{
  "id": 359222,
  "title": "CITE weights 0.661, MULTI weights 0.339 in !PUBLIC! correlation score",
  "url": "/competitions/open-problems-multimodal/discussion/359222",
  "author_name": "",
  "post_date": "2022-10-11T09:58:53.030154700Z",
  "votes": 18,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Greetings!<br>\nWe have some discussions here about weights of CITE part and MULTI part in submission correlation score.<br>\nMy opinion is: CITE weights 0.661, MULTI weights 0.339.<br>\nAnd explanation follows.<br>\nIf you make submission filled with zeroes, you will get score -1. That means negative correlation.<br>\nIf you make submission filled with random values, you will get score 0. That means no correlation.<br>\nIf you make submission with first 6812820 zeroes (CITE part) and all other values randomized, you will get score -0.661.<br>\nAnd finally if you make submission with first 6812820 random values (CITE part) and all other values equals to zero, you will get score -0.339.<br>\nI am ready to discuss these results.</p>\n<p>P.S.<br>\nThese results actual now, after leak problem solved.</p>\n<p>UPDATE:<br>\nThese results actual fo PUBLIC score only<br>\nFor both PUBLIC and PRIVATE scores weights look <a href=\"https://www.kaggle.com/kaggledummie007/msci-eda-score-weights-of-cite-multi\" target=\"_blank\">here</a></p>",
  "messages": [
    {
      "id": "1982177",
      "postDate": "10/11/2022 09:58:53",
      "content": "<p>Greetings!<br>\nWe have some discussions here about weights of CITE part and MULTI part in submission correlation score.<br>\nMy opinion is: CITE weights 0.661, MULTI weights 0.339.<br>\nAnd explanation follows.<br>\nIf you make submission filled with zeroes, you will get score -1. That means negative correlation.<br>\nIf you make submission filled with random values, you will get score 0. That means no correlation.<br>\nIf you make submission with first 6812820 zeroes (CITE part) and all other values randomized, you will get score -0.661.<br>\nAnd finally if you make submission with first 6812820 random values (CITE part) and all other values equals to zero, you will get score -0.339.<br>\nI am ready to discuss these results.</p>\n<p>P.S.<br>\nThese results actual now, after leak problem solved.</p>\n<p>UPDATE:<br>\nThese results actual fo PUBLIC score only<br>\nFor both PUBLIC and PRIVATE scores weights look <a href=\"https://www.kaggle.com/kaggledummie007/msci-eda-score-weights-of-cite-multi\" target=\"_blank\">here</a></p>",
      "rawMarkdown": "Greetings!\nWe have some discussions here about weights of CITE part and MULTI part in submission correlation score.\nMy opinion is: CITE weights 0.661, MULTI weights 0.339.\nAnd explanation follows.\nIf you make submission filled with zeroes, you will get score -1. That means negative correlation.\nIf you make submission filled with random values, you will get score 0. That means no correlation.\nIf you make submission with first 6812820 zeroes (CITE part) and all other values randomized, you will get score -0.661.\nAnd finally if you make submission with first 6812820 random values (CITE part) and all other values equals to zero, you will get score -0.339.\nI am ready to discuss these results.\n\nP.S.\nThese results actual now, after leak problem solved.\n\nUPDATE:\nThese results actual fo PUBLIC score only\nFor both PUBLIC and PRIVATE scores weights look [here](https://www.kaggle.com/kaggledummie007/msci-eda-score-weights-of-cite-multi)",
      "votes": null
    },
    {
      "id": "1982424",
      "postDate": "10/11/2022 12:21:36",
      "content": "<p>I did some probing and got similar results: CITE ~0.666 MULTI ~0.333</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8507%2Ffb957596a0047496833c7eed1594f9cb%2FRtF21ty.png?generation=1665490870002994&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I did some probing and got similar results: CITE ~0.666 MULTI ~0.333\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8507%2Ffb957596a0047496833c7eed1594f9cb%2FRtF21ty.png?generation=1665490870002994&alt=media)",
      "votes": null
    },
    {
      "id": "1982722",
      "postDate": "10/11/2022 15:26:22",
      "content": "<p>According my submissions I got  CITE ~0.689; MULTI ~0.311</p>",
      "rawMarkdown": "According my submissions I got  CITE ~0.689; MULTI ~0.311",
      "votes": null
    },
    {
      "id": "1982804",
      "postDate": "10/11/2022 16:30:05",
      "content": "<p>submission made with (any) model will not give you score of 1.0<br>\nand you can only guess what error do you have in CITE part and what error is in MULTI part<br>\nbut using random or zero values you can get scores equal to 0.000 and -1.0<br>\nif you use random and zero values for CITE and MULTI parts separately you will get 0.661 and 0.339</p>",
