{
  "id": 140082,
  "title": "A VERY SIMPLE APPROACH.....POWER OF LOGISTIC REGRESSION(0.78069 on LB)",
  "url": "/competitions/cat-in-the-dat-ii/discussion/140082",
  "author_name": "",
  "post_date": "2020-03-31T08:27:57.925610900Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Follow the link...\n <a href=\"https://www.kaggle.com/adegladius/my-first-kernel-on-cat-ii-with-0-78069-on-lb\">https://www.kaggle.com/adegladius/my-first-kernel-on-cat-ii-with-0-78069-on-lb</a></p>",
  "messages": [
    {
      "id": "792479",
      "postDate": "03/31/2020 08:27:57",
      "content": "<p>Follow the link...\n <a href=\"https://www.kaggle.com/adegladius/my-first-kernel-on-cat-ii-with-0-78069-on-lb\">https://www.kaggle.com/adegladius/my-first-kernel-on-cat-ii-with-0-78069-on-lb</a></p>",
      "rawMarkdown": "Follow the link...\n https://www.kaggle.com/adegladius/my-first-kernel-on-cat-ii-with-0-78069-on-lb",
      "votes": null
    },
    {
      "id": "797822",
      "postDate": "04/04/2020 23:38:02",
      "content": "<p>Nice notebook ! I like the trick you used to handle a table with 600000 rows and 5777 columns: a conversion to compressed sparse row format.\nThis solution can contribute to an blending.</p>\n\n<p>Some features (from nom_5 to nom_9) take some categories very few times. I think to gather all these rare categories into one categories (to consider them as NaN for example) could improve the model (that could reduce the overfitting). Indeed, the logistic regression have very few samples from these categories so it cannot learn efficiently the corresponding coefficients.\nWhat is more, the title mentions \"0.79377 accuracy\". It is actually an area under the ROC curve equal to 0.79377 with a 5-fold cross-validation on the training set.</p>\n\n<p>Thanks for sharing !</p>",
      "rawMarkdown": "Nice notebook ! I like the trick you used to handle a table with 600000 rows and 5777 columns: a conversion to compressed sparse row format.\nThis solution can contribute to an blending.\n\nSome features (from nom_5 to nom_9) take some categories very few times. I think to gather all these rare categories into one categories (to consider them as NaN for example) could improve the model (that could reduce the overfitting). Indeed, the logistic regression have very few samples from these categories so it cannot learn efficiently the corresponding coefficients.\nWhat is more, the title mentions \"0.79377 accuracy\". It is actually an area under the ROC curve equal to 0.79377 with a 5-fold cross-validation on the training set.\n\nThanks for sharing !",
      "votes": null
    },
    {
      "id": "797956",
      "postDate": "04/05/2020 04:14:02",
      "content": "<p>Thanks Robin. You're welcome 👍</p>",
      "rawMarkdown": "Thanks Robin. You're welcome 👍",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 797822,
      "author_name": "robinmunier",
      "author_url": "",
      "post_date": "04/04/2020 23:38:02",
      "content": "<p>Nice notebook ! I like the trick you used to handle a table with 600000 rows and 5777 columns: a conversion to compressed sparse row format.\nThis solution can contribute to an blending.</p>\n\n<p>Some features (from nom_5 to nom_9) take some categories very few times. I think to gather all these rare categories into one categories (to consider them as NaN for example) could improve the model (that could reduce the overfitting). Indeed, the logistic regression have very few samples from these categories so it cannot learn efficiently the corresponding coefficients.\nWhat is more, the title mentions \"0.79377 accuracy\". It is actually an area under the ROC curve equal to 0.79377 with a 5-fold cross-validation on the training set.</p>\n\n<p>Thanks for sharing !</p>",
      "votes": null,
      "replies": [
        {
          "id": 797956,
          "author_name": "adegladius",
          "author_url": "",
          "post_date": "04/05/2020 04:14:02",
          "content": "<p>Thanks Robin. You're welcome 👍</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "792479": "Follow the link...\n https://www.kaggle.com/adegladius/my-first-kernel-on-cat-ii-with-0-78069-on-lb",
    "797822": "Nice notebook ! I like the trick you used to handle a table with 600000 rows and 5777 columns: a conversion to compressed sparse row format.\nThis solution can contribute to an blending.\n\nSome features (from nom_5 to nom_9) take some categories very few times. I think to gather all these rare categories into one categories (to consider them as NaN for example) could improve the model (that could reduce the overfitting). Indeed, the logistic regression have very few samples from these categories so it cannot learn efficiently the corresponding coefficients.\nWhat is more, the title mentions \"0.79377 accuracy\". It is actually an area under the ROC curve equal to 0.79377 with a 5-fold cross-validation on the training set.\n\nThanks for sharing !",
    "797956": "Thanks Robin. You're welcome 👍"
  },
  "source": "meta"
}