{
  "id": 159580,
  "title": "Integration of image and tabular data (metadata) for fastai users ",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/159580",
  "author_name": "",
  "post_date": "2020-06-17T22:03:38.793701Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>There have been inspiring discussions in the forum about integrating image and tabular data for this competition. I described my approach to integrate the two data modalities and train a joint model using the integrated data using fastai and image_tabular in my Medium <a href=\"https://towardsdatascience.com/integrating-image-and-tabular-data-for-deep-learning-9281397c7318\">post</a>.</p>\n\n<p>There is also a jupyter notebook demonstrating the entire workflow for this competition in the image_tabular Github <a href=\"https://github.com/naity/image_tabular\">repo</a>. I hope this could serve as a framework for further experimentation and improvement for fastai users. </p>",
  "messages": [
    {
      "id": "891078",
      "postDate": "06/17/2020 22:03:38",
      "content": "<p>There have been inspiring discussions in the forum about integrating image and tabular data for this competition. I described my approach to integrate the two data modalities and train a joint model using the integrated data using fastai and image_tabular in my Medium <a href=\"https://towardsdatascience.com/integrating-image-and-tabular-data-for-deep-learning-9281397c7318\">post</a>.</p>\n\n<p>There is also a jupyter notebook demonstrating the entire workflow for this competition in the image_tabular Github <a href=\"https://github.com/naity/image_tabular\">repo</a>. I hope this could serve as a framework for further experimentation and improvement for fastai users. </p>",
      "rawMarkdown": "There have been inspiring discussions in the forum about integrating image and tabular data for this competition. I described my approach to integrate the two data modalities and train a joint model using the integrated data using fastai and image_tabular in my Medium [post](https://towardsdatascience.com/integrating-image-and-tabular-data-for-deep-learning-9281397c7318).\n\nThere is also a jupyter notebook demonstrating the entire workflow for this competition in the image_tabular Github [repo](https://github.com/naity/image_tabular). I hope this could serve as a framework for further experimentation and improvement for fastai users.",
      "votes": null
    },
    {
      "id": "894229",
      "postDate": "06/20/2020 08:46:21",
      "content": "<p>I have a few questions:\n1- Where did \"train_128\" come from?\n2- How were you able to run 15 epochs? It took me 5 hours with kaggle's GPU on to run epoch. Here's my <a href=\"https://www.kaggle.com/edkahara/fast-ai-v3-melanoma-classification\">notebook</a>.</p>",
      "rawMarkdown": "I have a few questions:\n1- Where did \"train_128\" come from?\n2- How were you able to run 15 epochs? It took me 5 hours with kaggle's GPU on to run epoch. Here's my [notebook](https://www.kaggle.com/edkahara/fast-ai-v3-melanoma-classification).",
      "votes": null
    },
    {
      "id": "894922",
      "postDate": "06/20/2020 22:45:17",
      "content": "<p>Thank you for your questions.</p>\n\n<ol>\n<li>I resized the images myself</li>\n<li>If you resize the images beforehand, it would take less time to train as the images are of different sizes and fastai would need to resize each batch on the fly</li>\n</ol>",
      "rawMarkdown": "Thank you for your questions.\n\n1. I resized the images myself\n2. If you resize the images beforehand, it would take less time to train as the images are of different sizes and fastai would need to resize each batch on the fly",
      "votes": null
    },
    {
      "id": "895157",
      "postDate": "06/21/2020 06:30:03",
      "content": "<p>Thank you!</p>",
      "rawMarkdown": "Thank you!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 894229,
      "author_name": "edkahara",
      "author_url": "",
      "post_date": "06/20/2020 08:46:21",
      "content": "<p>I have a few questions:\n1- Where did \"train_128\" come from?\n2- How were you able to run 15 epochs? It took me 5 hours with kaggle's GPU on to run epoch. Here's my <a href=\"https://www.kaggle.com/edkahara/fast-ai-v3-melanoma-classification\">notebook</a>.</p>",
      "votes": null,
      "replies": [
        {
          "id": 894922,
          "author_name": "yuantian",
          "author_url": "",
          "post_date": "06/20/2020 22:45:17",
          "content": "<p>Thank you for your questions.</p>\n\n<ol>\n<li>I resized the images myself</li>\n<li>If you resize the images beforehand, it would take less time to train as the images are of different sizes and fastai would need to resize each batch on the fly</li>\n</ol>",
          "votes": null,
          "replies": []
        },
        {
          "id": 895157,
          "author_name": "edkahara",
          "author_url": "",
          "post_date": "06/21/2020 06:30:03",
          "content": "<p>Thank you!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "891078": "There have been inspiring discussions in the forum about integrating image and tabular data for this competition. I described my approach to integrate the two data modalities and train a joint model using the integrated data using fastai and image_tabular in my Medium [post](https://towardsdatascience.com/integrating-image-and-tabular-data-for-deep-learning-9281397c7318).\n\nThere is also a jupyter notebook demonstrating the entire workflow for this competition in the image_tabular Github [repo](https://github.com/naity/image_tabular). I hope this could serve as a framework for further experimentation and improvement for fastai users.",
    "894229": "I have a few questions:\n1- Where did \"train_128\" come from?\n2- How were you able to run 15 epochs? It took me 5 hours with kaggle's GPU on to run epoch. Here's my [notebook](https://www.kaggle.com/edkahara/fast-ai-v3-melanoma-classification).",
    "894922": "Thank you for your questions.\n\n1. I resized the images myself\n2. If you resize the images beforehand, it would take less time to train as the images are of different sizes and fastai would need to resize each batch on the fly",
    "895157": "Thank you!"
  },
  "source": "meta"
}