{
  "id": 21054,
  "title": "Good news everyone!",
  "url": "/competitions/expedia-hotel-recommendations/discussion/21054",
  "author_name": "",
  "post_date": "2016-05-18T23:55:09.623Z",
  "votes": null,
  "comment_count": 6,
  "views": 1169,
  "content": "<p>Now for the low price of ~$96 per day* you can get an AWS X1 instance with almost 2TB(!) of RAM and 64 cores! You just might be able to train an xgboost model using the entire training set at once. :-P</p>\n\n<p><a href=\"https://aws.amazon.com/blogs/aws/x1-instances-for-ec2-ready-for-your-memory-intensive-workloads/\">https://aws.amazon.com/blogs/aws/x1-instances-for-ec2-ready-for-your-memory-intensive-workloads/</a></p>\n\n<p>*On a 3-year pricing agreement</p>",
  "messages": [
    {
      "id": "120545",
      "postDate": "05/18/2016 23:55:09",
      "content": "<p>Now for the low price of ~$96 per day* you can get an AWS X1 instance with almost 2TB(!) of RAM and 64 cores! You just might be able to train an xgboost model using the entire training set at once. :-P</p>\n\n<p><a href=\"https://aws.amazon.com/blogs/aws/x1-instances-for-ec2-ready-for-your-memory-intensive-workloads/\">https://aws.amazon.com/blogs/aws/x1-instances-for-ec2-ready-for-your-memory-intensive-workloads/</a></p>\n\n<p>*On a 3-year pricing agreement</p>",
      "rawMarkdown": "Now for the low price of ~$96 per day* you can get an AWS X1 instance with almost 2TB(!) of RAM and 64 cores! You just might be able to train an xgboost model using the entire training set at once. :-P\r\n\r\nhttps://aws.amazon.com/blogs/aws/x1-instances-for-ec2-ready-for-your-memory-intensive-workloads/\r\n\r\n*On a 3-year pricing agreement",
      "votes": null
    },
    {
      "id": "120549",
      "postDate": "05/19/2016 00:18:06",
      "content": "<p>$320 per day if you buy on demand.</p>\n\n<p>Ouch.</p>",
      "rawMarkdown": "$320 per day if you buy on demand.\r\n\r\nOuch.",
      "votes": null
    },
    {
      "id": "120567",
      "postDate": "05/19/2016 05:07:14",
      "content": "<p>@Brandon  - I tried xgboost , not holding good because  large number of classes.</p>",
      "rawMarkdown": "Brandon  - I tried xgboost , not holding good because  large number of classes.",
      "votes": null
    },
    {
      "id": "120568",
      "postDate": "05/19/2016 05:09:18",
      "content": "<p>Ya, but did you try it on a computer with 2 terabytes of RAM?</p>",
      "rawMarkdown": "Ya, but did you try it on a computer with 2 terabytes of RAM?",
      "votes": null
    },
    {
      "id": "120577",
      "postDate": "05/19/2016 07:32:05",
      "content": "<p>Well, xgb <em>does</em> complete (in ~ 12h, but still) on an i7 + 16gb macbook pro. On EC2, m4.large does work a bit faster, haven't checked on 4.4 or 4.10 yet.</p>",
      "rawMarkdown": "Well, xgb *does* complete (in ~ 12h, but still) on an i7 + 16gb macbook pro. On EC2, m4.large does work a bit faster, haven't checked on 4.4 or 4.10 yet.",
      "votes": null
    },
    {
      "id": "120579",
      "postDate": "05/19/2016 07:47:03",
      "content": "<p>I'm using an r3.8xlarge right now, but I added a few hundred features so it's pretty slow going. Using a million samples takes about 8-16 hours depending on my settings. </p>",
      "rawMarkdown": "I'm using an r3.8xlarge right now, but I added a few hundred features so it's pretty slow going. Using a million samples takes about 8-16 hours depending on my settings.",
      "votes": null
    },
    {
      "id": "120599",
      "postDate": "05/19/2016 11:49:06",
      "content": "<p>I tried to train xgboost by batch and blending the the result of each batch. For me, it doesn't work very well..</p>",
      "rawMarkdown": "I tried to train xgboost by batch and blending the the result of each batch. For me, it doesn't work very well..",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 120549,
      "author_name": "inversion",
      "author_url": "",
      "post_date": "05/19/2016 00:18:06",
      "content": "<p>$320 per day if you buy on demand.</p>\n\n<p>Ouch.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120567,
      "author_name": "himanshusinha",
      "author_url": "",
      "post_date": "05/19/2016 05:07:14",
      "content": "<p>@Brandon  - I tried xgboost , not holding good because  large number of classes.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120568,
      "author_name": "brandenkmurray",
      "author_url": "",
      "post_date": "05/19/2016 05:09:18",
      "content": "<p>Ya, but did you try it on a computer with 2 terabytes of RAM?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120577,
      "author_name": "konradb",
      "author_url": "",
      "post_date": "05/19/2016 07:32:05",
      "content": "<p>Well, xgb <em>does</em> complete (in ~ 12h, but still) on an i7 + 16gb macbook pro. On EC2, m4.large does work a bit faster, haven't checked on 4.4 or 4.10 yet.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120579,
      "author_name": "brandenkmurray",
      "author_url": "",
      "post_date": "05/19/2016 07:47:03",
      "content": "<p>I'm using an r3.8xlarge right now, but I added a few hundred features so it's pretty slow going. Using a million samples takes about 8-16 hours depending on my settings. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 120599,
      "author_name": "kuanchen",
      "author_url": "",
      "post_date": "05/19/2016 11:49:06",
      "content": "<p>I tried to train xgboost by batch and blending the the result of each batch. For me, it doesn't work very well..</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "120545": "Now for the low price of ~$96 per day* you can get an AWS X1 instance with almost 2TB(!) of RAM and 64 cores! You just might be able to train an xgboost model using the entire training set at once. :-P\r\n\r\nhttps://aws.amazon.com/blogs/aws/x1-instances-for-ec2-ready-for-your-memory-intensive-workloads/\r\n\r\n*On a 3-year pricing agreement",
    "120549": "$320 per day if you buy on demand.\r\n\r\nOuch.",
    "120567": "Brandon  - I tried xgboost , not holding good because  large number of classes.",
    "120568": "Ya, but did you try it on a computer with 2 terabytes of RAM?",
    "120577": "Well, xgb *does* complete (in ~ 12h, but still) on an i7 + 16gb macbook pro. On EC2, m4.large does work a bit faster, haven't checked on 4.4 or 4.10 yet.",
    "120579": "I'm using an r3.8xlarge right now, but I added a few hundred features so it's pretty slow going. Using a million samples takes about 8-16 hours depending on my settings.",
    "120599": "I tried to train xgboost by batch and blending the the result of each batch. For me, it doesn't work very well.."
  },
  "source": "meta"
}