{
  "id": 162398,
  "title": "Additional tabular data",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/162398",
  "author_name": "Marcelo Kittlein",
  "post_date": "2020-06-28T18:28:37.448000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I have just used tabular data using metrics derived from landscape analysis (basically it extract landscape features from melanoma images) it is available for download <a href=\"https://www.kaggle.com/kittlein/landscape\">here</a>\nBasic calculations are availabe in <a href=\"https://www.kaggle.com/kittlein/landscape-metrics-for-melanomas\">this notebook</a></p>\n\n<p>It raised the LB score of a very simple cnn on 128x128 resized images from 0.78~ to 0.90~ by adding a layer which averages the outputs of the 2DCNN (images) and a 1D CNN (this tabular).\nPerhaps someone with an already high scoring model would want to give it a try and comment on it. If you concatenate both sets of data there is 259 features to use...</p>",
  "messages": [
    {
      "id": 905717,
      "postDate": "2020-06-28T18:28:37.450Z",
      "content": "<p>I have just used tabular data using metrics derived from landscape analysis (basically it extract landscape features from melanoma images) it is available for download <a href=\"https://www.kaggle.com/kittlein/landscape\">here</a>\nBasic calculations are availabe in <a href=\"https://www.kaggle.com/kittlein/landscape-metrics-for-melanomas\">this notebook</a></p>\n\n<p>It raised the LB score of a very simple cnn on 128x128 resized images from 0.78~ to 0.90~ by adding a layer which averages the outputs of the 2DCNN (images) and a 1D CNN (this tabular).\nPerhaps someone with an already high scoring model would want to give it a try and comment on it. If you concatenate both sets of data there is 259 features to use...</p>",
      "rawMarkdown": "I have just used tabular data using metrics derived from landscape analysis (basically it extract landscape features from melanoma images) it is available for download [here](https://www.kaggle.com/kittlein/landscape)\nBasic calculations are availabe in [this notebook](https://www.kaggle.com/kittlein/landscape-metrics-for-melanomas)\n\nIt raised the LB score of a very simple cnn on 128x128 resized images from 0.78~ to 0.90~ by adding a layer which averages the outputs of the 2DCNN (images) and a 1D CNN (this tabular).\nPerhaps someone with an already high scoring model would want to give it a try and comment on it. If you concatenate both sets of data there is 259 features to use...",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "905717": "I have just used tabular data using metrics derived from landscape analysis (basically it extract landscape features from melanoma images) it is available for download [here](https://www.kaggle.com/kittlein/landscape)\nBasic calculations are availabe in [this notebook](https://www.kaggle.com/kittlein/landscape-metrics-for-melanomas)\n\nIt raised the LB score of a very simple cnn on 128x128 resized images from 0.78~ to 0.90~ by adding a layer which averages the outputs of the 2DCNN (images) and a 1D CNN (this tabular).\nPerhaps someone with an already high scoring model would want to give it a try and comment on it. If you concatenate both sets of data there is 259 features to use..."
  }
}