{
  "id": 24473,
  "title": "What is your best single model can do?",
  "url": "/competitions/bosch-production-line-performance/discussion/24473",
  "author_name": "",
  "post_date": "2016-10-16T19:42:14.540Z",
  "votes": null,
  "comment_count": 4,
  "views": 1040,
  "content": "<p>Hi folks,</p>\n\n<p>So far our team members only use xgboost separately and we can only get 0.456 on LB with our best single model. We would like to know if we should continue on feature engineering or start to work on ensemble and stacking methods.</p>\n\n<p>Let's us know what is your best single model's score, please ! :-)  And even better if you can specify which algorithm for that model (xgboost, random forest or deep learning ...)</p>\n\n<p>Thanks,</p>\n\n<p>Titan team.</p>\n\n<p>(PS: Sorry there is a typo in the title but I am unable to edit that).</p>",
  "messages": [
    {
      "id": "139818",
      "postDate": "10/16/2016 19:42:14",
      "content": "<p>Hi folks,</p>\n\n<p>So far our team members only use xgboost separately and we can only get 0.456 on LB with our best single model. We would like to know if we should continue on feature engineering or start to work on ensemble and stacking methods.</p>\n\n<p>Let's us know what is your best single model's score, please ! :-)  And even better if you can specify which algorithm for that model (xgboost, random forest or deep learning ...)</p>\n\n<p>Thanks,</p>\n\n<p>Titan team.</p>\n\n<p>(PS: Sorry there is a typo in the title but I am unable to edit that).</p>",
      "rawMarkdown": "Hi folks,\r\n\r\nSo far our team members only use xgboost separately and we can only get 0.456 on LB with our best single model. We would like to know if we should continue on feature engineering or start to work on ensemble and stacking methods.\r\n\r\nLet's us know what is your best single model's score, please ! :-)  And even better if you can specify which algorithm for that model (xgboost, random forest or deep learning ...)\r\n\r\nThanks,\r\n\r\nTitan team.\r\n\r\n(PS: Sorry there is a typo in the title but I am unable to edit that).",
      "votes": null
    },
    {
      "id": "139821",
      "postDate": "10/16/2016 19:46:02",
      "content": "<p>You can definitely improve your best single model.\nYou should check this thread:\n<a href=\"https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24050/best-single-model\">https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24050/best-single-model</a></p>",
      "rawMarkdown": "You can definitely improve your best single model.\r\nYou should check this thread:\r\nhttps://www.kaggle.com/c/bosch-production-line-performance/forums/t/24050/best-single-model",
      "votes": null
    },
    {
      "id": "139822",
      "postDate": "10/16/2016 19:51:46",
      "content": "<p>Thanks @beluga. I weren't aware of that thread.</p>",
      "rawMarkdown": "Thanks @beluga. I weren't aware of that thread.",
      "votes": null
    },
    {
      "id": "139854",
      "postDate": "10/16/2016 23:33:14",
      "content": "<p>My best Single Model (with XGBoost) is my current score! not used any ensembling or bagging yet. The score is .4929 and used 540 features (including engineered ones). You can definitely improve your single model, keep digging ... </p>",
      "rawMarkdown": "My best Single Model (with XGBoost) is my current score! not used any ensembling or bagging yet. The score is .4929 and used 540 features (including engineered ones). You can definitely improve your single model, keep digging ...",
      "votes": null
    },
    {
      "id": "140020",
      "postDate": "10/17/2016 23:50:16",
      "content": "<p>In our current model we are using about 350 features and only 280 features are present in feature importance (in python), so other features didn't contribute to any split.</p>\n\n<p>But I read in other threads where all numeric features are used. I am not sure how this is going add any value if no split can be made using those features. </p>",
      "rawMarkdown": "In our current model we are using about 350 features and only 280 features are present in feature importance (in python), so other features didn't contribute to any split.\r\n\r\nBut I read in other threads where all numeric features are used. I am not sure how this is going add any value if no split can be made using those features.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 139821,
      "author_name": "gaborfodor",
      "author_url": "",
      "post_date": "10/16/2016 19:46:02",
      "content": "<p>You can definitely improve your best single model.\nYou should check this thread:\n<a href=\"https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24050/best-single-model\">https://www.kaggle.com/c/bosch-production-line-performance/forums/t/24050/best-single-model</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 139822,
      "author_name": "tarobxl",
      "author_url": "",
      "post_date": "10/16/2016 19:51:46",
      "content": "<p>Thanks @beluga. I weren't aware of that thread.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 139854,
      "author_name": "nimashahbazi",
      "author_url": "",
      "post_date": "10/16/2016 23:33:14",
      "content": "<p>My best Single Model (with XGBoost) is my current score! not used any ensembling or bagging yet. The score is .4929 and used 540 features (including engineered ones). You can definitely improve your single model, keep digging ... </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 140020,
      "author_name": "kumaran5555",
      "author_url": "",
      "post_date": "10/17/2016 23:50:16",
      "content": "<p>In our current model we are using about 350 features and only 280 features are present in feature importance (in python), so other features didn't contribute to any split.</p>\n\n<p>But I read in other threads where all numeric features are used. I am not sure how this is going add any value if no split can be made using those features. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "139818": "Hi folks,\r\n\r\nSo far our team members only use xgboost separately and we can only get 0.456 on LB with our best single model. We would like to know if we should continue on feature engineering or start to work on ensemble and stacking methods.\r\n\r\nLet's us know what is your best single model's score, please ! :-)  And even better if you can specify which algorithm for that model (xgboost, random forest or deep learning ...)\r\n\r\nThanks,\r\n\r\nTitan team.\r\n\r\n(PS: Sorry there is a typo in the title but I am unable to edit that).",
    "139821": "You can definitely improve your best single model.\r\nYou should check this thread:\r\nhttps://www.kaggle.com/c/bosch-production-line-performance/forums/t/24050/best-single-model",
    "139822": "Thanks @beluga. I weren't aware of that thread.",
    "139854": "My best Single Model (with XGBoost) is my current score! not used any ensembling or bagging yet. The score is .4929 and used 540 features (including engineered ones). You can definitely improve your single model, keep digging ...",
    "140020": "In our current model we are using about 350 features and only 280 features are present in feature importance (in python), so other features didn't contribute to any split.\r\n\r\nBut I read in other threads where all numeric features are used. I am not sure how this is going add any value if no split can be made using those features."
  },
  "source": "meta"
}