{
  "id": 236028,
  "title": "removing jaccard similarity lowers my score",
  "url": "/competitions/coleridgeinitiative-show-us-the-data/discussion/236028",
  "author_name": "",
  "post_date": "2021-05-02T13:15:37.145229Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have found when I remove any predictions with jaccard similarity &gt;.5 my scores go down about 3%. Has anyone else experienced this?</p>\n<p>I am working on baseline test with public code to demonstrate score with and without jaccard similarity removal.</p>",
  "messages": [
    {
      "id": "1290835",
      "postDate": "05/02/2021 13:15:37",
      "content": "<p>I have found when I remove any predictions with jaccard similarity &gt;.5 my scores go down about 3%. Has anyone else experienced this?</p>\n<p>I am working on baseline test with public code to demonstrate score with and without jaccard similarity removal.</p>",
      "rawMarkdown": "I have found when I remove any predictions with jaccard similarity >.5 my scores go down about 3%. Has anyone else experienced this?\n\nI am working on baseline test with public code to demonstrate score with and without jaccard similarity removal.",
      "votes": null
    },
    {
      "id": "1290863",
      "postDate": "05/02/2021 13:36:22",
      "content": "<p>Yes I experienced the same.. IMO this happens because we need to predict all unique dataset mentions (if a dataset is mentioned with exactly the same name twice, we only need to include it once).</p>",
      "rawMarkdown": "Yes I experienced the same.. IMO this happens because we need to predict all unique dataset mentions (if a dataset is mentioned with exactly the same name twice, we only need to include it once).",
      "votes": null
    },
    {
      "id": "1291070",
      "postDate": "05/02/2021 17:17:50",
      "content": "<p>Yes.</p>\n<p>I've also experienced that training my model, using machine learning, and in general, doing anything besides brute force matching, lowers my score.</p>",
      "rawMarkdown": "Yes.\n\nI've also experienced that training my model, using machine learning, and in general, doing anything besides brute force matching, lowers my score.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1290863,
      "author_name": "hannes82",
      "author_url": "",
      "post_date": "05/02/2021 13:36:22",
      "content": "<p>Yes I experienced the same.. IMO this happens because we need to predict all unique dataset mentions (if a dataset is mentioned with exactly the same name twice, we only need to include it once).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1291070,
      "author_name": "verracodeguacas",
      "author_url": "",
      "post_date": "05/02/2021 17:17:50",
      "content": "<p>Yes.</p>\n<p>I've also experienced that training my model, using machine learning, and in general, doing anything besides brute force matching, lowers my score.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1290835": "I have found when I remove any predictions with jaccard similarity >.5 my scores go down about 3%. Has anyone else experienced this?\n\nI am working on baseline test with public code to demonstrate score with and without jaccard similarity removal.",
    "1290863": "Yes I experienced the same.. IMO this happens because we need to predict all unique dataset mentions (if a dataset is mentioned with exactly the same name twice, we only need to include it once).",
    "1291070": "Yes.\n\nI've also experienced that training my model, using machine learning, and in general, doing anything besides brute force matching, lowers my score."
  },
  "source": "meta"
}