{
  "id": 9430,
  "title": "Hyper-parameter tuning with Python ",
  "url": "/competitions/acquire-valued-shoppers-challenge/discussion/9430",
  "author_name": "",
  "post_date": "2014-06-10T07:53:34.227Z",
  "votes": 3,
  "comment_count": 2,
  "views": 2137,
  "content": "<p style=\"text-align: justify\">Hello everyone,</p>\n<p style=\"text-align: justify\">&nbsp;</p>\n<p style=\"text-align: justify\">recently I wrote a <a href=\"https://github.com/fmfn/Bayesian-Optimization-with-Gaussian-Processes\">python package to perform global optimization with gaussian processes</a>. My main motivation was precisely to find a good set of hyper-parameters when doing cross validation, in particular when the number of parameters is large, or&nbsp;the CV is expensive, and grid search is completely out of question.</p>\n<p style=\"text-align: justify\">&nbsp;</p>\n<p style=\"text-align: justify\">From what I can tell the code is working as intended, however more testing is always welcomed. So if you feel like giving it a try, or if you have parameters you think&nbsp;can be optimized, check out the&nbsp;<a href=\"https://github.com/fmfn/Bayesian-Optimization-with-Gaussian-Processes\">repository</a>. In the examples folder you can find two small examples that hopefully are enough to understand the general idea of how to operate it. However I would gladly answer any questions.<br><br>Also, if you have suggestions, corrections, comments, whatever, I would like to hear it.</p>\n<p style=\"text-align: justify\">&nbsp;</p>\n<p style=\"text-align: justify\">ps: This is still in its early stages, so expect some loose ends here and there, maybe a bug or two, hopefully nothing too serious.<br><br>Thanks,</p>",
  "messages": [
    {
      "id": "48936",
      "postDate": "06/10/2014 07:53:34",
      "content": "<p style=\"text-align: justify\">Hello everyone,</p>\n<p style=\"text-align: justify\">&nbsp;</p>\n<p style=\"text-align: justify\">recently I wrote a <a href=\"https://github.com/fmfn/Bayesian-Optimization-with-Gaussian-Processes\">python package to perform global optimization with gaussian processes</a>. My main motivation was precisely to find a good set of hyper-parameters when doing cross validation, in particular when the number of parameters is large, or&nbsp;the CV is expensive, and grid search is completely out of question.</p>\n<p style=\"text-align: justify\">&nbsp;</p>\n<p style=\"text-align: justify\">From what I can tell the code is working as intended, however more testing is always welcomed. So if you feel like giving it a try, or if you have parameters you think&nbsp;can be optimized, check out the&nbsp;<a href=\"https://github.com/fmfn/Bayesian-Optimization-with-Gaussian-Processes\">repository</a>. In the examples folder you can find two small examples that hopefully are enough to understand the general idea of how to operate it. However I would gladly answer any questions.<br><br>Also, if you have suggestions, corrections, comments, whatever, I would like to hear it.</p>\n<p style=\"text-align: justify\">&nbsp;</p>\n<p style=\"text-align: justify\">ps: This is still in its early stages, so expect some loose ends here and there, maybe a bug or two, hopefully nothing too serious.<br><br>Thanks,</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "49046",
      "postDate": "06/12/2014 11:17:23",
      "content": "<p>Hi Fernando,</p>\n\n<p>This is quite interesting, thank you for sharing. I wonder, though, how this compares to the authors <a href=\"https://github.com/JasperSnoek/spearmint/\">own implementation</a>?</p>\n\n<p>Thanks</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "49070",
      "postDate": "06/12/2014 19:40:54",
      "content": "<p>Hi James.&nbsp;</p>\n\n<p>I am in the process of comparing it to the author's code, hopefully it is comparable. In the next few days I want to turn it more towards a tool for cross validation, and maybe some integration with sklearn even. But as it stands it does not add much to spearmint, I did initially mostly for (self) educational purposes.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 49046,
      "author_name": "jpetterson",
      "author_url": "",
      "post_date": "06/12/2014 11:17:23",
      "content": "<p>Hi Fernando,</p>\n\n<p>This is quite interesting, thank you for sharing. I wonder, though, how this compares to the authors <a href=\"https://github.com/JasperSnoek/spearmint/\">own implementation</a>?</p>\n\n<p>Thanks</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 49070,
      "author_name": "fnogueira",
      "author_url": "",
      "post_date": "06/12/2014 19:40:54",
      "content": "<p>Hi James.&nbsp;</p>\n\n<p>I am in the process of comparing it to the author's code, hopefully it is comparable. In the next few days I want to turn it more towards a tool for cross validation, and maybe some integration with sklearn even. But as it stands it does not add much to spearmint, I did initially mostly for (self) educational purposes.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "48936": "",
    "49046": "",
    "49070": ""
  },
  "source": "meta"
}