{
  "id": 75879,
  "title": "zeros can get 3.87 in training set",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/75879",
  "author_name": "",
  "post_date": "2018-12-27T08:58:25.803037500Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I found that even set predictions to zeros, the RMSE in training set is only 3.87. \nHowever, most submissions are around 3.7, just a little bit lower than zeros. \nI think we are going the wrong way :(</p>",
  "messages": [
    {
      "id": "445947",
      "postDate": "12/27/2018 08:58:25",
      "content": "<p>I found that even set predictions to zeros, the RMSE in training set is only 3.87. \nHowever, most submissions are around 3.7, just a little bit lower than zeros. \nI think we are going the wrong way :(</p>",
      "rawMarkdown": "I found that even set predictions to zeros, the RMSE in training set is only 3.87. \nHowever, most submissions are around 3.7, just a little bit lower than zeros. \nI think we are going the wrong way :(",
      "votes": null
    },
    {
      "id": "446071",
      "postDate": "12/27/2018 13:04:43",
      "content": "<p>yes, you're right! RMSE is sensitive to outliers(target=-33.x) which can not be handled by regression. So <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/74519\">A large ratio of RMSE is caused by outlier</a></p>",
      "rawMarkdown": "yes, you're right! RMSE is sensitive to outliers(target=-33.x) which can not be handled by regression. So [A large ratio of RMSE is caused by outlier](https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/74519)",
      "votes": null
    },
    {
      "id": "446119",
      "postDate": "12/27/2018 14:37:49",
      "content": "<p>Outliers are so critical, you need to build a Classifier to identify them and after a Regressor to predict the target, also there are some patterns in these outliers that I manage to find but my model is ignoring them</p>",
      "rawMarkdown": "Outliers are so critical, you need to build a Classifier to identify them and after a Regressor to predict the target, also there are some patterns in these outliers that I manage to find but my model is ignoring them",
      "votes": null
    },
    {
      "id": "446122",
      "postDate": "12/27/2018 14:47:31",
      "content": "<p>I really did a classifier to identify outliers, but it only helps a little.\nsee this kernel:\n<a href=\"https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier\">https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier</a></p>",
      "rawMarkdown": "I really did a classifier to identify outliers, but it only helps a little.\nsee this kernel:\nhttps://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 446071,
      "author_name": "jiazhuang",
      "author_url": "",
      "post_date": "12/27/2018 13:04:43",
      "content": "<p>yes, you're right! RMSE is sensitive to outliers(target=-33.x) which can not be handled by regression. So <a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/74519\">A large ratio of RMSE is caused by outlier</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 446119,
      "author_name": "cv13j0",
      "author_url": "",
      "post_date": "12/27/2018 14:37:49",
      "content": "<p>Outliers are so critical, you need to build a Classifier to identify them and after a Regressor to predict the target, also there are some patterns in these outliers that I manage to find but my model is ignoring them</p>",
      "votes": null,
      "replies": [
        {
          "id": 446122,
          "author_name": "sunyc11",
          "author_url": "",
          "post_date": "12/27/2018 14:47:31",
          "content": "<p>I really did a classifier to identify outliers, but it only helps a little.\nsee this kernel:\n<a href=\"https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier\">https://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "445947": "I found that even set predictions to zeros, the RMSE in training set is only 3.87. \nHowever, most submissions are around 3.7, just a little bit lower than zeros. \nI think we are going the wrong way :(",
    "446071": "yes, you're right! RMSE is sensitive to outliers(target=-33.x) which can not be handled by regression. So [A large ratio of RMSE is caused by outlier](https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/74519)",
    "446119": "Outliers are so critical, you need to build a Classifier to identify them and after a Regressor to predict the target, also there are some patterns in these outliers that I manage to find but my model is ignoring them",
    "446122": "I really did a classifier to identify outliers, but it only helps a little.\nsee this kernel:\nhttps://www.kaggle.com/waitingli/combining-your-model-with-a-model-without-outlier"
  },
  "source": "meta"
}