{
  "id": 181167,
  "title": "Additional powerful hardware is necessary?",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/181167",
  "author_name": "",
  "post_date": "2020-09-07T21:44:01.597230100Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hello everyone<br>\nI want to know that ,<br>\nDoes this challenge can be executed by kaggle servers or It needs additional more powerful hardware that must be prepared by ourselves?</p>",
  "messages": [
    {
      "id": "1002133",
      "postDate": "09/07/2020 21:44:01",
      "content": "<p>Hello everyone<br>\nI want to know that ,<br>\nDoes this challenge can be executed by kaggle servers or It needs additional more powerful hardware that must be prepared by ourselves?</p>",
      "rawMarkdown": "Hello everyone\nI want to know that ,\nDoes this challenge can be executed by kaggle servers or It needs additional more powerful hardware that must be prepared by ourselves?",
      "votes": null
    },
    {
      "id": "1002533",
      "postDate": "09/08/2020 07:53:50",
      "content": "<p>the provided TPU resources are very powerful, I believe you have a good chance with that</p>\n<p>And usually, clever solutions win over raw computing power</p>",
      "rawMarkdown": "the provided TPU resources are very powerful, I believe you have a good chance with that\n\nAnd usually, clever solutions win over raw computing power",
      "votes": null
    },
    {
      "id": "1003039",
      "postDate": "09/08/2020 16:08:15",
      "content": "<p>I feel the same way. Especially a lot of time is spent in just pulling data out of the Zarr files. I am not sure how other folks are able to make any predictions at this rate. </p>",
      "rawMarkdown": "I feel the same way. Especially a lot of time is spent in just pulling data out of the Zarr files. I am not sure how other folks are able to make any predictions at this rate.",
      "votes": null
    },
    {
      "id": "1003146",
      "postDate": "09/08/2020 17:34:50",
      "content": "<p><a href=\"https://www.kaggle.com/behrouzbo\" target=\"_blank\">@behrouzbo</a> and <a href=\"https://www.kaggle.com/srinivasgopalkrishna\" target=\"_blank\">@srinivasgopalkrishna</a> there are methods to store training weights and resume training. So you can do multiple sessions on Kaggle itself and train model. Just wanted to let you know in case you aren’t aware of intermittent training. </p>",
      "rawMarkdown": "behrouzbo and @srinivasgopalkrishna there are methods to store training weights and resume training. So you can do multiple sessions on Kaggle itself and train model. Just wanted to let you know in case you aren’t aware of intermittent training.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1002533,
      "author_name": "ilu000",
      "author_url": "",
      "post_date": "09/08/2020 07:53:50",
      "content": "<p>the provided TPU resources are very powerful, I believe you have a good chance with that</p>\n<p>And usually, clever solutions win over raw computing power</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1003039,
      "author_name": "srinivasgopalkrishna",
      "author_url": "",
      "post_date": "09/08/2020 16:08:15",
      "content": "<p>I feel the same way. Especially a lot of time is spent in just pulling data out of the Zarr files. I am not sure how other folks are able to make any predictions at this rate. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1003146,
          "author_name": "deepakrajpurushothaman",
          "author_url": "",
          "post_date": "09/08/2020 17:34:50",
          "content": "<p><a href=\"https://www.kaggle.com/behrouzbo\" target=\"_blank\">@behrouzbo</a> and <a href=\"https://www.kaggle.com/srinivasgopalkrishna\" target=\"_blank\">@srinivasgopalkrishna</a> there are methods to store training weights and resume training. So you can do multiple sessions on Kaggle itself and train model. Just wanted to let you know in case you aren’t aware of intermittent training. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1002133": "Hello everyone\nI want to know that ,\nDoes this challenge can be executed by kaggle servers or It needs additional more powerful hardware that must be prepared by ourselves?",
    "1002533": "the provided TPU resources are very powerful, I believe you have a good chance with that\n\nAnd usually, clever solutions win over raw computing power",
    "1003039": "I feel the same way. Especially a lot of time is spent in just pulling data out of the Zarr files. I am not sure how other folks are able to make any predictions at this rate.",
    "1003146": "behrouzbo and @srinivasgopalkrishna there are methods to store training weights and resume training. So you can do multiple sessions on Kaggle itself and train model. Just wanted to let you know in case you aren’t aware of intermittent training."
  },
  "source": "meta"
}