) -> (#layers, batch, hidden_size)\n",
        "        context, attn_weights = self.attn(hidden, encoder_outputs)\n",
        "        # context -> (batch, 1, hidden_size)\n",
        "        # attn_weights -> (batch, 1, time)\n",
        "        o_input = torch.cat([hidden.squeeze(1), context.squeeze(1)], dim=1) # (batch, 2*hidden_size)\n",
        "        on = torch.tanh(self.o(o_input))\n",
        "        # next_o -> (batch, hidden_size)\n",
        "        output = F.log_softmax(self.out(on), dim=1)\n",
        "        # output: log softmax of symbol scores -> (batch, output_size)\n",
        "        return output, (hn, cn), on, attn_weights"
      ],
      "execution_count": 34,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LFK_p6r_4Eup"
      },
      "source": [
        "from torch import optim\n",
        "import numpy as np\n",
        "\n",
        "class GreedyDecoder:\n",
        "    def __init__(self, model, max_len, hn, cn, on, encoder_outputs, device):\n",
        "        self.model = model\n",
        "        self.max_len = max_len\n",
        "        self.hn = hn\n",
        "        self.cn = cn\n",
        "        self.on = on\n",
        "        self.encoder_outputs = encoder_outputs\n",
        "        self.device = device\n",
        "\n",
        "    def generate(self, start_token):\n",
        "        next_token = torch.Tensor([start_token]).long().to(self.device) # (1)\n",
        "        b = self.encoder_outputs.shape[0]\n",
        "        next_token = next_token.repeat(b) # (batch)\n",
        "        logits = []\n",
        "        preds = [next_token]\n",
        "        attn_weights = []\n",
        "        hn = self.hn\n",
        "        cn = self.cn\n",
        "        on = self.on\n",
        "        for t in range(1, self.max_len):\n",
        "            output, (hn, cn), on, attn_weight = self.model.decoder_step(next_token, hn, cn, on, self.encoder_outputs)\n",
        "            attn_weights.append(attn_weight.squeeze(1))\n",
        "            logits.append(output)\n",
        "            next_token = output.argmax(dim=1) # (batch)\n",
        "            preds.append(next_token)\n",
        "        logits = [torch.zeros(logits[1].shape[0], logits[1].shape[1], device=self.device)] + logits\n",
        "        logits[0][:, start_token] = 1\n",
        "        logits = torch.stack(logits, dim=1)\n",
        "        preds = torch.stack(preds, dim=1)\n",
        "        attn_weights = torch.stack(attn_weights, dim=1)\n",
        "        return preds, logits, attn_weights\n",
        "\n",
        "class Model(nn.Module):\n",
        "    def __init__(self, cnn, encoder, embedding, decoder, device):\n",
        "        super(Model, self).__init__()\n",
        "        self.cnn = cnn\n",
        "        self.encoder = encoder\n",
        "        self.embedding = embedding\n",
        "        self.decoder = decoder\n",
        "        self.device = device\n",
        "\n",
        "        directions = 2 if self.encoder.bidirectional else 1\n",
        "\n",
        "        self.init_h = nn.Linear(directions*self.encoder.hidden_size, self.decoder.hidden_size).to(self.device)\n",
        "        self.init_o = nn.Linear(directions*self.encoder.hidden_size, self.decoder.hidden_size).to(self.device)\n",
        "\n",
        "    def get_cnn_feature(self, x):\n",
        "        '''\n",
        "            input:\n",
        "                > x: image (batch, 1, H, W)\n",
        "            return:\n",
        "                > cnn feature map (batch, C, H', W')\n",
        "        '''\n",
        "        x = self.cnn(x)\n",
        "        return x\n",
        "\n",
        "    def get_encoder_anotation_grid(self, x):\n",
        "        '''\n",
        "            input:\n",
        "                > x: cnn feature map (batch, C, H', W')\n",
        "\n",
        "            return:\n",
        "                > encoder_outputs: top layer hidden states of encoder applied on H' row of x\n",
        "                                   (batch, H', W', directions * encoder_hidden_size)\n",
        "                > hn, cn: last hidden states of encoder applied on H' row of x\n",
        "                          (#layers, batch, H', decoder_hidden_size)\n",
        "\n",
        "        '''\n",
        "        B, C, H_prime, W_prime = x.shape\n",
        "        x = x.permute(0, 2, 3, 1).contiguous().view(B*H_prime, W_prime, C) # (batch * H', W', C)\n",
        "        encoder_outputs, (hn, cn) = self.encoder(x)\n",
        "        # encoder_outputs -> (batch * H', W', directions * encoder_hidden_size)\n",
        "        # hn, cn -> (#layers, batch * H', directions * encoder_hidden_size)\n",
        "        encoder_outputs = encoder_outputs.view(B, H_prime, W_prime, -1).contiguous()\n",
        "        hn = hn.view(hn.shape[0], B, H_prime, -1).contiguous()\n",
        "        cn = hn.view(hn.shape[0], B, H_prime, -1).contiguous()\n",
        "        return encoder_outputs, (hn, cn)\n",
        "\n",
        "    def add_positional_embedding(self, x):\n",
        "        '''\n",
        "            input:\n",
        "                > x: (batch, W'*H', directions * encoder_hidden_size)\n",
        "            return:\n",
        "                > x + positional_embedding\n",
        "        '''\n",
        "        n_position = x.shape[1]\n",
        "        emb_dim = x.shape[2]\n",
        "        position_enc = np.array([\n",
        "                        [pos / np.power(10000, 2 * (j // 2) / emb_dim) for j in range(emb_dim)]\n",
        "                         for pos in range(n_position)])\n",
        "        position_enc[1:, 0::2] = np.sin(position_enc[1:, 0::2])\n",
        "        position_enc[1:, 1::2] = np.cos(position_enc[1:, 1::2])\n",
        "        return x + torch.from_numpy(position_enc).type(torch.FloatTensor).to(self.device)\n",
        "\n",
        "    def decoder_step(self, input_token, hp, cp, op, encoder_outputs):\n",
        "        '''\n",
        "            input:\n",
        "                > input_token: one hot input token (batch)\n",
        "                > hp, cp: previous states (#layers, batch, decoder_hidden_size)\n",
        "                > op: previous o (batch, decoder_hidden_size)\n",
        "                > encoder_outputs: top layer encoder states (batch, H'*W', directions * encoder_hidden_size)\n",
        "            return:\n",
        "                > output: log probability of next token (batch, vocab_size)\n",
        "                > hn, cn: next hidden states (#layers, batch, decoder_hidden_size)\n",
        "                > on: next (batch, decoder_hidden_size)\n",
        "                > attn_weight: attention alignments (batch, 1, H'*W')\n",
        "        '''\n",
        "        token_embedding = self.embedding(input_token).unsqueeze(1) # (batch, 1, embedding_size)\n",
        "        output, (hn, cn), on, attn_weight = self.decoder(token_embedding, hp, cp, op, encoder_outputs)\n",
        "        return output, (hn, cn), on, attn_weight\n",
        "\n",
        "    def forward(self, x, y, teacher_forcing_ratio):\n",
        "        '''\n",
        "            input:\n",
        "                > x: image (batch, 1, H, W)\n",
        "                > y: one hot formulas (batch, len)\n",
        "                > teacher_forcing_ratio\n",
        "            return:\n",
        "                > logits: log probability of tokens (batch, len-1, vocab_size) [excluding start token]\n",
        "        '''\n",
        "        x = self.get_cnn_feature(x)\n",
        "\n",
