{
  "id": 21528,
  "title": "What's is your best grid?",
  "url": "/competitions/facebook-v-predicting-check-ins/discussion/21528",
  "author_name": "",
  "post_date": "2016-06-09T05:47:25.123Z",
  "votes": null,
  "comment_count": 4,
  "views": 610,
  "content": "<p>I assume most of you guys use grid in making predictions. I just wonder what is your best grid so far;) I haven't try anything other than 10 * 10 and I am no sure how much does it affect the predicting performance.</p>",
  "messages": [
    {
      "id": "123010",
      "postDate": "06/09/2016 05:47:25",
      "content": "<p>I assume most of you guys use grid in making predictions. I just wonder what is your best grid so far;) I haven't try anything other than 10 * 10 and I am no sure how much does it affect the predicting performance.</p>",
      "rawMarkdown": "I assume most of you guys use grid in making predictions. I just wonder what is your best grid so far;) I haven't try anything other than 10 * 10 and I am no sure how much does it affect the predicting performance.",
      "votes": null
    },
    {
      "id": "123016",
      "postDate": "06/09/2016 06:58:30",
      "content": "<p>Hi,</p>\n\n<p>I have got my best results with much smaller grids (squares of 100 meters). The problem is that with this grid size you have to train 10.000 models... I have published an script that allows you to test different grid sizes and configurations, to see which works better. </p>\n\n<p><a href=\"https://www.kaggle.com/shdgrao/facebook-v-predicting-check-ins/parallel-training-over-grids\">https://www.kaggle.com/shdgrao/facebook-v-predicting-check-ins/parallel-training-over-grids</a></p>",
      "rawMarkdown": "Hi,\r\n\r\nI have got my best results with much smaller grids (squares of 100 meters). The problem is that with this grid size you have to train 10.000 models... I have published an script that allows you to test different grid sizes and configurations, to see which works better. \r\n\r\nhttps://www.kaggle.com/shdgrao/facebook-v-predicting-check-ins/parallel-training-over-grids",
      "votes": null
    },
    {
      "id": "123017",
      "postDate": "06/09/2016 07:05:26",
      "content": "<p>@shdgrao\nSo you mean a 100 * 100 grid for your best score?</p>",
      "rawMarkdown": "shdgrao\r\nSo you mean a 100 * 100 grid for your best score?",
      "votes": null
    },
    {
      "id": "123019",
      "postDate": "06/09/2016 07:17:56",
      "content": "<p>It depends on models. Some models are sensitive to grid size while others are not. If changing the grid size is significant for a model, smaller grid size tends to work better.</p>",
      "rawMarkdown": "It depends on models. Some models are sensitive to grid size while others are not. If changing the grid size is significant for a model, smaller grid size tends to work better.",
      "votes": null
    },
    {
      "id": "123020",
      "postDate": "06/09/2016 07:25:11",
      "content": "<p>@Jiahuan He. Yes thats what I have used.</p>",
      "rawMarkdown": "Jiahuan He. Yes thats what I have used.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 123016,
      "author_name": "shdgrao",
      "author_url": "",
      "post_date": "06/09/2016 06:58:30",
      "content": "<p>Hi,</p>\n\n<p>I have got my best results with much smaller grids (squares of 100 meters). The problem is that with this grid size you have to train 10.000 models... I have published an script that allows you to test different grid sizes and configurations, to see which works better. </p>\n\n<p><a href=\"https://www.kaggle.com/shdgrao/facebook-v-predicting-check-ins/parallel-training-over-grids\">https://www.kaggle.com/shdgrao/facebook-v-predicting-check-ins/parallel-training-over-grids</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 123017,
      "author_name": "gospursgo",
      "author_url": "",
      "post_date": "06/09/2016 07:05:26",
      "content": "<p>@shdgrao\nSo you mean a 100 * 100 grid for your best score?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 123019,
      "author_name": "citynight",
      "author_url": "",
      "post_date": "06/09/2016 07:17:56",
      "content": "<p>It depends on models. Some models are sensitive to grid size while others are not. If changing the grid size is significant for a model, smaller grid size tends to work better.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 123020,
      "author_name": "shdgrao",
      "author_url": "",
      "post_date": "06/09/2016 07:25:11",
      "content": "<p>@Jiahuan He. Yes thats what I have used.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "123010": "I assume most of you guys use grid in making predictions. I just wonder what is your best grid so far;) I haven't try anything other than 10 * 10 and I am no sure how much does it affect the predicting performance.",
    "123016": "Hi,\r\n\r\nI have got my best results with much smaller grids (squares of 100 meters). The problem is that with this grid size you have to train 10.000 models... I have published an script that allows you to test different grid sizes and configurations, to see which works better. \r\n\r\nhttps://www.kaggle.com/shdgrao/facebook-v-predicting-check-ins/parallel-training-over-grids",
    "123017": "shdgrao\r\nSo you mean a 100 * 100 grid for your best score?",
    "123019": "It depends on models. Some models are sensitive to grid size while others are not. If changing the grid size is significant for a model, smaller grid size tends to work better.",
    "123020": "Jiahuan He. Yes thats what I have used."
  },
  "source": "meta"
}