{
  "id": 198160,
  "title": "Using Tuned Parameters on Larger Sample? ",
  "url": "/competitions/riiid-test-answer-prediction/discussion/198160",
  "author_name": "Kemal Emre Çolak",
  "post_date": "2020-11-20T02:07:02.321000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi,<br>\nI just wanted to know, the I create model using 10^5 data and I want to know what if I use this model hyperparameters on larger sample size like 10^7? Is there any assumptions about that? </p>\n<p>I mean, if two distribution are similar, we can say yes but, I don't want to calculate statistical properties of two different samples (assuming that we have 1.3M row) to know whether they are similar,same or not?</p>\n<p>Conclusion: Can I use hyperparameters that tuned on smaller sample size without calculating any statistical properties? Is there any guarantee that this is going to be good as previous one?</p>\n<p>Sorry if I miss something that I don't know to ask this question.</p>\n<p>Regards,<br>\nEmre</p>",
  "messages": [
    {
      "id": 1084391,
      "postDate": "2020-11-20T02:07:02.320Z",
      "content": "<p>Hi,<br>\nI just wanted to know, the I create model using 10^5 data and I want to know what if I use this model hyperparameters on larger sample size like 10^7? Is there any assumptions about that? </p>\n<p>I mean, if two distribution are similar, we can say yes but, I don't want to calculate statistical properties of two different samples (assuming that we have 1.3M row) to know whether they are similar,same or not?</p>\n<p>Conclusion: Can I use hyperparameters that tuned on smaller sample size without calculating any statistical properties? Is there any guarantee that this is going to be good as previous one?</p>\n<p>Sorry if I miss something that I don't know to ask this question.</p>\n<p>Regards,<br>\nEmre</p>",
      "rawMarkdown": "Hi,\nI just wanted to know, the I create model using 10^5 data and I want to know what if I use this model hyperparameters on larger sample size like 10^7? Is there any assumptions about that? \n\nI mean, if two distribution are similar, we can say yes but, I don't want to calculate statistical properties of two different samples (assuming that we have 1.3M row) to know whether they are similar,same or not?\n\nConclusion: Can I use hyperparameters that tuned on smaller sample size without calculating any statistical properties? Is there any guarantee that this is going to be good as previous one?\n\nSorry if I miss something that I don't know to ask this question.\n\nRegards,\nEmre"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1084391": "Hi,\nI just wanted to know, the I create model using 10^5 data and I want to know what if I use this model hyperparameters on larger sample size like 10^7? Is there any assumptions about that? \n\nI mean, if two distribution are similar, we can say yes but, I don't want to calculate statistical properties of two different samples (assuming that we have 1.3M row) to know whether they are similar,same or not?\n\nConclusion: Can I use hyperparameters that tuned on smaller sample size without calculating any statistical properties? Is there any guarantee that this is going to be good as previous one?\n\nSorry if I miss something that I don't know to ask this question.\n\nRegards,\nEmre"
  }
}