        "        encoder_outputs, (hn, cn) = self.get_encoder_anotation_grid(x)\n",
        "\n",
        "        b, h, w, _ = encoder_outputs.shape\n",
        "        encoder_outputs = encoder_outputs.view(b, h*w, -1).contiguous() # (batch, H'*W', directions * encoder_hidden_size)\n",
        "        encoder_outputs = self.add_positional_embedding(encoder_outputs) # (batch, H'*W', directions * encoder_hidden_size)\n",
        "\n",
        "        hn = torch.tanh(self.init_h(torch.mean(encoder_outputs, dim=1))) # (batch, decoder_hidden_size)\n",
        "        hn = hn.view(1, hn.shape[0], hn.shape[1]).contiguous() # (1, batch, decoder_hidden_size)\n",
        "        cn = torch.zeros(1, b, self.decoder.hidden_size, device=self.device) # (1, batch, decoder_hidden_size)\n",
        "        on = torch.tanh(self.init_o(torch.mean(encoder_outputs, dim=1))) # (batch, decoder_hidden_size)\n",
        "\n",
        "        next_token = y[:, 0] # (batch)\n",
        "        logits = torch.zeros(y.shape[0], y.shape[1], self.decoder.output_size, device=self.device) # (batch, len, vocab_size)\n",
        "        logits[:, 0, y[0, 0]] = 1\n",
        "        for t in range(1, y.shape[1]):\n",
        "            output, (hn, cn), on, attn_weight = self.decoder_step(next_token, hn, cn, on, encoder_outputs)\n",
        "            logits[:, t, :] = output\n",
        "            if teacher_forcing_ratio < torch.rand(1).item():\n",
        "                next_token = output.exp().multinomial(1).squeeze(1) # (batch)\n",
        "            else:\n",
        "                next_token = y[:, t] # (batch)\n",
        "        return logits\n",
        "\n",
        "    def generate(self, x, start_token, max_len, method):\n",
        "        '''\n",
        "            input:\n",
        "                > x: image (batch, 1, H, W)\n",
        "                > start_token: start token index\n",
        "                > max_len: length of all generated formulas would be eless than max_len\n",
        "                > method: 'greedy' or 'beam-seach'\n",
        "            return:\n",
        "                > preds: generated sequences (batch, max_len)\n",
        "                > logits: log probability of tokens (batch, max_len-1, vocab_size)\n",
        "                > attn_weights: attention alignments (batch, max_len-1, H'*W')\n",
        "        '''\n",
        "        x = self.get_cnn_feature(x)\n",
        "        encoder_outputs, (hn, cn) = self.get_encoder_anotation_grid(x)\n",
        "        b, h, w, _ = encoder_outputs.shape\n",
        "        encoder_outputs = encoder_outputs.view(b, h*w, -1)\n",
        "        encoder_outputs = self.add_positional_embedding(encoder_outputs) # (batch, H'*W', directions * encoder_hidden_size)\n",
        "\n",
        "        hn = torch.tanh(self.init_h(torch.mean(encoder_outputs, dim=1))) # (batch, decoder_hidden_size)\n",
        "        hn = hn.view(1, hn.shape[0], hn.shape[1]).contiguous() # (1, batch, decoder_hidden_size)\n",
        "        cn = torch.zeros(1, b, self.decoder.hidden_size, device=self.device) # (1, batch, decoder_hidden_size)\n",
        "        on = torch.tanh(self.init_o(torch.mean(encoder_outputs, dim=1))) # (batch, decoder_hidden_size)\n",
        "\n",
        "        if method == 'greedy':\n",
        "            generator = GreedyDecoder(self, max_len, hn, cn, on, encoder_outputs, self.device)\n",
        "\n",
        "        elif method == 'beam-search':\n",
        "            raise NotImplementedError('beam-search method is not implemented yet.')\n",
        "            generator = BeamSearchDecoder()\n",
        "\n",
        "        preds, logits, attn_weights = generator.generate(start_token)\n",
        "        return preds, logits, attn_weights\n"
      ],
      "execution_count": 35,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Hp8Kyl2t482V"
      },
      "source": [
        "class Trainer:\n",
        "    def __init__(self, model, train_loader, config):\n",
        "        '''\n",
        "            config: device, checkpoints_dir, log_dir, print_every_batch, clip\n",
        "                    learning_rate, learning_rate_decay, learning_rate_decay_step, learning_rate_min\n",
        "                    teacher_forcing_ratio, teacher_forcing_ratio_decay, teacher_forcing_ratio_decay_step, teacher_forcing_ratio_min\n",
        "        '''\n",
        "        self.model = model\n",
        "        self.train_loader = train_loader\n",
        "        self.device = config['device']\n",
        "\n",
        "        self.criterian = nn.NLLLoss(ignore_index=train_loader.vocab.pad_token)\n",
        "        self.optimizer = optim.Adam(self.model.parameters(), lr=config['learning_rate'])\n",
        "        self.scheduler = optim.lr_scheduler.StepLR(self.optimizer, step_size=config['learning_rate_decay_step'],\n",
        "                                                    gamma=config['learning_rate_decay'])\n",
        "        self.learning_rate_min = config['learning_rate_min']\n",
        "\n",
        "        self.checkpoints_dir = config['checkpoints_dir']\n",
        "\n",
        "        logger_config = {\n",
        "            'owner': 'Trainer.py',\n",
        "            'log_dir' : config['log_dir'],\n",
        "            'timezone' : 'Iran',\n",
        "        }\n",
        "        self.logger = utils.Logger(logger_config)\n",
        "\n",
        "        self.clip = config['clip']\n",
        "\n",
        "        self.checkpoint_name = lambda id: 'snapshot-{:02d}.pt'.format(id)\n",
        "\n",
        "        self.print_every_batch = config['print_every_batch']\n",
        "        self.current_epoch = 0\n",
        "\n",
        "        self.teacher_forcing_ratio = config['teacher_forcing_ratio']\n",
        "        self.teacher_forcing_ratio_decay = config['teacher_forcing_ratio_decay']\n",
        "        self.teacher_forcing_ratio_decay_step = config['teacher_forcing_ratio_decay_step']\n",
        "        self.teacher_forcing_ratio_min = config['teacher_forcing_ratio_min']\n",
        "\n",
        "    def train_one_epoch(self):\n",
        "        self.logger('start training one epoch')\n",
        "        self.model.train()\n",
        "        epoch_ended = False\n",
        "        batch_counter = 0\n",
        "        epoch_loss = 0\n",
        "        epoch_acc = 0\n",
        "        predictions = []\n",
        "        while not epoch_ended:\n",
        "            x_train, y_train, epoch_ended = self.train_loader.get_next_batch()\n",
        "            x_train = x_train.to(self.device)\n",
        "            y_train = torch.from_numpy(y_train).long().to(self.device)\n",
        "            self.optimizer.zero_grad()\n",
        "            logits = self.model(x_train, y_train, self.teacher_forcing_ratio)\n",
        "            loss = self.criterian(logits[:, 1:, :].contiguous().view(-1, logits.shape[-1]),\n",
        "                                  y_train[:, 1:].contiguous().view(-1))\n",
        "            loss.backward()\n",
        "            torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.clip)\n",
        "            self.optimizer.step()\n",
        "            mask = (y_train != self.train_loader.vocab.pad_token)\n",
        "            preds = logits.argmax(dim=2)\n",
        "            acc = ((y_train == preds) * mask).sum().item() / mask.sum().item()\n",
        "            if batch_counter % self.print_every_batch == 0:\n",
        "                self.logger('Batch {}: loss={}, acc={}, lr={}'.format(batch_counter, loss, acc, get_lr(self.optimizer)))\n",
        "            epoch_loss += loss.item()\n",
