{
  "id": 451660,
  "title": "Are Naive Methods Exist?",
  "url": "/competitions/open-problems-single-cell-perturbations/discussion/451660",
  "author_name": "",
  "post_date": "2023-10-30T00:14:31.802131300Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Anybody get high ranks on LB with only using de_train.parquet?</p>",
  "messages": [
    {
      "id": "2504506",
      "postDate": "10/30/2023 00:14:31",
      "content": "<p>Anybody get high ranks on LB with only using de_train.parquet?</p>",
      "rawMarkdown": "Anybody get high ranks on LB with only using de_train.parquet?",
      "votes": null
    },
    {
      "id": "2504515",
      "postDate": "10/30/2023 00:36:32",
      "content": "<p>Not sure to what extent it is considered native. </p>\n<p>I use only <code>de_train.parquet</code>, but don't think the model is naive</p>",
      "rawMarkdown": "Not sure to what extent it is considered native. \n\nI use only `de_train.parquet`, but don't think the model is naive",
      "votes": null
    },
    {
      "id": "2504517",
      "postDate": "10/30/2023 00:44:31",
      "content": "<p>Yeah thank you. I mean \"Naive\" term likes \"Use small part of dataset to handle prob\". Your excellent ranking comes from tree-based model or deep learning model ??? Or maybe ensemble of two these kinds???</p>",
      "rawMarkdown": "Yeah thank you. I mean \"Naive\" term likes \"Use small part of dataset to handle prob\". Your excellent ranking comes from tree-based model or deep learning model ??? Or maybe ensemble of two these kinds???",
      "votes": null
    },
    {
      "id": "2504564",
      "postDate": "10/30/2023 02:27:05",
      "content": "<p>So far only deep learning. Learning about tree-based methods and trying to see if it can help to improve the score.</p>\n<p>And I am curious if anyone can use other files in the dataset to improve their score. Maybe that's the gap between my model and the ones on top of the LB…</p>",
      "rawMarkdown": "So far only deep learning. Learning about tree-based methods and trying to see if it can help to improve the score.\n\nAnd I am curious if anyone can use other files in the dataset to improve their score. Maybe that's the gap between my model and the ones on top of the LB...",
      "votes": null
    },
    {
      "id": "2504574",
      "postDate": "10/30/2023 02:53:36",
      "content": "<p>I see. May be competition has big gap btw LB and Private Board. Anyway, thanks for detailed answer</p>",
      "rawMarkdown": "I see. May be competition has big gap btw LB and Private Board. Anyway, thanks for detailed answer",
      "votes": null
    },
    {
      "id": "2504975",
      "postDate": "10/30/2023 09:38:52",
      "content": "<p>I initially experimented with straightforward methods, but it seems that deep neural networks are more effective in this particular case. My attempt at using the naive methods for multiple regression might have been flawed. You could consider performing separate regression analyses for each label and then combining the results for a potentially better outcome.</p>",
      "rawMarkdown": "I initially experimented with straightforward methods, but it seems that deep neural networks are more effective in this particular case. My attempt at using the naive methods for multiple regression might have been flawed. You could consider performing separate regression analyses for each label and then combining the results for a potentially better outcome.",
      "votes": null
    },
    {
      "id": "2505225",
      "postDate": "10/30/2023 13:20:28",
      "content": "<p>And inclusion, your deep learning for this particular case is efficient without coping with small dataset or overfitting in neural network.</p>",
      "rawMarkdown": "And inclusion, your deep learning for this particular case is efficient without coping with small dataset or overfitting in neural network.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2504515,
      "author_name": "qihuaz",
      "author_url": "",
      "post_date": "10/30/2023 00:36:32",
      "content": "<p>Not sure to what extent it is considered native. </p>\n<p>I use only <code>de_train.parquet</code>, but don't think the model is naive</p>",
      "votes": null,
      "replies": [
        {
          "id": 2504517,
          "author_name": "jjleesunny",
          "author_url": "",
          "post_date": "10/30/2023 00:44:31",
          "content": "<p>Yeah thank you. I mean \"Naive\" term likes \"Use small part of dataset to handle prob\". Your excellent ranking comes from tree-based model or deep learning model ??? Or maybe ensemble of two these kinds???</p>",
          "votes": null,
          "replies": [
            {
              "id": 2504564,
              "author_name": "qihuaz",
              "author_url": "",
              "post_date": "10/30/2023 02:27:05",
              "content": "<p>So far only deep learning. Learning about tree-based methods and trying to see if it can help to improve the score.</p>\n<p>And I am curious if anyone can use other files in the dataset to improve their score. Maybe that's the gap between my model and the ones on top of the LB…</p>",
              "votes": null,
              "replies": [
                {
                  "id": 2504574,
                  "author_name": "jjleesunny",
                  "author_url": "",
                  "post_date": "10/30/2023 02:53:36",
                  "content": "<p>I see. May be competition has big gap btw LB and Private Board. Anyway, thanks for detailed answer</p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    },
    {
      "id": 2504975,
      "author_name": "eliork",
      "author_url": "",
      "post_date": "10/30/2023 09:38:52",
      "content": "<p>I initially experimented with straightforward methods, but it seems that deep neural networks are more effective in this particular case. My attempt at using the naive methods for multiple regression might have been flawed. You could consider performing separate regression analyses for each label and then combining the results for a potentially better outcome.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2505225,
          "author_name": "jjleesunny",
          "author_url": "",
          "post_date": "10/30/2023 13:20:28",
          "content": "<p>And inclusion, your deep learning for this particular case is efficient without coping with small dataset or overfitting in neural network.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2504506": "Anybody get high ranks on LB with only using de_train.parquet?",
    "2504515": "Not sure to what extent it is considered native. \n\nI use only `de_train.parquet`, but don't think the model is naive",
    "2504517": "Yeah thank you. I mean \"Naive\" term likes \"Use small part of dataset to handle prob\". Your excellent ranking comes from tree-based model or deep learning model ??? Or maybe ensemble of two these kinds???",
    "2504564": "So far only deep learning. Learning about tree-based methods and trying to see if it can help to improve the score.\n\nAnd I am curious if anyone can use other files in the dataset to improve their score. Maybe that's the gap between my model and the ones on top of the LB...",
    "2504574": "I see. May be competition has big gap btw LB and Private Board. Anyway, thanks for detailed answer",
    "2504975": "I initially experimented with straightforward methods, but it seems that deep neural networks are more effective in this particular case. My attempt at using the naive methods for multiple regression might have been flawed. You could consider performing separate regression analyses for each label and then combining the results for a potentially better outcome.",
    "2505225": "And inclusion, your deep learning for this particular case is efficient without coping with small dataset or overfitting in neural network."
  },
  "source": "meta"
}