{
  "id": 169531,
  "title": "exhausted with errors",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/169531",
  "author_name": "",
  "post_date": "2020-07-24T05:41:44.116778600Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p><a href=\"https://www.kaggle.com/vaidicjain/siim-isic-melanoma-classification?scriptVersionId=36235854\">my failed tries</a>\nAfter many tries a failed and getting error while training with TPU with both metadata and image data\ngot .87 with only image data <a href=\"https://www.kaggle.com/vaidicjain/siim-deeplearning-basic-model-score-of-87?scriptVersionId=38475602\">image kernel</a></p>",
  "messages": [
    {
      "id": "943044",
      "postDate": "07/24/2020 05:41:44",
      "content": "<p><a href=\"https://www.kaggle.com/vaidicjain/siim-isic-melanoma-classification?scriptVersionId=36235854\">my failed tries</a>\nAfter many tries a failed and getting error while training with TPU with both metadata and image data\ngot .87 with only image data <a href=\"https://www.kaggle.com/vaidicjain/siim-deeplearning-basic-model-score-of-87?scriptVersionId=38475602\">image kernel</a></p>",
      "rawMarkdown": "[my failed tries](https://www.kaggle.com/vaidicjain/siim-isic-melanoma-classification?scriptVersionId=36235854)\nAfter many tries a failed and getting error while training with TPU with both metadata and image data\ngot .87 with only image data [image kernel](https://www.kaggle.com/vaidicjain/siim-deeplearning-basic-model-score-of-87?scriptVersionId=38475602)",
      "votes": null
    },
    {
      "id": "949630",
      "postDate": "07/28/2020 18:57:05",
      "content": "<p>You can just train a seperate model based on metadata (e.g. XGBClassifier) and blend it with your submissions by giving the image recognition results higher \"weights\" than the results for the metadata model. I have seen some people even claiming its the better way than using the metadata within the image-model. Doing this increased my LB Score significantly.</p>\n\n<p>A quick and easy notebook to learn is e.g.:\n<a href=\"https://www.kaggle.com/blurredmachine/siim-isic-an-ensemble-beginner-s-approach\">https://www.kaggle.com/blurredmachine/siim-isic-an-ensemble-beginner-s-approach</a></p>",
      "rawMarkdown": "You can just train a seperate model based on metadata (e.g. XGBClassifier) and blend it with your submissions by giving the image recognition results higher \"weights\" than the results for the metadata model. I have seen some people even claiming its the better way than using the metadata within the image-model. Doing this increased my LB Score significantly.\n\nA quick and easy notebook to learn is e.g.:\nhttps://www.kaggle.com/blurredmachine/siim-isic-an-ensemble-beginner-s-approach",
      "votes": null
    },
    {
      "id": "949686",
      "postDate": "07/28/2020 20:00:01",
      "content": "<p>thanks for reply and helping,\nI will definately try it😀 </p>",
      "rawMarkdown": "thanks for reply and helping,\nI will definately try it😀",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 949630,
      "author_name": "aliabdin1",
      "author_url": "",
      "post_date": "07/28/2020 18:57:05",
      "content": "<p>You can just train a seperate model based on metadata (e.g. XGBClassifier) and blend it with your submissions by giving the image recognition results higher \"weights\" than the results for the metadata model. I have seen some people even claiming its the better way than using the metadata within the image-model. Doing this increased my LB Score significantly.</p>\n\n<p>A quick and easy notebook to learn is e.g.:\n<a href=\"https://www.kaggle.com/blurredmachine/siim-isic-an-ensemble-beginner-s-approach\">https://www.kaggle.com/blurredmachine/siim-isic-an-ensemble-beginner-s-approach</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 949686,
          "author_name": "vaidicjain",
          "author_url": "",
          "post_date": "07/28/2020 20:00:01",
          "content": "<p>thanks for reply and helping,\nI will definately try it😀 </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "943044": "[my failed tries](https://www.kaggle.com/vaidicjain/siim-isic-melanoma-classification?scriptVersionId=36235854)\nAfter many tries a failed and getting error while training with TPU with both metadata and image data\ngot .87 with only image data [image kernel](https://www.kaggle.com/vaidicjain/siim-deeplearning-basic-model-score-of-87?scriptVersionId=38475602)",
    "949630": "You can just train a seperate model based on metadata (e.g. XGBClassifier) and blend it with your submissions by giving the image recognition results higher \"weights\" than the results for the metadata model. I have seen some people even claiming its the better way than using the metadata within the image-model. Doing this increased my LB Score significantly.\n\nA quick and easy notebook to learn is e.g.:\nhttps://www.kaggle.com/blurredmachine/siim-isic-an-ensemble-beginner-s-approach",
    "949686": "thanks for reply and helping,\nI will definately try it😀"
  },
  "source": "meta"
}