        "            epoch_acc += acc\n",
        "            for i in range(preds.shape[0]):\n",
        "                pred = preds[i].cpu().numpy()\n",
        "                predictions.append(self.train_loader.vocab.tensor2formula(pred, pretty=True))\n",
        "            batch_counter += 1\n",
        "        self.logger('Epoch finished, loss={} acc={}, lr={}'.format(epoch_loss/batch_counter, epoch_acc/batch_counter, get_lr(self.optimizer)))\n",
        "        if get_lr(self.optimizer) > self.learning_rate_min:\n",
        "            self.scheduler.step()\n",
        "        if self.current_epoch % self.teacher_forcing_ratio_decay_step == 0:\n",
        "            self.teacher_forcing_ratio = max(self.teacher_forcing_ratio_min,\n",
        "                                            self.teacher_forcing_ratio_decay * self.teacher_forcing_ratio)\n",
        "        self.current_epoch += 1\n",
        "        self.logger('training one epoch finished.')\n",
        "        return predictions, epoch_loss/batch_counter, epoch_acc/batch_counter\n",
        "\n",
        "    def evaluate(self, eval_loader, method):\n",
        "        self.logger('evaluation starts.')\n",
        "        self.model.eval()\n",
        "        start_token = self.train_loader.vocab.start_token\n",
        "        max_len = self.train_loader.max_len\n",
        "        predictions = []\n",
        "        with torch.no_grad():\n",
        "            epoch_ended = False\n",
        "            batch_counter = 0\n",
        "            epoch_loss = 0\n",
        "            epoch_acc = 0\n",
        "            while not epoch_ended:\n",
        "                if eval_loader.has_label:\n",
        "                    x_eval, y_eval, epoch_ended = eval_loader.get_next_batch()\n",
        "                    y_eval = torch.from_numpy(y_eval).long().to(self.device)\n",
        "                else:\n",
        "                    x_eval, epoch_ended = eval_loader.get_next_batch()\n",
        "                # x_eval = torch.from_numpy(x_eval).float().to(self.device)\n",
        "                x_eval = x_eval.to(self.device)\n",
        "                if eval_loader.has_label:\n",
        "                    max_len = y_eval.shape[1]\n",
        "                preds, logits, attn_weights = self.model.generate(x_eval, start_token, max_len, method)\n",
        "                if eval_loader.has_label:\n",
        "                    epoch_loss += self.criterian(logits[:, 1:, :].contiguous().view(-1, logits.shape[-1]),\n",
        "                                                 y_eval[:, 1:].contiguous().view(-1))\n",
        "                    mask = (y_eval != self.train_loader.vocab.pad_token)\n",
        "                    acc = ((y_eval == preds) * mask).sum().item() / mask.sum().item()\n",
        "                    epoch_acc += acc\n",
        "                for i in range(preds.shape[0]):\n",
        "                    pred = preds[i].cpu().numpy()\n",
        "                    predictions.append(self.train_loader.vocab.tensor2formula(pred, pretty=True))\n",
        "                batch_counter += 1\n",
        "                if batch_counter % self.print_every_batch == 0:\n",
        "                    self.logger('batch {} completed.'.format(batch_counter))\n",
        "            self.logger('evaluation finished.')\n",
        "            if eval_loader.has_label:\n",
        "                return predictions, attn_weights, epoch_loss/batch_counter, epoch_acc/batch_counter\n",
        "            else:\n",
        "                return predictions, attn_weights\n",
        "\n",
        "    def save(self, epoch, loss, acc):\n",
        "        if not os.path.isdir(self.checkpoints_dir):\n",
        "            os.mkdir(self.checkpoints_dir)\n",
        "        torch.save({\n",
        "            'epoch' : epoch,\n",
        "            'acc' : acc,\n",
        "            'loss' : loss,\n",
        "            'model_state_dict' : self.model.state_dict(),\n",
        "            'optimizer_state_dict' : self.optimizer.state_dict(),\n",
        "            'scheduler_state_dict' : self.scheduler.state_dict(),\n",
        "        }, os.path.join(self.checkpoints_dir, self.checkpoint_name(epoch)))\n",
        "        self.logger('Epoch {} saved.'.format(epoch))\n",
        "        return os.path.join(self.checkpoints_dir, self.checkpoint_name(epoch))\n",
        "\n",
        "    def load(self, checkpoint_name, cpu=False):\n",
        "        checkpoint = torch.load(checkpoint_name, map_location=torch.device('cpu') if cpu else None)\n",
        "        self.model.load_state_dict(checkpoint['model_state_dict'])\n",
        "        self.optimizer.load_state_dict(checkpoint['optimizer_state_dict'])\n",
        "        self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])\n",
        "        self.current_epoch = checkpoint['epoch']\n",
        "        self.logger('{} loaded.'.format(checkpoint_name))\n",
        "\n",
        "def get_lr(optimizer):\n",
        "    for param_group in optimizer.param_groups:\n",
        "        return param_group['lr']\n"
      ],
      "execution_count": 36,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "275fMMyw-61j"
      },
      "source": [
        "def get_seq2seq_model(config, vocab, checkpoint_name, cpu=True):\n",
        "    cnn_model = CNN(config['cnn_params']).to(config['device'])\n",
        "\n",
        "    encoder_model = Encoder(config['cnn_params']['conv6_c'], config['encoder_hidden_size'],\n",
        "                                    config['bidirectional'], config['device']).to(config['device'])\n",
        "    vocab_size = len(vocab.token2idx.keys())\n",
        "    embedding_model = Embedding(vocab_size, config['embedding_size'], vocab.pad_token).to(config['device'])\n",
        "\n",
        "    decoder_model = AttnDecoder(config['embedding_size'], config['decoder_hidden_size'],\n",
        "                                        config['encoder_hidden_size']*(2 if config['bidirectional'] else 1), vocab_size,\n",
        "                                        config['device']).to(config['device'])\n",
        "\n",
        "    model = Model(cnn_model, encoder_model, embedding_model, decoder_model, config['device'])\n",
        "\n",
        "    checkpoint = torch.load(checkpoint_name, map_location=torch.device('cpu') if cpu else None)\n",
        "    model.load_state_dict(checkpoint['model_state_dict'])\n",
        "\n",
        "    return model"
      ],
      "execution_count": 37,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G7UUl7BC9hav"
      },
      "source": [
        "seq2seq_config = {\n",
        "    # ======================\n",
        "    #        pathes\n",
        "    # ======================\n",
        "\n",
        "    'formulas_train_path' : 'train.txt',\n",
        "    'formulas_validation_path' : 'validate.txt',\n",
        "    'formulas_test_path' : 'test.txt',\n",
        "    'images_train_path' : 'train_imgs',\n",
        "    'images_validation_path' : 'val_imgs',\n",
        "    'images_test_path' : 'test_imgs',\n",
        "\n",
        "    'checkpoints_dir' : 'checkpoints/',\n",
        "\n",
        "    'log_dir' : 'logs/log.txt',\n",
        "    # ======================\n",
        "    #   general parameters\n",
        "    # ======================\n",
        "    'batch_size' : 32,\n",
        "    'epochs' : 40,\n",
        "    'print_every_batch' : 100,\n",
        "    'learning_rate' : 1e-3,\n",
        "    'learning_rate_decay' : .5,\n",
        "    'learning_rate_decay_step' : 3,\n",