      "rawMarkdown": "submission made with (any) model will not give you score of 1.0\nand you can only guess what error do you have in CITE part and what error is in MULTI part\nbut using random or zero values you can get scores equal to 0.000 and -1.0\nif you use random and zero values for CITE and MULTI parts separately you will get 0.661 and 0.339",
      "votes": null
    },
    {
      "id": "1983767",
      "postDate": "10/12/2022 08:15:36",
      "content": "<p>Your inference is quite reasonable.I have a question that Which part is now easier to improve public list scores，CITE or Mul?</p>",
      "rawMarkdown": "Your inference is quite reasonable.I have a question that Which part is now easier to improve public list scores，CITE or Mul?",
      "votes": null
    },
    {
      "id": "1983814",
      "postDate": "10/12/2022 08:51:49",
      "content": "<p>thanx<br>\nin terms of influence on result score SITE part is twice more important<br>\nbut i am not shure that it is easier to improve in terms of data / model</p>",
      "rawMarkdown": "thanx\nin terms of influence on result score SITE part is twice more important\nbut i am not shure that it is easier to improve in terms of data / model",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1982424,
      "author_name": "antorsae",
      "author_url": "",
      "post_date": "10/11/2022 12:21:36",
      "content": "<p>I did some probing and got similar results: CITE ~0.666 MULTI ~0.333</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8507%2Ffb957596a0047496833c7eed1594f9cb%2FRtF21ty.png?generation=1665490870002994&amp;alt=media\" alt=\"\"></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1982722,
      "author_name": "josecarmona",
      "author_url": "",
      "post_date": "10/11/2022 15:26:22",
      "content": "<p>According my submissions I got  CITE ~0.689; MULTI ~0.311</p>",
      "votes": null,
      "replies": [
        {
          "id": 1982804,
          "author_name": "kaggledummie007",
          "author_url": "",
          "post_date": "10/11/2022 16:30:05",
          "content": "<p>submission made with (any) model will not give you score of 1.0<br>\nand you can only guess what error do you have in CITE part and what error is in MULTI part<br>\nbut using random or zero values you can get scores equal to 0.000 and -1.0<br>\nif you use random and zero values for CITE and MULTI parts separately you will get 0.661 and 0.339</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1983767,
      "author_name": "zhangyuan2000",
      "author_url": "",
      "post_date": "10/12/2022 08:15:36",
      "content": "<p>Your inference is quite reasonable.I have a question that Which part is now easier to improve public list scores，CITE or Mul?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1983814,
          "author_name": "kaggledummie007",
          "author_url": "",
          "post_date": "10/12/2022 08:51:49",
          "content": "<p>thanx<br>\nin terms of influence on result score SITE part is twice more important<br>\nbut i am not shure that it is easier to improve in terms of data / model</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1982177": "Greetings!\nWe have some discussions here about weights of CITE part and MULTI part in submission correlation score.\nMy opinion is: CITE weights 0.661, MULTI weights 0.339.\nAnd explanation follows.\nIf you make submission filled with zeroes, you will get score -1. That means negative correlation.\nIf you make submission filled with random values, you will get score 0. That means no correlation.\nIf you make submission with first 6812820 zeroes (CITE part) and all other values randomized, you will get score -0.661.\nAnd finally if you make submission with first 6812820 random values (CITE part) and all other values equals to zero, you will get score -0.339.\nI am ready to discuss these results.\n\nP.S.\nThese results actual now, after leak problem solved.\n\nUPDATE:\nThese results actual fo PUBLIC score only\nFor both PUBLIC and PRIVATE scores weights look [here](https://www.kaggle.com/kaggledummie007/msci-eda-score-weights-of-cite-multi)",
    "1982424": "I did some probing and got similar results: CITE ~0.666 MULTI ~0.333\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8507%2Ffb957596a0047496833c7eed1594f9cb%2FRtF21ty.png?generation=1665490870002994&alt=media)",
    "1982722": "According my submissions I got  CITE ~0.689; MULTI ~0.311",
    "1982804": "submission made with (any) model will not give you score of 1.0\nand you can only guess what error do you have in CITE part and what error is in MULTI part\nbut using random or zero values you can get scores equal to 0.000 and -1.0\nif you use random and zero values for CITE and MULTI parts separately you will get 0.661 and 0.339",
    "1983767": "Your inference is quite reasonable.I have a question that Which part is now easier to improve public list scores，CITE or Mul?",
    "1983814": "thanx\nin terms of influence on result score SITE part is twice more important\nbut i am not shure that it is easier to improve in terms of data / model"
  },
  "source": "meta"
}