{
  "id": 61297,
  "title": "Trackml library is not mandatory",
  "url": "/competitions/trackml-particle-identification/discussion/61297",
  "author_name": "",
  "post_date": "2018-07-17T12:30:01.349038900Z",
  "votes": 9,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I see a couple of posts about issues with the trackml library and kernels.  It is possible to make it work, it is explained somewhere in the forum.  But you can also do without it.  Indeed, the library provides basically two features: a python scoring function, and a file loading function., also in python</p>\n\n<p>I shared a python scoring function <a href=\"https://www.kaggle.com/cpmpml/a-faster-python-scoring-function\">here</a>, and I load files without the library in python <a href=\"https://www.kaggle.com/cpmpml/how-weights-are-computed\">here</a>.</p>",
  "messages": [
    {
      "id": "358045",
      "postDate": "07/17/2018 12:30:01",
      "content": "<p>I see a couple of posts about issues with the trackml library and kernels.  It is possible to make it work, it is explained somewhere in the forum.  But you can also do without it.  Indeed, the library provides basically two features: a python scoring function, and a file loading function., also in python</p>\n\n<p>I shared a python scoring function <a href=\"https://www.kaggle.com/cpmpml/a-faster-python-scoring-function\">here</a>, and I load files without the library in python <a href=\"https://www.kaggle.com/cpmpml/how-weights-are-computed\">here</a>.</p>",
      "rawMarkdown": "I see a couple of posts about issues with the trackml library and kernels.  It is possible to make it work, it is explained somewhere in the forum.  But you can also do without it.  Indeed, the library provides basically two features: a python scoring function, and a file loading function., also in python\n\nI shared a python scoring function [here][1], and I load files without the library in python [here][2].\n\n\n  [1]: https://www.kaggle.com/cpmpml/a-faster-python-scoring-function\n  [2]: https://www.kaggle.com/cpmpml/how-weights-are-computed",
      "votes": null
    },
    {
      "id": "358099",
      "postDate": "07/17/2018 14:34:17",
      "content": "<p>A file loading function in trackML sets types for columns in a dataframe. It can be useful I think.</p>",
      "rawMarkdown": "A file loading function in trackML sets types for columns in a dataframe. It can be useful I think.",
      "votes": null
    },
    {
      "id": "358161",
      "postDate": "07/17/2018 16:21:17",
      "content": "<p>Sure, thanks, I changed the title of my post as I did not want to imply the library was not useful.  I'm just providing a workaround in case people cannot use it.</p>",
      "rawMarkdown": "Sure, thanks, I changed the title of my post as I did not want to imply the library was not useful.  I'm just providing a workaround in case people cannot use it.",
      "votes": null
    },
    {
      "id": "358166",
      "postDate": "07/17/2018 16:29:17",
      "content": "<p>The only issue we are aware is the fact that one cannot use the library on GPU with kernel (Kaggle people are aware of this). Does you code work with GPU ? Is it just because your code is inlined instead of loaded from a library ? </p>",
      "rawMarkdown": "The only issue we are aware is the fact that one cannot use the library on GPU with kernel (Kaggle people are aware of this). Does you code work with GPU ? Is it just because your code is inlined instead of loaded from a library ?",
      "votes": null
    },
    {
      "id": "358175",
      "postDate": "07/17/2018 16:46:45",
      "content": "<p>@David, are you asking me?  </p>\n\n<p>From <a href=\"https://www.kaggle.com/c/trackml-particle-identification/discussion/55935\">this discussion</a>, it seems the issue is with installing a custom package on the docker image that supports gpu. I don't think the code of the library is the issue.</p>\n\n<p>My code runs fine on my local machine that has 2 gpus, and I don't see why having gpus would break any Python code actually.  Still, I tried to test my code in Kaggle kernel, but I've been waiting for the gpu kernel to start for quite a while now, I can't wait more unfortunately  </p>",
      "rawMarkdown": "David, are you asking me?  \n\nFrom [this discussion][1], it seems the issue is with installing a custom package on the docker image that supports gpu. I don't think the code of the library is the issue.\n\nMy code runs fine on my local machine that has 2 gpus, and I don't see why having gpus would break any Python code actually.  Still, I tried to test my code in Kaggle kernel, but I've been waiting for the gpu kernel to start for quite a while now, I can't wait more unfortunately  \n\n  [1]: https://www.kaggle.com/c/trackml-particle-identification/discussion/55935",