        "    'learning_rate_min': 1e-6,\n",
        "    'teacher_forcing_ratio' : 1,\n",
        "    'teacher_forcing_ratio_decay' : 0.95,\n",
        "    'teacher_forcing_ratio_decay_step' : 2,\n",
        "    'teacher_forcing_ratio_min': 0.7,\n",
        "    'clip' : 5,\n",
        "    'unk_token_threshold' : 10,\n",
        "    'generation_method' : 'greedy', # or 'beam-search'\n",
        "    'device' : torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\"),\n",
        "    # ======================\n",
        "    #   CNN parameters\n",
        "    # ======================\n",
        "    'cnn_params' : {\n",
        "        'conv1_c' : 64,\n",
        "        'conv1_k' : (3, 3),\n",
        "        'conv1_s' : (1, 1),\n",
        "        'conv1_p' : (1, 1),\n",
        "\n",
        "        'pool1_k' : (2, 2),\n",
        "        'pool1_s' : (2, 2),\n",
        "        'pool1_p' : (0, 0),\n",
        "\n",
        "        'conv2_c' : 128,\n",
        "        'conv2_k' : (3, 3),\n",
        "        'conv2_s' : (1, 1),\n",
        "        'conv2_p' : (1, 1),\n",
        "\n",
        "        'pool2_k' : (2, 2),\n",
        "        'pool2_s' : (2, 2),\n",
        "        'pool2_p' : (0, 0),\n",
        "\n",
        "        'conv3_c' : 256,\n",
        "        'conv3_k' : (3, 3),\n",
        "        'conv3_s' : (1, 1),\n",
        "        'conv3_p' : (1, 1),\n",
        "\n",
        "        'conv4_c' : 256,\n",
        "        'conv4_k' : (3, 3),\n",
        "        'conv4_s' : (1, 1),\n",
        "        'conv4_p' : (1, 1),\n",
        "\n",
        "        'pool3_k' : (2, 1),\n",
        "        'pool3_s' : (2, 1),\n",
        "        'pool3_p' : (0, 0),\n",
        "\n",
        "        'conv5_c' : 512,\n",
        "        'conv5_k' : (3, 3),\n",
        "        'conv5_s' : (1, 1),\n",
        "        'conv5_p' : (1, 1),\n",
        "\n",
        "        'pool4_k' : (1, 2),\n",
        "        'pool4_s' : (1, 2),\n",
        "        'pool4_p' : (0, 0),\n",
        "\n",
        "        'conv6_c' : 512,\n",
        "        'conv6_k' : (3, 3),\n",
        "        'conv6_s' : (1, 1),\n",
        "        'conv6_p' : (1, 1),\n",
        "    },\n",
        "    # ======================\n",
        "    #   seq2seq parameters\n",
        "    # ======================\n",
        "    'embedding_size' : 80,\n",
        "    'decoder_hidden_size' : 512,\n",
        "    'encoder_hidden_size' : 256,\n",
        "    'bidirectional' : True,\n",
        "}"
      ],
      "execution_count": 38,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zUbWQBWiccmP",
        "outputId": "e2abf2c2-f366-4e50-d2f2-b941bbeda520",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!wget https://students.mimuw.edu.pl/~mt406390/machine_learning/armin/armin_aug_100k/snapshot-15.pt\n",
        "!mv snapshot-15.pt models\n",
        "model_weights_file = 'snapshot-15.pt'"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-11-02 17:22:22--  https://students.mimuw.edu.pl/~mt406390/machine_learning/armin/armin_aug_100k/snapshot-15.pt\n",
            "Resolving students.mimuw.edu.pl (students.mimuw.edu.pl)... 193.0.96.129, 2001:6a0:5001:1::3\n",
            "Connecting to students.mimuw.edu.pl (students.mimuw.edu.pl)|193.0.96.129|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 119714352 (114M)\n",
            "Saving to: ‘snapshot-15.pt’\n",
            "\n",
            "snapshot-15.pt      100%[===================>] 114.17M  3.66MB/s    in 22s     \n",
            "\n",
            "2020-11-02 17:22:45 (5.14 MB/s) - ‘snapshot-15.pt’ saved [119714352/119714352]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k8oO9wJMB3WS",
        "outputId": "a260d013-6e98-4340-e40f-267a66849100",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!wget https://students.mimuw.edu.pl/~mt406390/machine_learning/vocab.pkl\n",
        "!mkdir vocab\n",
        "!mv vocab.pkl vocab"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-11-02 17:22:45--  https://students.mimuw.edu.pl/~mt406390/machine_learning/vocab.pkl\n",
            "Resolving students.mimuw.edu.pl (students.mimuw.edu.pl)... 193.0.96.129, 2001:6a0:5001:1::3\n",
            "Connecting to students.mimuw.edu.pl (students.mimuw.edu.pl)|193.0.96.129|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 2399 (2.3K)\n",
            "Saving to: ‘vocab.pkl’\n",
            "\n",
            "\rvocab.pkl             0%[                    ]       0  --.-KB/s               \rvocab.pkl           100%[===================>]   2.34K  --.-KB/s    in 0s      \n",
            "\n",
            "2020-11-02 17:22:45 (74.1 MB/s) - ‘vocab.pkl’ saved [2399/2399]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "slWSihDfBGKb"
      },
      "source": [
        "def get_vocab():\n",
        "    seq2seq_config['pathes'] = ''\n",
        "    vocab = utils.Vocab(seq2seq_config)\n",
        "    return vocab"
      ],
      "execution_count": 41,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "10rTF8YMFnlz"
      },
      "source": [
        "vocab = get_vocab()"
      ],
      "execution_count": 42,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dC2wT6RKue7n"
      },
      "source": [
        "from pathlib import Path\n",
        "models = Path('models')\n",
        "mdl = get_seq2seq_model(seq2seq_config, vocab, str(models / model_weights_file), cpu=True)"
      ],
      "execution_count": 43,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mVnsZWHgBqr5"
      },
      "source": [
        "class LatexProducerWrapper(Predictor):\n",
        "    def __init__(self, learn: Learner, seq2seq_model, vocab, gen_method, max_len):\n",
        "        super().__init__(learn)\n",
        "        self.model = seq2seq_model\n",
        "        self.vocab = vocab\n",
        "        self.gen_method = gen_method\n",
        "        self.max_len = max_len\n",
        "\n",
        "    def _get_single_pred(self, img_tensor):\n",
        "        with torch.no_grad():\n",
        "            preds, logits, attn_weights = self.model.generate(\n",
        "                img_tensor, self.vocab.start_token, self.max_len, self.gen_method\n",
        "        )\n",
        "        pred = preds[0].cpu().numpy()\n",
        "\n",
        "        return vocab.tensor2formula(pred, pretty=True)\n",
        "\n",
        "    def predict(self, img: PILImageBW):\n",
        "        return self._get_single_pred(to_device(self._transform_img(img), default_device()))\n",
        "\n",
        "seq2seq_wrap = LatexProducerWrapper(learn, mdl, vocab, 'greedy', max_len=50)"
      ],
      "execution_count": 44,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bAMaEmW-Dyr9"
      },
      "source": [
        "# Final inference"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t5abFwAuswkI"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "from matplotlib.patches import Rectangle\n",
        "\n",
        "class Inference:\n",
        "    def __init__(self, fastai_model: Model, leaf_producer: LatexProducerWrapper,\n",
        "                leaf_threshold=None):\n",
        "        self.model = fastai_model\n",
        "        self.leaf_producer = leaf_producer\n",
        "        self.MAX_DEPTH = 8\n",
        "        self.leaf_threshold = self.MAX_DEPTH if leaf_threshold is None \\\n",
        "            else leaf_threshold\n",
        "\n",
        "    def _show_pred(self, input: PILImageBW, pred: Prediction):\n",
        "        print(f'pred: {pred.op.op_id}')\n",
        "        for box in pred.bboxes:\n",
        "            plt.gca().add_patch(Rectangle((box.x, box.y), box.dx, box.dy,linewidth=1,edgecolor='green',facecolor='none'))\n",
