{
  "id": 74379,
  "title": "Why is MLP model more overfitting rather than LGB model?",
  "url": "/competitions/PLAsTiCC-2018/discussion/74379",
  "author_name": "",
  "post_date": "2018-12-11T17:44:17.259960200Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello,</p>\n\n<p>My LGB model with 5 fold has 1.006 LB and 0.6080 CV, which is derived from <a href=\"https://www.kaggle.com/cttsai/forked-lgbm-w-ideas-from-kernels-and-discuss\">here</a>.\nMy MLP model with 5 fold has 1.119 LB and 0.5269, derived from <a href=\"https://www.kaggle.com/meaninglesslives/simple-neural-net-for-time-series-classification\">here</a>.</p>\n\n<p>The difference in LGB is approximately 0.40.\nThe difference in MLP is approximately 0.59.</p>\n\n<p>Can anyone explain me why MLP has more overfitting?</p>\n\n<p>I wish success to all participants.</p>",
  "messages": [
    {
      "id": "437311",
      "postDate": "12/11/2018 17:44:17",
      "content": "<p>Hello,</p>\n\n<p>My LGB model with 5 fold has 1.006 LB and 0.6080 CV, which is derived from <a href=\"https://www.kaggle.com/cttsai/forked-lgbm-w-ideas-from-kernels-and-discuss\">here</a>.\nMy MLP model with 5 fold has 1.119 LB and 0.5269, derived from <a href=\"https://www.kaggle.com/meaninglesslives/simple-neural-net-for-time-series-classification\">here</a>.</p>\n\n<p>The difference in LGB is approximately 0.40.\nThe difference in MLP is approximately 0.59.</p>\n\n<p>Can anyone explain me why MLP has more overfitting?</p>\n\n<p>I wish success to all participants.</p>",
      "rawMarkdown": "Hello,\n\nMy LGB model with 5 fold has 1.006 LB and 0.6080 CV, which is derived from [here](https://www.kaggle.com/cttsai/forked-lgbm-w-ideas-from-kernels-and-discuss).\nMy MLP model with 5 fold has 1.119 LB and 0.5269, derived from [here](https://www.kaggle.com/meaninglesslives/simple-neural-net-for-time-series-classification).\n\nThe difference in LGB is approximately 0.40.\nThe difference in MLP is approximately 0.59.\n\n\nCan anyone explain me why MLP has more overfitting?\n\nI wish success to all participants.",
      "votes": null
    },
    {
      "id": "437316",
      "postDate": "12/11/2018 17:50:29",
      "content": "<p>Did you put 'checkPoint' within the loop? If not, first fold's model's weights are used in other folds' models and it's sort of over fitting. </p>",
      "rawMarkdown": "Did you put 'checkPoint' within the loop? If not, first fold's model's weights are used in other folds' models and it's sort of over fitting.",
      "votes": null
    },
    {
      "id": "437356",
      "postDate": "12/11/2018 19:06:20",
      "content": "<p><a href=\"/fatihozturk\">@fatihozturk</a>, yeah. I noticed that I put it out of the loop. The loss expectedly came to where it should be :(( .</p>",
      "rawMarkdown": "fatihozturk, yeah. I noticed that I put it out of the loop. The loss expectedly came to where it should be :(( .",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 437316,
      "author_name": "fatihozturk",
      "author_url": "",
      "post_date": "12/11/2018 17:50:29",
      "content": "<p>Did you put 'checkPoint' within the loop? If not, first fold's model's weights are used in other folds' models and it's sort of over fitting. </p>",
      "votes": null,
      "replies": [
        {
          "id": 437356,
          "author_name": "mbkinaci",
          "author_url": "",
          "post_date": "12/11/2018 19:06:20",
          "content": "<p><a href=\"/fatihozturk\">@fatihozturk</a>, yeah. I noticed that I put it out of the loop. The loss expectedly came to where it should be :(( .</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "437311": "Hello,\n\nMy LGB model with 5 fold has 1.006 LB and 0.6080 CV, which is derived from [here](https://www.kaggle.com/cttsai/forked-lgbm-w-ideas-from-kernels-and-discuss).\nMy MLP model with 5 fold has 1.119 LB and 0.5269, derived from [here](https://www.kaggle.com/meaninglesslives/simple-neural-net-for-time-series-classification).\n\nThe difference in LGB is approximately 0.40.\nThe difference in MLP is approximately 0.59.\n\n\nCan anyone explain me why MLP has more overfitting?\n\nI wish success to all participants.",
    "437316": "Did you put 'checkPoint' within the loop? If not, first fold's model's weights are used in other folds' models and it's sort of over fitting.",
    "437356": "fatihozturk, yeah. I noticed that I put it out of the loop. The loss expectedly came to where it should be :(( ."
  },
  "source": "meta"
}