      "votes": null
    },
    {
      "id": "358253",
      "postDate": "07/17/2018 19:02:07",
      "content": "<p>Thanks @CPMP,  just wanted to be sure. </p>",
      "rawMarkdown": "Thanks @CPMP,  just wanted to be sure.",
      "votes": null
    },
    {
      "id": "359561",
      "postDate": "07/20/2018 10:58:50",
      "content": "<p>Thank you for sharing, just on time</p>",
      "rawMarkdown": "Thank you for sharing, just on time",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 358099,
      "author_name": "sergeyzlobin",
      "author_url": "",
      "post_date": "07/17/2018 14:34:17",
      "content": "<p>A file loading function in trackML sets types for columns in a dataframe. It can be useful I think.</p>",
      "votes": null,
      "replies": [
        {
          "id": 358161,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "07/17/2018 16:21:17",
          "content": "<p>Sure, thanks, I changed the title of my post as I did not want to imply the library was not useful.  I'm just providing a workaround in case people cannot use it.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 358166,
          "author_name": "droussea",
          "author_url": "",
          "post_date": "07/17/2018 16:29:17",
          "content": "<p>The only issue we are aware is the fact that one cannot use the library on GPU with kernel (Kaggle people are aware of this). Does you code work with GPU ? Is it just because your code is inlined instead of loaded from a library ? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 358175,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "07/17/2018 16:46:45",
          "content": "<p>@David, are you asking me?  </p>\n\n<p>From <a href=\"https://www.kaggle.com/c/trackml-particle-identification/discussion/55935\">this discussion</a>, it seems the issue is with installing a custom package on the docker image that supports gpu. I don't think the code of the library is the issue.</p>\n\n<p>My code runs fine on my local machine that has 2 gpus, and I don't see why having gpus would break any Python code actually.  Still, I tried to test my code in Kaggle kernel, but I've been waiting for the gpu kernel to start for quite a while now, I can't wait more unfortunately  </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 358253,
          "author_name": "droussea",
          "author_url": "",
          "post_date": "07/17/2018 19:02:07",
          "content": "<p>Thanks @CPMP,  just wanted to be sure. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 359561,
      "author_name": "blondinka",
      "author_url": "",
      "post_date": "07/20/2018 10:58:50",
      "content": "<p>Thank you for sharing, just on time</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "358045": "I see a couple of posts about issues with the trackml library and kernels.  It is possible to make it work, it is explained somewhere in the forum.  But you can also do without it.  Indeed, the library provides basically two features: a python scoring function, and a file loading function., also in python\n\nI shared a python scoring function [here][1], and I load files without the library in python [here][2].\n\n\n  [1]: https://www.kaggle.com/cpmpml/a-faster-python-scoring-function\n  [2]: https://www.kaggle.com/cpmpml/how-weights-are-computed",
    "358099": "A file loading function in trackML sets types for columns in a dataframe. It can be useful I think.",
    "358161": "Sure, thanks, I changed the title of my post as I did not want to imply the library was not useful.  I'm just providing a workaround in case people cannot use it.",
    "358166": "The only issue we are aware is the fact that one cannot use the library on GPU with kernel (Kaggle people are aware of this). Does you code work with GPU ? Is it just because your code is inlined instead of loaded from a library ?",
    "358175": "David, are you asking me?  \n\nFrom [this discussion][1], it seems the issue is with installing a custom package on the docker image that supports gpu. I don't think the code of the library is the issue.\n\nMy code runs fine on my local machine that has 2 gpus, and I don't see why having gpus would break any Python code actually.  Still, I tried to test my code in Kaggle kernel, but I've been waiting for the gpu kernel to start for quite a while now, I can't wait more unfortunately  \n\n  [1]: https://www.kaggle.com/c/trackml-particle-identification/discussion/55935",
    "358253": "Thanks @CPMP,  just wanted to be sure.",
    "359561": "Thank you for sharing, just on time"
  },
  "source": "meta"
}