        "        plt.imshow(input,  cmap='Greys_r')\n",
        "        # plt.axis('off')\n",
        "        plt.show()\n",
        "\n",
        "    def __call__(self, input: PILImageBW, depth=0, debug=False) -> str:\n",
        "        if depth >= self.MAX_DEPTH:\n",
        "            raise ValueError('stack overflow - your model too bad bro')\n",
        "\n",
        "        if depth >= self.leaf_threshold:\n",
        "            return self.leaf_producer.predict(input)\n",
        "\n",
        "        pred = self.model.predict(input)\n",
        "\n",
        "        if debug:\n",
        "            self._show_pred(input, pred)\n",
        "            \n",
        "\n",
        "        operands = ops_cfg[pred.op.op_id]['operands']\n",
        "        latex = ops_cfg[pred.op.op_id]['latex']\n",
        "\n",
        "        if operands == 0:\n",
        "            return self.leaf_producer.predict(input)\n",
        "        \n",
        "        inputs = [resize_and_center_pred(input, pred.bboxes[i]) for i in range(operands)]\n",
        "\n",
        "        return latex.format(*(self(inp, depth + 1, debug) for inp in inputs))"
      ],
      "execution_count": 45,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QRnufpeefvJH",
        "outputId": "3162a52d-b743-4abc-e60f-4510254cfa5a",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!wget https://students.mimuw.edu.pl/~mt406390/machine_learning/dataset_sota_1k_100_100_1-4.zip\n",
        "!unzip -qq dataset_sota_1k_100_100_1-4.zip -d test_set/"
      ],
      "execution_count": 46,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-11-02 17:22:46--  https://students.mimuw.edu.pl/~mt406390/machine_learning/dataset_sota_1k_100_100_1-4.zip\n",
            "Resolving students.mimuw.edu.pl (students.mimuw.edu.pl)... 193.0.96.129, 2001:6a0:5001:1::3\n",
            "Connecting to students.mimuw.edu.pl (students.mimuw.edu.pl)|193.0.96.129|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 7614642 (7.3M) [application/zip]\n",
            "Saving to: ‘dataset_sota_1k_100_100_1-4.zip’\n",
            "\n",
            "dataset_sota_1k_100 100%[===================>]   7.26M  11.0MB/s    in 0.7s    \n",
            "\n",
            "2020-11-02 17:22:47 (11.0 MB/s) - ‘dataset_sota_1k_100_100_1-4.zip’ saved [7614642/7614642]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-x9Rgxzgfyev",
        "outputId": "31249a38-0a2c-4b15-9490-4f3b6d6b0f74",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "import glob\n",
        "\n",
        "test_set = Path('test_set')\n",
        "test_images = glob.glob(str(test_set / 'test_imgs' / '*.png'))\n",
        "test_images= sorted(test_images, key=lambda x: int(x.split('/')[-1].split('.')[0]))\n",
        "print(len(test_images))"
      ],
      "execution_count": 47,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AAbHeVLb8-U3"
      },
      "source": [
        "# Recursive CNN Inference example\n",
        "\n",
        "---\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W3wdtrPZF-8z",
        "outputId": "a65a430e-0e08-44f0-cc68-a551424e7d86",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 241
        }
      },
      "source": [
        "def img_from_path(path: Path) -> PILImageBW:\n",
        "    return PILImageBW.create(path)\n",
        "\n",
        "img = img_from_path(test_images[600])\n",
        "img"
      ],
      "execution_count": 48,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "PILImageBW mode=L size=224x224"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UVdaH19j8Ml1",
        "outputId": "18c86ed5-171d-46c9-db19-81cef85867a1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "example_infer = Inference(fastai_model, seq2seq_wrap)\n",
        "example_infer(img, debug=True)"
      ],
      "execution_count": 49,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "pred: star\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: par\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: mul\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: gpar\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: leaf\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: leaf\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: frac\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: gpar\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAD8CAYAAAB3lxGOAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAWIklEQVR4nO3de5BU5ZnH8e8zzTATbgPDjFPAAAOCJiqXkKmIlSga47W2QryUShIkiVliGati1q1Vk5Qxmz/WzWqyWptoaWmCW17ixmRDlbhqSCqiAbEBITJcxRkEBgZBBhhgrs/+0WfGnpsM0910d97fp2qqT7/ndPfTNvOb95zTnsfcHREJV0G2CxCR7FIIiAROISASOIWASOAUAiKBUwiIBC5jIWBmV5rZFjPbbmZ3Z+p1RCQ1lonvCZhZDNgKXAbsAt4CFrh7TdpfTERSkqmZwGeB7e6+w91bgOeA+Rl6LRFJwZAMPe8E4P2k+7uA8/vbuKyszKuqqjJUiogArFmz5gN3L+85nqkQOCkzWwwsBpg0aRLxeDxbpYgEwczq+hrP1O7AbmBi0v3KaKyLuz/m7tXuXl1e3iucROQ0yVQIvAVMN7MpZjYUuAlYmqHXEpEUZGR3wN3bzOx24GUgBjzp7hsz8VoikpqMHRNw92XAskw9v4ikh74xKBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoEbdAiY2UQz+7OZ1ZjZRjP7bjR+n5ntNrO3o5+r01euiKRbKhcVaQPudPe1ZjYSWGNmr0brfu7uD6Renohk2qBDwN3rgfpo+YiZbSJxqXERySNpOSZgZlXAp4E3o6HbzWyDmT1pZmPS8Roikhkph4CZjQBeAO5w98PAI8CZwGwSM4UH+3ncYjOLm1l8//79qZYhIoOUUgiYWSGJAHja3X8H4O773L3d3TuAx0m0JOtFfQdEckMqZwcMeALY5O4/Sxofl7TZNcA7gy9PRDItlbMDnwMWAn8zs7ejse8DC8xsNuBALfDtlCoUkYxK5ezA64D1sUq9BkTyiL4xKBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgELpUrCwFgZrXAEaAdaHP3ajMrBX4DVJG4utAN7v5hqq8lIumXrpnAJe4+292ro/t3A8vdfTqwPLovIjkoU7sD84El0fIS4MsZeh0RSVE6QsCBV8xsjZktjsYqog5FAHuBip4PUt8BkdyQ8jEB4PPuvtvMzgBeNbPNySvd3c3Mez7I3R8DHgOorq7utV5ETo+UZwLuvju6bQB+T6LZyL7O/gPRbUOqryMimZFqB6LhUUdizGw4cDmJZiNLgUXRZouAP6TyOiKSOanuDlQAv080I2II8Iy7/5+ZvQU8b2a3AHXADSm+johkSEoh4O47gFl9jB8ALk3luUXk9NA3BkUCpxAQCZxCQCRwCgGRwCkERAKnEBAJnEJAJHAKAZHAKQREAqcQEAmcQkAkcAoBkcApBEQCpxAQCZxCQCRwg76egJmdTaK3QKepwL3AaOAfgc6rh37f3ZcNukIRyahBh4C7bwFmA5hZDNhN4hqD3wB+7u4PpKVCEcmodO0OXAq86+51aXo+ETlN0hUCNwHPJt2/3cw2mNmTZjYmTa8hIhmQcgiY2VDgS8D/REOPAGeS2FWoBx7s53FqPiKSA9IxE7gKWOvu+wDcfZ+7t7t7B/A4iT4Evbj7Y+5e7e7V5eXlaShDRAYjHSGwgKRdgc6mI5FrSPQhEJEcldIlx6OGI5cB304a/qmZzSbRo7C2xzoRyTGp9h1oAsb2GFuYUkUiclrpG4MigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASuAGFQHTB0AYzeydprNTMXjWzbdHtmGjczOxhM9seXWx0TqaKF5HUDXQm8Gvgyh5jdwPL3X06sDy6D4lrDk6PfhaTuPCoiOSoAYWAu78GHOwxPB9YEi0vAb6cNP6UJ6wCRve47qCI5JBUjglUuHt9tLwXqIiWJwDvJ223KxoTkRyUlgOD7u4kLiw6YOo7IJIbUgmBfZ3T/Oi2IRrfDUxM2q4yGutGfQdEckMqIbAUWBQtLwL+kDR+c3SWYC7QmLTbICI5ZkCXHDezZ4GLgTIz2wX8CLgfeN7MbgHqgBuizZcBVwPbgWMkuhRLmjU1NbFmzRr27NnD1KlTmT17NgUFBRw/fpyioiKGDh3abXt3Z+fOnbS2tjJ58mQKCwuzVHnqWlpaiMfjzJ49m2HDhmW7nLxnid357KqurvZ4PJ7tMvLC0aNHefHFF4nH4yxYsIDJkyfzwQcf8Mc//pEZM2awZ88errnmGoqKiro97uDBg2zbto2qqioqKir6efb84O7s2rWL5cuXs3DhQmKxWLZLygtmtsbdq3uOp9R8RE6v+vp67r//fs477zzuvfdehg0bRkFBAaNHj2bo0KHcdtttfPOb3+z1V76trY36+nrMjJKSkq7xqv+soq4xP7vJTy6ZzJ2xO4nH45x//vnZLievKQTyxMGDB7n11luZNm0aX/va1/jEJz7RtS4Wi1FeXs4ll1zCtGnTKCjofqinqamJ+vp6Kisru80Q6hrr8B+dvpng3r17WbduHfPmzUt5Gm8/Nm6+42YWLVrE888/32v3RwZO/+9AHujo6ODxxx/njTfe4Cc/+Um3AOgUi8WYNWsWEydO7LWuvr6euro6xo4di5l1W9fW1sbH7RK2t7fT3t7e7/q2tjY6OjoG/D5aWlo+9vX64u4cP368Vx0lJSWMGDGCnTt3ntLzSXcKgTzQ0NDA008/zV133dXvX9CioiLmzZtHaWlpn49vbGxk+PDhQOKX6tChQwA89NBDrFixghMnTgCwbNkyXnvtNVpbWzl8+DDxeJw777yTbdu24e60tbVx4403smPHDjo6Orjrrrv41re+lZH3ffToUV566SWeeeYZ1q5dyy9/+Uuam5u7bVNdXU1NTU1GXj8UCoE8sGfPHmpqarj++uv73aagoIDi4uJeuwIdHR00NjYCdB0raG1t7RoD2Lp1a9df2VdeeQUzw8xYs2YNn/zkJzl69GjXL1p9fT0rVqzgjDPOwMy49tpruz3XyfScifTn+PHjPProoxw+fJjrrruOCy64gJEjR/aaRUyYMIEdO3YM+PWlNx0TyAM7d+6krKys11/55uZmfvGLXwAwZMgQysvLufzyyxk79qNG0a2trRw6dIji4uKuo+ixWKzrDMHmzZv56le/ytChQzly5AgHDx5k3LhxFBQUMGvWLI4dO8aOHTv4whe+gJlRU1PDZz7zGYYPH46ZMXPmTC688MJeNTc3N/P666+zadOmrrHdu3dTW1tLTU0NI0aMAGDs2LFceOGFjB8/vluAbd26lZUrV3LuueeyefNmtm/fzrRp03od9CwrK9NMIEUKgTxQWlrK8OHDe/2VLyoq4rbbbuOBBx7g2WefZdWqVYwcObLX4zs6Orrtt8disa5AcHemTp1KYWEhf/3rXxk+fDhnnHEGBQUFlJaWsmLFCs4++2yGDRuGu7Nq1Souu+yyrr/oe/fu7XOGMmTIEGbNmsXUqVO7xurq6njzzTe54oorus5SFBUVMWbMmF4zhHg8zrhx46ioqKCyspIZM2b0eyqwuLh4IP8ZpR8KgTxQWVmJu3PkyJFev+RtbW00NDRw8cUX9xkAQ4YMoaSkhAMHDtDe3t4rSKqqqhg1ahQAS5cuZdq0aTQ1NVFcXMyQIUPYsmULc+bMwcxobm6mtraW6667DkiES11dHV/84hd7vW4sFqOsrIyysrKusaKiIhobG5k+fXrX8Yn+dHR08KlPfYqzzz67a9v29nba2tq6neE4cOBAt5mPnDodE8gD48eP5/rrr+ehhx6itbUVd+86SPfee++xfv16Lr300j4fG4vFuv6y9zyoBolZRiwW49ChQ7zxxhtMmTKFXbt20dzcjLtjZowePRp3p6mpiR07dhCLxXB3Dhw4wOTJkzPynmfOnElBQQHt7e24O62traxfv77X2YUtW7Zw5plnZqSGUCgE8kBxcTH33XcfI0eO5Ctf+QqrV69m69at/Pa3v2XDhg1873vfY968ef0+vqKigtLSUo4cOdJr3YEDB3j55ZdZv349N9xwA1u3bqWlpYWhQ4cSi8W48soricfjrFu3jj/96U8sXLiQhx9+mJUrV7J27VoqKysz8p5nzJhBSUkJK1euZPXq1bz44ouUl5d3HYuAxK7MunXrOOusszJSQyj0teE8c/jwYWpra3F3pk2bdtJpNSS+LPTWW29RWlrKeeed17VLYD82mu9ppqWlhaKiIgoLCzlx4gSFhYXd9r/b2to4dOgQJSUlFBYW0tLSwrFjxxg1alSv3YuP09DQwMaNG5k7d26f33XoqXP2EYvFKC4u7nbcwH5svH/L+/zwhz/kV7/61YDPOoRMXxv+OzFq1Chmzpx5So8pLi5m/PjxfPjhh5w4caLruwaTSyZT9G9FJ3l0BryW+lNMLpnMkiVLuPfeexUAKVIIBCAWizF+/HgaGxvZv38/kyZNwsyovaM226UNSuduwJEjR5gyZUq2y8l7OiYQiBEjRjBlyhSamppoamrKdjkpOXr0KB0dHVxwwQWaBaSBZgIB6XnKLl+NHDmS6upeu7YySJoJiATupCHQT+OR/zCzzVFzkd+b2ehovMrMjpvZ29HPo5ksXkRSN5CZwK/p3XjkVeA8d58JbAXuSVr3rrvPjn5uTU+ZIpIpJw2BvhqPuPsr7t4W3V1F4orCIpKH0nFM4JvAS0n3p5jZOjP7i5n1/t/LIuo7IJIbUgoBM/sB0AY8HQ3VA5Pc/dPAPwHPmNmovh6rvgMiuWHQIWBmXwf+Afhq1IEId2929wPR8hrgXUBf7BbJYYMKATO7EvgX4EvufixpvNzMYtHyVBKdiXXZF5EcdtIvC/XTeOQeoAh4NfrG1qroTMBFwL+aWSvQAdzq7j27GYtIDjlpCLj7gj6Gn+hn2xeAF1ItSkROH31jUCRwCgGRwCkERAKnEBAJnEJAJHAKAZHAKQREAqcQEAmcQkAkcAoBkcApBEQCpxAQCZxCQCRwCgGRwCkERAI32L4D95nZ7qT+AlcnrbvHzLab2RYzuyJThYtIegy27wDAz5P6CywDMLNzgJuAc6PH/LLzcmMikpsG1XfgY8wHnosuOPoesB34bAr1iUiGpXJM4PaoDdmTZjYmGpsAvJ+0za5orBf1HRDJDYMNgUeAM4HZJHoNPHiqT6C+AyK5YVAh4O773L3d3TuAx/loyr8bmJi0aWU0JiI5arB9B8Yl3b0G6DxzsBS4ycyKzGwKib4Dq1MrUUQyabB9By42s9mAA7XAtwHcfaOZPQ/UkGhP9h13b89M6SKSDhZ1EMuq6upqj8fj2S5D5O+ama1x9+qe4/rGoEjgFAIigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiARusH0HfpPUc6DWzN6OxqvM7HjSukczWbyIpO6kVxYi0Xfgv4CnOgfc/cbOZTN7EGhM2v5dd5+drgJFJLNOGgLu/pqZVfW1zswMuAH4QnrLEpHTJdVjAhcC+9x9W9LYFDNbZ2Z/MbMLU3x+EcmwgewOfJwFwLNJ9+uBSe5+wMw+A/yvmZ3r7od7PtDMFgOLASZNmpRiGSIyWIOeCZjZEOBa4DedY1H7sQPR8hrgXeCsvh6v5iMiuSGV3YEvApvdfVfngJmVdzYgNbOpJPoO7EitRBHJpIGcInwWWAmcbWa7zOyWaNVNdN8VALgI2BCdMvwtcKu7D7SZqYhkwUDODizoZ/zrfYy9ALyQelkicrroG4MigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASuIFcVGSimf3ZzGrMbKOZfTcaLzWzV81sW3Q7Jho3M3vYzLab2QYzm5PpNyEigzeQmUAbcKe7nwPMBb5jZucAdwPL3X06sDy6D3AVicuKTSdxIdFH0l61iKTNSUPA3evdfW20fATYBEwA5gNLos2WAF+OlucDT3nCKmC0mY1Le+UikhandEwgakLyaeBNoMLd66NVe4GKaHkC8H7Sw3ZFYyKSgwYcAmY2gsT1A+/o2UfA3R3wU3lhM1tsZnEzi+/fv/9UHioiaTSgEDCzQhIB8LS7/y4a3tc5zY9uG6Lx3cDEpIdXRmPdqO+ASG4YyNkBA54ANrn7z5JWLQUWRcuLgD8kjd8cnSWYCzQm7TaISI4ZSBuyzwELgb91tiAHvg/cDzwf9SGoI9GYFGAZcDWwHTgGfCOtFYtIWg2k78DrgPWz+tI+tnfgOynWJSKnib4xKBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgLHE1sCwXYbYfaAI+yHYtKSgjv+uH/H8P+V4/ZPY9THb3Xpf2zokQADCzuLtXZ7uOwcr3+iH/30O+1w/ZeQ/aHRAJnEJAJHC5FAKPZbuAFOV7/ZD/7yHf64csvIecOSYgItmRSzMBEcmCrIeAmV1pZlvMbLuZ3Z3tegbKzGrN7G9m9raZxaOxUjN71cy2Rbdjsl1nMjN70swazOydpLE+a456ST4cfS4bzGxO9irvqrWv+u8zs93R5/C2mV2dtO6eqP4tZnZFdqr+iJlNNLM/m1mNmW00s+9G49n9DNw9az9ADHgXmAoMBdYD52SzplOovRYo6zH2U+DuaPlu4N+zXWeP+i4C5gDvnKxmEv0kXyLRgm4u8GaO1n8f8M99bHtO9O+pCJgS/TuLZbn+ccCcaHkksDWqM6ufQbZnAp8Ftrv7DndvAZ4D5me5plTMB5ZEy0uAL2exll7c/TXgYI/h/mqeDzzlCauA0Z2t6LOln/r7Mx94zt2b3f09Eg1yP5ux4gbA3evdfW20fATYBEwgy59BtkNgAvB+0v1d0Vg+cOAVM1tjZoujsQr/qA37XqAiO6Wdkv5qzqfP5vZouvxk0i5YTtdvZlXAp4E3yfJnkO0QyGefd/c5wFXAd8zsouSVnpjP5dWpl3ysGXgEOBOYDdQDD2a3nJMzsxHAC8Ad7n44eV02PoNsh8BuYGLS/cpoLOe5++7otgH4PYmp5r7O6Vp025C9Cgesv5rz4rNx933u3u7uHcDjfDTlz8n6zayQRAA87e6/i4az+hlkOwTeAqab2RQzGwrcBCzNck0nZWbDzWxk5zJwOfAOidoXRZstAv6QnQpPSX81LwVujo5QzwUak6asOaPHPvI1JD4HSNR/k5kVmdkUYDqw+nTXl8zMDHgC2OTuP0tald3PIJtHS5OOgG4lcfT2B9muZ4A1TyVx5Hk9sLGzbmAssBzYBvwRKM12rT3qfpbElLmVxP7lLf3VTOKI9C+iz+VvQHWO1v/fUX0bol+acUnb/yCqfwtwVQ7U/3kSU/0NwNvRz9XZ/gz0jUGRwGV7d0BEskwhIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigft/Nj0KaCkVVV0AAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: add\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: leaf\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: leaf\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAD8CAYAAAB3lxGOAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAPxElEQVR4nO3dbYwd5XnG8f+FMUi81Jh4ayFjew1yEFC1xl1Ri/CSlpKAVWWhH6gtBE6KapCMBGqqykCUon5K0wASaksEwoqpCC+pTfEHp8W1TFBQDazBMRhjsIkte2Vsh1RgQgje9d0P8ywM693s2TPn7JzNc/2kozPnmTk799HApXnz3IoIzCxfJ9RdgJnVyyFgljmHgFnmHAJmmXMImGXOIWCWubaFgKSrJe2UtEvSynatx8yqUTvuE5A0BXgLuArYD7wMLI2IN1q+MjOrpF17AhcDuyLinYj4BHgC6G3TusysghPb9HdnAftKn/cDfzLawjNmzIju7u42lWJmAFu2bPlFRHQNH29XCIxJ0nJgOcCcOXPo6+urqxSzLEjaO9J4uw4H+oHZpc9np7FPRcRDEdETET1dXceFk5lNkHaFwMvAfEnzJJ0ELAHWtWldZlZBWw4HImJA0m3AfwNTgFURsb0d6zKzatp2TiAi1gPr2/X3zaw1fMegWeYcAmaZcwiYZc4hYJY5h4BZ5hwCZplzCJhlziFgljmHgFnmHAJmmXMImGXOIWCWOYeAWeYcAmaZcwiYZa7pEJA0W9ImSW9I2i7p9jR+j6R+SVvTa3HryjWzVqvyUJEB4JsR8Yqk04EtkjakefdHxPeql2dm7dZ0CETEAeBAmj4iaQfFo8bNbBJpyTkBSd3ARcCLaeg2SdskrZI0vRXrMLP2qBwCkk4D1gB3RMQHwIPAucACij2Fe0f53nJJfZL6Dh8+XLUMM2tSpRCQNJUiAB6LiLUAEXEwIgYj4hjwMEVLsuO474BZZ6hydUDAI8COiLivNH5WabHrgNebL8/M2q3K1YEvATcCr0namsbuApZKWgAEsAe4pVKFZtZWVa4O/BTQCLPca8BsEvEdg2aZcwiYZc4hYJY5h4BZ5hwCZplzCJhlziFgljmHgFnmHAJmmXMImGXOIWCWOYeAWeYcAmaZcwiYZc4hYJY5h4BZ5qo8WQgASXuAI8AgMBARPZLOBJ4EuimeLnR9RPxf1XWZWeu1ak/gTyNiQUT0pM8rgY0RMR/YmD6bWQdq1+FAL7A6Ta8Grm3TesysolaEQADPStoiaXkam5k6FAG8C8wc/iX3HTDrDJXPCQCXRkS/pN8HNkh6szwzIkJSDP9SRDwEPATQ09Nz3HwzmxiV9wQioj+9HwKepmg2cnCo/0B6P1R1PWbWHlU7EJ2aOhIj6VTgKxTNRtYBy9Jiy4BnqqzHzNqn6uHATODpohkRJwI/jIj/kvQy8JSkm4G9wPUV12NmbVIpBCLiHeCPRhh/D7iyyt82s4nhOwbNMucQMMucQ8Ascw4Bs8w5BMwy5xAwy5xDwCxzDgGzzDkEzDLnEDDLnEPALHMOAbPMOQTMMucQMMucQ8Asc00/T0DSeRS9BYacA3wbOAP4G2Do6aF3RcT6pis0s7ZqOgQiYiewAEDSFKCf4hmD3wDuj4jvtaRCM2urVh0OXAnsjoi9Lfp7ZjZBWhUCS4DHS59vk7RN0ipJ01u0DjNrg8ohIOkk4GvAj9LQg8C5FIcKB4B7R/mem4+YdYBW7AlcA7wSEQcBIuJgRAxGxDHgYYo+BMeJiIcioicierq6ulpQhpk1oxUhsJTSocBQ05HkOoo+BGbWoSo9cjw1HLkKuKU0/F1JCyh6FO4ZNs/MOkzVvgO/Ar4wbOzGShWZ2YTyHYNmmXMImGXOIWCWOYeAWeYcAmaZcwiYZc4hYJY5h4BZ5hwCZplzCJhlziFgljmHgFnmHAJmmXMImGXOIWCWuYZCID0w9JCk10tjZ0raIOnt9D49jUvSA5J2pYeNLmxX8WZWXaN7Aj8Arh42thLYGBHzgY3pMxTPHJyfXsspHjxqZh2qoRCIiOeBXw4b7gVWp+nVwLWl8UejsBk4Y9hzB82sg1Q5JzAzIg6k6XeBmWl6FrCvtNz+NGZmHaglJwYjIigeLNow9x0w6wxVQuDg0G5+ej+UxvuB2aXlzk5jn+O+A2adoUoIrAOWpellwDOl8ZvSVYJFwPulwwYz6zANPXJc0uPAl4EZkvYD/wB8B3hK0s3AXuD6tPh6YDGwC/iIokuxmXWohkIgIpaOMuvKEZYNYEWVosxs4viOQbPMOQTMMucQsKYdO3aMDz/8kMHBwbpLsQoq9SK0/EQEc++fy74j+46bN3faXPbcsWfii7JKvCdg43LkyBH2HdnH6nmrOfbtYxy9+yjfGvgWn9z1CXvf31t3edYE7wnYuKxfvx6A7u5uXnjhBV566SWuuOIKpk6dWnNl1iyHgI3L2rVr4UKYMWMGXV1dXHLJJZxwgncoJzOHgI3L0aNHATj//PORBMDAwAADAwN1lmUVOMJtXK666ioABgcHiQg+/vhjnnvuOV8hmMQcAjYuN9xwA1CcG9i0aROrVq3ivPPO45RTTqm5MmuWDwdsXKZNm8bcaXPpfbX307EVq4q7xOdOm1tXWVaBQ8DGzfcC/G7x4YBZ5hwCZplzCJhlziFglrkxQ2CUxiP/LOnN1FzkaUlnpPFuSb+WtDW9vt/O4s2sukb2BH7A8Y1HNgB/EBF/CLwF3FmatzsiFqTXra0p08zaZcwQGKnxSEQ8GxFD94lupniisJlNQq04J/DXwI9Ln+dJelXSTyRdNtqX3HfArDNUCgFJdwMDwGNp6AAwJyIuAv4W+KGk3xvpu+47YNYZmg4BSV8H/gK4IT1hmIj4TUS8l6a3ALuBL7agTjNrk6ZCQNLVwN8DX4uIj0rjXZKmpOlzKDoTv9OKQs2sPcb8twOjNB65EzgZ2JD+TfnmdCXgcuAfJR0FjgG3RsTwbsZm1kHGDIFRGo88Msqya4A1VYsys4njOwbNMucQMMucQ8Ascw4Bs8w5BMwy5xAwy5xDwCxzDgGzzDkEzDLnEDDLnEPALHMOAbPMOQTMMucQMMucQ8Asc832HbhHUn+pv8Di0rw7Je2StFPSV9tVuJm1RrN9BwDuL/UXWA8g6QJgCXBh+s6/DT1uzMw6U1N9B36LXuCJ9MDRnwO7gIsr1GdmbVblnMBtqQ3ZKknT09gsYF9pmf1p7DjuO2DWGZoNgQeBc4EFFL0G7h3vH3DfAbPO0FQIRMTBiBiMiGPAw3y2y98PzC4tenYaM7MO1WzfgbNKH68Dhq4crAOWSDpZ0jyKvgMvVSvRzNqp2b4DX5a0AAhgD3ALQERsl/QU8AZFe7IVETHYntLNrBWUOojVqqenJ/r6+uouw+x3mqQtEdEzfNx3DJplziFgljmHgFnmHAJmmXMImGXOIWCWOYeAWeYcAmaZcwiYZc4hYJY5h4BZ5hwCZplzCJhlziFgljmHgFnmmu078GSp58AeSVvTeLekX5fmfb+dxZtZdWM+WYii78C/AI8ODUTEXw1NS7oXeL+0/O6IWNCqAs2svcYMgYh4XlL3SPMkCbge+LPWlmVmE6XqOYHLgIMR8XZpbJ6kVyX9RNJlFf++mbVZI4cDv81S4PHS5wPAnIh4T9IfA/8p6cKI+GD4FyUtB5YDzJkzp2IZZtaspvcEJJ0I/CXw5NBYaj/2XpreAuwGvjjS9918xKwzVDkc+HPgzYjYPzQgqWuoAamkcyj6DrxTrUQza6dGLhE+DvwvcJ6k/ZJuTrOW8PlDAYDLgW3pkuF/ALdGRKPNTM2sBo1cHVg6yvjXRxhbA6ypXpaZTRTfMWiWOYeAWeYcAmaZcwiYZc4hYJY5h4BZ5hwCZplzCJhlziFgljmHgFnmHAJmmXMImGXOIWCWOYeAWeYcAmaZa+ShIrMlbZL0hqTtkm5P42dK2iDp7fQ+PY1L0gOSdknaJmlhu3+EmTWvkT2BAeCbEXEBsAhYIekCYCWwMSLmAxvTZ4BrKB4rNp/iQaIPtrxqM2uZMUMgIg5ExCtp+giwA5gF9AKr02KrgWvTdC/waBQ2A2dIOqvllZtZS4zrnEBqQnIR8CIwMyIOpFnvAjPT9CxgX+lr+9OYmXWghkNA0mkUzw+8Y3gfgYgIIMazYknLJfVJ6jt8+PB4vmpmLdRQCEiaShEAj0XE2jR8cGg3P70fSuP9wOzS189OY5/jvgNmnaGRqwMCHgF2RMR9pVnrgGVpehnwTGn8pnSVYBHwfumwwcw6TCNtyL4E3Ai8NtSCHLgL+A7wVOpDsJeiMSnAemAxsAv4CPhGSys2s5ZqpO/ATwGNMvvKEZYPYEXFusxsgviOQbPMOQTMMucQMMucQ8Ascw4Bs8w5BMwy5xAwy5xDwCxzDgGzzDkEzDLnEDDLnEPALHMOAbPMOQTMMucQMMucQ8Ascw4Bs8w5BMwyp+JpYDUXIR0GfgX8ou5aKpjB5K4fJv9vmOz1Q3t/w9yIOO7R3h0RAgCS+iKip+46mjXZ64fJ/xsme/1Qz2/w4YBZ5hwCZpnrpBB4qO4CKprs9cPk/w2TvX6o4Td0zDkBM6tHJ+0JmFkNag8BSVdL2ilpl6SVddfTKEl7JL0maaukvjR2pqQNkt5O79PrrrNM0ipJhyS9XhobsebUS/KBtF22SVpYX+Wf1jpS/fdI6k/bYaukxaV5d6b6d0r6aj1Vf0bSbEmbJL0habuk29N4vdsgImp7AVOA3cA5wEnAz4AL6qxpHLXvAWYMG/susDJNrwT+qe46h9V3ObAQeH2smin6Sf6YogXdIuDFDq3/HuDvRlj2gvTf08nAvPTf2ZSa6z8LWJimTwfeSnXWug3q3hO4GNgVEe9ExCfAE0BvzTVV0QusTtOrgWtrrOU4EfE88Mthw6PV3As8GoXNwBlDrejrMkr9o+kFnoiI30TEzyka5F7ctuIaEBEHIuKVNH0E2AHMouZtUHcIzAL2lT7vT2OTQQDPStoiaXkamxmftWF/F5hZT2njMlrNk2nb3JZ2l1eVDsE6un5J3cBFwIvUvA3qDoHJ7NKIWAhcA6yQdHl5ZhT7c5Pq0stkrBl4EDgXWAAcAO6tt5yxSToNWAPcEREflOfVsQ3qDoF+YHbp89lprONFRH96PwQ8TbGreXBody29H6qvwoaNVvOk2DYRcTAiBiPiGPAwn+3yd2T9kqZSBMBjEbE2Dde6DeoOgZeB+ZLmSToJWAKsq7mmMUk6VdLpQ9PAV4DXKWpflhZbBjxTT4XjMlrN64Cb0hnqRcD7pV3WjjHsGPk6iu0ARf1LJJ0saR4wH3hpousrkyTgEWBHRNxXmlXvNqjzbGnpDOhbFGdv7667ngZrPofizPPPgO1DdQNfADYCbwP/A5xZd63D6n6cYpf5KMXx5c2j1UxxRvpf03Z5Dejp0Pr/PdW3Lf1Pc1Zp+btT/TuBazqg/kspdvW3AVvTa3Hd28B3DJplru7DATOrmUPALHMOAbPMOQTMMucQMMucQ8Ascw4Bs8w5BMwy9/9nsZayA1pmAAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: ffpar\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAD8CAYAAAB3lxGOAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAWKElEQVR4nO3de2xc9ZnG8e+b8SWN7RI7CUkanBskrZLAmuByaQntwqaFaNuE/YMlWrXQRaQVoBYtqxW00i5FqdTtlnbFwlIFgRoQEC4pBVV0IUXQNjShOJBAbpCrE1vGMYHGsePrzLt/zBkzviWOZyYz7u/5SNac+Z1zZt7JZB6fy/i85u6ISLjG5bsAEckvhYBI4BQCIoFTCIgETiEgEjiFgEjgchYCZna1mb1nZnvN7M5cPY+IZMZy8T0BM4sB7wNLgQbgTWClu+/M+pOJSEZytSVwMbDX3fe7ezewDlieo+cSkQwU5ehxZwCH0+43AJcMt/DkyZN99uzZOSpFRAC2bNnyobtPGTieqxA4JTNbBawCmDlzJnV1dfkqRSQIZlY/1Hiudgcageq0++dEY33cfY2717p77ZQpg8JJRM6QXIXAm8A8M5tjZiXA9cALOXouEclATnYH3L3XzG4DXgJiwCPuviMXzyUimcnZMQF3fxF4MVePLyLZoW8MigROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOBGHQJmVm1mr5rZTjPbYWbfi8bvNrNGM9sa/SzLXrkikm2ZXFSkF7jD3d8yswpgi5ltiOb93N1/mnl5IpJrow4Bd28CmqLp42a2i+SlxkVkDMnKMQEzmw1cCLwRDd1mZu+Y2SNmVpmN5xCR3Mg4BMysHFgP3O7urcCDwLlADckthXuHWW+VmdWZWV1LS0umZYjIKGUUAmZWTDIAHnf3XwG4e7O7x909ATxEsiXZIOo7IFIYMjk7YMDDwC53/1na+PS0xa4Fto++PBHJtUzODnwR+Abwrpltjca+D6w0sxrAgYPAtzOqUERyKpOzAxsBG2KWeg2IjCH6xqBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgErhMriwEgJkdBI4DcaDX3WvNrAp4CphN8upC17n7x5k+l4hkX7a2BP7W3WvcvTa6fyfwirvPA16J7otIAcrV7sByYG00vRZYkaPnEZEMZSMEHHjZzLaY2apobGrUoQjgA2DqwJXUd0CkMGR8TAC43N0bzexsYIOZ7U6f6e5uZj5wJXdfA6wBqK2tHTRfRM6MjLcE3L0xuj0CPEey2Uhzqv9AdHsk0+cRkdzItANRWdSRGDMrA75CstnIC8AN0WI3AM9n8jwikjuZ7g5MBZ5LNiOiCHjC3f/PzN4Enjazm4B64LoMn0dEciSjEHD3/cDfDDF+FLgqk8cWkTND3xgUCZxCQCRwCgGRwCkERAKnEBAJnEJAJHAKAZHAKQREAqcQEAmcQkAkcAoBkcApBEQCpxAQCZxCQCRwCgGRwI36egJm9lmSvQVS5gL/DkwEbgZSVw/9vru/OOoKRSSnRh0C7v4eUANgZjGgkeQ1Br8F/Nzdf5qVCkUkp7K1O3AVsM/d67P0eCJyhmQrBK4Hnky7f5uZvWNmj5hZZZaeQ0RyIOMQMLMS4OvAM9HQg8C5JHcVmoB7h1lPzUdECkA2tgSuAd5y92YAd29297i7J4CHSPYhGMTd17h7rbvXTpkyJQtliMhoZCMEVpK2K5BqOhK5lmQfAhEpUBldcjxqOLIU+Hba8E/MrIZkj8KDA+aJSIHJtO9AOzBpwNg3MqpIRM4ofWNQJHAKAZHAKQREAqcQEAmcQkAkcAoBkcApBEQCpxAQCZxCQCRwCgGRwCkERAKnEBAJnEJAJHAKAZHAKQREAjeiEIguGHrEzLanjVWZ2QYz2xPdVkbjZmb3mdne6GKji3NVvIhkbqRbAr8Erh4wdifwirvPA16J7kPymoPzop9VJC88KiIFakQh4O5/AD4aMLwcWBtNrwVWpI0/6kmbgYkDrjsoIgUkk2MCU929KZr+AJgaTc8ADqct1xCNiUgBysqBQXd3khcWHTH1HRApDJmEQHNqMz+6PRKNNwLVacudE431o74DIoUhk6sNvwDcAPw4un0+bfw2M1sHXAIcS9ttkALQ29tLT08PAJs2bWLjxo10d3dTWVnJhAkTMLNB6zQ3N9PT00NxcTGrVq1i+vShD/O4Oz09PcTjcT71qU+dVl2dnZ0AlJSUMG7cJ7+furu7SSQSlJaWDlmbZMaSW/KnWMjsSeDLwGSgGfgP4NfA08BMoB64zt0/suS7dD/JswkngG+5e93JHr+2ttbr6k66iAzQ0dFBIpEYcp6ZUVRURElJyaB58Xic/fv3U1payvTp0znv/vM41Hoo1+WOyqyzZnHw9oPE43FeffVVLr30UsrKyhQEo2RmW9y9duD4iLYE3H3lMLOuGmJZB249vfLkdMTjcTZu3MgzzzzDoUOHWLVqFZWVlfT29tLY2Mhbb73F/PnzufnmmyktLe237tGjRzl27BjnnXceRUVFHGo9xO8u/x0333wza9as4Utf+hLFxcWDnvPKK6/k85//PPfcc8+gx0xJJBLs2bOHlpYWLrvsMmKx2LCvob6+nscee4wrrriCL3zhCxQVJf8rtra2UldXR01NDZP+J9nSIhaLcckll/DAAw9wyy23UFFRMdp/OhlCRs1HJD9isRhLlizhtddeY+rUqVxzzTWUlJTg7rg7S5cu5e233+77YKX09vayf/9+JkyYQHl5ed9v1OLiYszspL9hzYySkpKTLtPe3s66deu44447ThoAHR0d7Nixgx07dnD++eeTvjVaUVFBZWUlBw4c6LdOeXk5S5cu5aWXXmLFihWDXpuMnr42PEaZGdu2baO2tpbx48cTi8UoKiqiuLiYT3/608yaNWvQB/Hw4cM0NDQwefLkIT9EI9k1HE48Hmfz5s3MmzePsrKyYZdzdxoaGmhqamL27Nm0trb2e14zY/r06TQ0NPSrycy44IIL2LVr16B1JDMKgTGqs7OT+vp65s6di5nR1dXVd/Bu/PjxzJo1a9A6dXV1HDlyhMrKyqzXE4/Heeyxx1iyZMlJtxY+/vhjOjs7Wbp0KWVlZXR2dg76QJeXl3P06FGgfzAVFRUxdepUDh48mPX6Q6YQGIMSiQR/+tOfOOusswBoaWlh165d1NfXE4vF+rYG0nV1dXHo0CHKy8uH3afPRHt7O0eOHBn2rAEkj/IfP36ciRMnUlVVRXFxMd3d3YNCoKSkhHg8DgzeOpk1axYNDQ3aEsgihcAYlEgkeOaZZ5g8eTInTpxg586drF+/nuLi4n6n1tL19PSQSCSYPHnysMuMlrvT2tpKZWXlsPvq7k5HRwddXV1MmjSJcePGMWHCBDo7O+nt7e23bPqZjVQYpEycOJHGxsZB4zJ6OroyBvX29vL888/zox/9iNraWnp6epg4cSKf+cxnhl2np6cHd2fChAk5qam9vf2kxwJSByU3bNjAggULaGtrY+vWrZx99tl0dXVRXl7et+y4ceP6Hmvgb/yKigra29u1JZBFCoExxt05cOAA3d3dXH755X0H/6qrq/t2D4YTj8dz9hvUzAb9Rk9xd/7yl7/w4YcfsnLlSoqLi+ns7KSpqYmmpiZOnDjBpEmT+i2fqnPg8YXUl5D0XYHsUQiMMe7O66+/TllZGXPnzgWSpwyrqqpOul7qNGBbW9uwy2TywSorK+PYsWO4+6DH6ejo4NChQ5x//vlMmzYNSG6ZLFq0iMbGxr5vCqYkEgna29sBBu26tLW1UVVVlfVdmpDpX3IMcXe6u7t5/fXXueyyyxg/fvyIN4vHjx9PeXk5zc3NJ/2NPRpmRlVVFe3t7Zw4caLfvM7OTnbs2EFzczNnn312v3VKSkpobW0ddMovdfwitVy6pqYmpk2bpi2BLFIIjCE7d+7ku9/9Llu3bqWhoYEf/vCH7N27d0TrFhUVsWjRImKx2KAPakomH6zS0lIWL17Mvn37+sba2tpYvXo1N910E0888UTfF4ASiQTbtm3jvvvuY9OmTdx///1s27at7+8Zurq6+o5dDKypoaGBc845RyGQRdodGEMWLlzImjVrRr1+TU0NbW1ttLS0UFFRMeiDdLItgVNtJcRiMW688UaeffZZFi5cSCwWo7y8nNWrV7N69ep+y44bN46LLrqI9evXD/lYH330UXKrYU//3YG2tjaOHz/OjBkzFAJZpC2BgJx11lnMmTOHY8eO0dXV1Tfe1dVFIpEY8o+SUgfpurq66OzsHPaPlsaNG8ecOXOYNm0aTU1No961iMfjHD58eNCXnRKJBE899RRf+9rXGD9+/KgeW4amEAiImTFz5sy+o/WpYwOvvfYaXV1dvPzyy3zwwQf9jhm0t7ezadMm9u/fzxtvvMHu3bvp7u4e8vFLSkpYsWIFu3bt6hcyI+Xu7N69m+LiYmbMmNFvfM+ePUyZMoXPfe5zOiiYZdodCExFRQXV1dX09PT0fVArKiq45ZZbANi2bRtTpkwhFothZmzfvp0//vGP3Hpr8g9Df/Ob3zBp0iSqq6sHPbaZUVlZyfz582ltbT3t39gdHR10d3ezaNGift9n6OrqoqOjg2XLlukPh3JgRNcTyDVdTyB/Zv/3bOqP1ee7jCGlricg2ZHR9QTkr5c+ZHLKnathGo/8l5ntjpqLPGdmE6Px2WbWYWZbo59f5LJ4EcncSI6w/JLBjUc2AIvc/QLgfeCutHn73L0m+vlOdsoUkVw5ZQgM1XjE3V9299Qh5M0krygsImNQNs61/DPw27T7c8zsbTP7vZktGW4l9R0QKQwZhYCZ/QDoBR6PhpqAme5+IfAvwBNm9umh1lXfAZHCMOoQMLMbgb8H/im6wjDu3uXuR6PpLcA+YH4W6hSRHBlVCJjZ1cC/AV939xNp41PMLBZNzyXZmXh/NgoVkdw45fcE0huPmFkDycYjdwGlwIboDzk2R2cCrgDuMbMeIAF8x90HdjMWkQJyyhAYpvHIw8Msux4Y+k/DRKQg6S8xRAKnEBAJnEJAJHAKAZHAKQREAqcQEAmcQkAkcAoBkcApBEQCpxAQCZxCQCRwCgGRwCkERAKnEBAJnEJAJHCj7Ttwt5k1pvUXWJY27y4z22tm75nZV3NVuIhkx2j7DgD8PK2/wIsAZrYAuB5YGK3zv6nLjYlIYRpV34GTWA6siy44egDYC1ycQX0ikmOZHBO4LWpD9oiZVUZjM4DDacs0RGODqO+ASGEYbQg8CJwL1JDsNXDv6T6A+g6IFIZRhYC7N7t73N0TwEN8ssnfCKQ3rj8nGhORAjXavgPT0+5eC6TOHLwAXG9mpWY2h2TfgT9nVqKI5NJo+w582cxqAAcOAt8GcPcdZvY0sJNke7Jb3T2em9JFJBss6iCWV7W1tV5XV5fvMkT+qpnZFnevHTiubwyKBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOAUAiKBUwiIBE4hIBI4hYBI4Ebbd+CptJ4DB81sazQ+28w60ub9IpfFi0jmTnllIZJ9B+4HHk0NuPs/pqbN7F7gWNry+9y9JlsFikhunTIE3P0PZjZ7qHlmZsB1wJXZLUtEzpRMjwksAZrdfU/a2Bwze9vMfm9mSzJ8fBHJsZHsDpzMSuDJtPtNwEx3P2pmFwG/NrOF7t46cEUzWwWsApg5c2aGZYjIaI16S8DMioB/AJ5KjUXtx45G01uAfcD8odZX8xGRwpDJ7sDfAbvdvSE1YGZTUg1IzWwuyb4D+zMrUURyaSSnCJ8ENgGfNbMGM7spmnU9/XcFAK4A3olOGT4LfMfdR9rMVETyYCRnB1YOM37jEGPrgfWZlyUiZ4q+MSgSOIWASOAUAiKBUwiIBE4hIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigVMIiAROISASOIWASOAUAiKBG8lFRarN7FUz22lmO8zse9F4lZltMLM90W1lNG5mdp+Z7TWzd8xsca5fhIiM3ki2BHqBO9x9AXApcKuZLQDuBF5x93nAK9F9gGtIXlZsHskLiT6Y9apFJGtOGQLu3uTub0XTx4FdwAxgObA2WmwtsCKaXg486kmbgYlmNj3rlYtIVpzWMYGoCcmFwBvAVHdvimZ9AEyNpmcAh9NWa4jGRKQAjTgEzKyc5PUDbx/YR8DdHfDTeWIzW2VmdWZW19LScjqrikgWjSgEzKyYZAA87u6/ioabU5v50e2RaLwRqE5b/ZxorB/1HRApDCM5O2DAw8Aud/9Z2qwXgBui6RuA59PGvxmdJbgUOJa22yAiBWYkbci+CHwDeDfVghz4PvBj4OmoD0E9ycakAC8Cy4C9wAngW1mtWESyaiR9BzYCNszsq4ZY3oFbM6xLRM4QfWNQJHAKAZHAKQREAqcQEAmcQkAkcAoBkcApBEQCpxAQCZxCQCRwCgGRwCkERAKnEBAJnEJAJHAKAZHAKQREAqcQEAmcQkAkcAoBkcBZ8mpgeS7CrAVoBz7Mdy0ZmMzYrh/G/msY6/VDbl/DLHcfdGnvgggBADOrc/fafNcxWmO9fhj7r2Gs1w/5eQ3aHRAJnEJAJHCFFAJr8l1AhsZ6/TD2X8NYrx/y8BoK5piAiORHIW0JiEge5D0EzOxqM3vPzPaa2Z35rmekzOygmb1rZlvNrC4aqzKzDWa2J7qtzHed6czsETM7Ymbb08aGrDnqJXlf9L68Y2aL81d5X61D1X+3mTVG78NWM1uWNu+uqP73zOyr+an6E2ZWbWavmtlOM9thZt+LxvP7Hrh73n6AGLAPmAuUANuABfms6TRqPwhMHjD2E+DOaPpO4D/zXeeA+q4AFgPbT1UzyX6SvyXZgu5S4I0Crf9u4F+HWHZB9P+pFJgT/T+L5bn+6cDiaLoCeD+qM6/vQb63BC4G9rr7fnfvBtYBy/NcUyaWA2uj6bXAijzWMoi7/wH4aMDwcDUvBx71pM3AxFQr+nwZpv7hLAfWuXuXux8g2SD34pwVNwLu3uTub0XTx4FdwAzy/B7kOwRmAIfT7jdEY2OBAy+b2RYzWxWNTfVP2rB/AEzNT2mnZbiax9J7c1u0ufxI2i5YQddvZrOBC4E3yPN7kO8QGMsud/fFwDXArWZ2RfpMT27PjalTL2OxZuBB4FygBmgC7s1vOadmZuXAeuB2d29Nn5eP9yDfIdAIVKfdPycaK3ju3hjdHgGeI7mp2ZzaXItuj+SvwhEbruYx8d64e7O7x909ATzEJ5v8BVm/mRWTDIDH3f1X0XBe34N8h8CbwDwzm2NmJcD1wAt5rumUzKzMzCpS08BXgO0ka78hWuwG4Pn8VHhahqv5BeCb0RHqS4FjaZusBWPAPvK1JN8HSNZ/vZmVmtkcYB7w5zNdXzozM+BhYJe7/yxtVn7fg3weLU07Avo+yaO3P8h3PSOseS7JI8/bgB2puoFJwCvAHuB3QFW+ax1Q95MkN5l7SO5f3jRczSSPSD8QvS/vArUFWv9jUX3vRB+a6WnL/yCq/z3gmgKo/3KSm/rvAFujn2X5fg/0jUGRwOV7d0BE8kwhIBI4hYBI4BQCIoFTCIgETiEgEjiFgEjgFAIigft/tRJ0GAEs5vsAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: pi\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "pred: leaf\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'\\\\left( G \\\\left( X O \\\\right) \\\\cdot \\\\delta \\\\beta \\\\right) \\\\star \\\\frac { G \\\\left( u w + e \\\\right) } { F \\\\left( \\\\prod \\\\left( \\\\frac { \\\\sqrt { y } } { x } \\\\right) \\\\right) }'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 49
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "11tTHk5Ioavn",
        "outputId": "bafdee5f-1cc7-4a90-a87d-93a1c1b1a8ca",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!git clone https://github.com/kakainet/im2latex-1.git\n",
        "!pip install -q distance"
      ],
      "execution_count": 50,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Cloning into 'im2latex-1'...\n",
            "remote: Enumerating objects: 46, done.\u001b[K\n",
            "remote: Counting objects: 100% (46/46), done.\u001b[K\n",
            "remote: Compressing objects: 100% (37/37), done.\u001b[K\n",
            "remote: Total 84 (delta 23), reused 20 (delta 9), pack-reused 38\u001b[K\n",
            "Unpacking objects: 100% (84/84), done.\n",
            "\u001b[K     |████████████████████████████████| 184kB 8.0MB/s \n",
            "\u001b[?25h  Building wheel for distance (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "di7ivBVCogGx"
      },
      "source": [
        "import subprocess\n",
        "\n",
        "preds_path = test_set / 'preds.txt'\n",
        "predicted_path = test_set / 'test.txt'\n",
        "\n",
        "def bleu_and_edit_distance(predictions, predicted_path, target_path):\n",
        "    with open(predicted_path, 'w+') as f:\n",
        "        for pred in predictions:\n",
        "            f.write(pred+'\\n')\n",
        "    output = subprocess.check_output('python im2latex-1/src/Evaluation/bleu_score.py --target-formulas {} --predicted-formulas {} --ngram 5'.\n",
        "                format(target_path, predicted_path), shell=True)\n",
        "    output = str(output)\n",
        "    bleu_message = output[output.find('BLEU'):-3]\n",
        "    output = subprocess.check_output('python im2latex-1/src/Evaluation/edit_distance.py --target-formulas {} --predicted-formulas {}'.\n",
        "                format(target_path, predicted_path), shell=True)\n",
        "    output = str(output)\n",
        "    edit_message = output[output.find('Edit'):-3]\n",
        "    return bleu_message, edit_message"
      ],
      "execution_count": 51,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cpAkvOseszZs"
      },
      "source": [
        "from tqdm import tqdm\n",
        "\n",
        "def get_results(leaf_threshold):\n",
        "    infer = Inference(fastai_model, seq2seq_wrap, leaf_threshold=leaf_threshold)\n",
        "\n",
        "    predictions = [infer(img_from_path(img_path)) for img_path in tqdm(test_images)]\n",
        "\n",
        "    return bleu_and_edit_distance(predictions, str(preds_path), str(predicted_path))"
      ],
      "execution_count": 52,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0d_Bx_pf7-BC"
      },
      "source": [
        "# Results"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SpNHDjC-tSA8",
        "outputId": "7e8ec354-feed-4054-d3bf-17cef89a6d9c",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "# Vanilla seq2seq model\n",
        "get_results(leaf_threshold=0)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 1000/1000 [16:52<00:00,  1.01s/it]\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "('BLEU Score: 0.812304', 'Edit Distance Accuracy: 0.781447')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 47
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CHoW11Q1uAsq",
        "outputId": "204ac164-4475-4262-bf7a-677b2f545753",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "# One prediction of our CNN model, then seq2seq model\n",
        "get_results(leaf_threshold=1)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 1000/1000 [25:35<00:00,  1.54s/it]\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "('BLEU Score: 0.891221', 'Edit Distance Accuracy: 0.854896')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zb-MBDiauJC5",
        "outputId": "58be9b30-a4fe-47df-d24c-901a5a8f9f74",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "get_results(leaf_threshold=2)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 1000/1000 [30:55<00:00,  1.86s/it]\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "('BLEU Score: 0.871733', 'Edit Distance Accuracy: 0.824075')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 49
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GFZOr402nNbi",
        "outputId": "8f9f56bc-521a-47cc-d937-7749855c2956",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "get_results(leaf_threshold=3)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 1000/1000 [40:18<00:00,  2.42s/it]\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "('BLEU Score: 0.832564', 'Edit Distance Accuracy: 0.755858')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 50
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zt6wmfleuUMf",
        "outputId": "61d89bec-377b-48f0-f482-b5152101ad5f",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "# CNN performs all predictions - seq2seq used only for leaves\n",
        "get_results(leaf_threshold=4)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 1000/1000 [44:22<00:00,  2.66s/it]\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "('BLEU Score: 0.802975', 'Edit Distance Accuracy: 0.682920')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 51
        }
      ]
    }
  ]
}