{
  "id": 58560,
  "title": "Do I need GPU or cloud computation?",
  "url": "/competitions/avito-demand-prediction/discussion/58560",
  "author_name": "",
  "post_date": "2018-06-10T15:04:02.591051500Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have no experience of running ML algorithms on hundred of G data including photos.  It seems to me that deep NN is needed as one ingredient.   Is it OK I can just use a mac (no GPU) for the computation ?</p>",
  "messages": [
    {
      "id": "340888",
      "postDate": "06/10/2018 15:04:02",
      "content": "<p>I have no experience of running ML algorithms on hundred of G data including photos.  It seems to me that deep NN is needed as one ingredient.   Is it OK I can just use a mac (no GPU) for the computation ?</p>",
      "rawMarkdown": "I have no experience of running ML algorithms on hundred of G data including photos.  It seems to me that deep NN is needed as one ingredient.   Is it OK I can just use a mac (no GPU) for the computation ?",
      "votes": null
    },
    {
      "id": "341718",
      "postDate": "06/12/2018 05:56:06",
      "content": "<p>You can get pretty far using kaggle kernel or google colab which both offer free GPU power (I use them for running tests and tuning the network when my own GPU is blocked by training), but have restrictions on RAM and space. I am afraid including the images directly will be hard to organize due to disk space restrictions. Some shared extracted image features in kernels. So you could start incorporating those. Hope that helps.</p>",
      "rawMarkdown": "You can get pretty far using kaggle kernel or google colab which both offer free GPU power (I use them for running tests and tuning the network when my own GPU is blocked by training), but have restrictions on RAM and space. I am afraid including the images directly will be hard to organize due to disk space restrictions. Some shared extracted image features in kernels. So you could start incorporating those. Hope that helps.",
      "votes": null
    },
    {
      "id": "343584",
      "postDate": "06/15/2018 16:06:54",
      "content": "<p>Thanks a lot!  That sure helped, especially for pointing out the extracted images features shared by some.  I will check it out and most likely start from there.</p>",
      "rawMarkdown": "Thanks a lot!  That sure helped, especially for pointing out the extracted images features shared by some.  I will check it out and most likely start from there.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 341718,
      "author_name": "christofhenkel",
      "author_url": "",
      "post_date": "06/12/2018 05:56:06",
      "content": "<p>You can get pretty far using kaggle kernel or google colab which both offer free GPU power (I use them for running tests and tuning the network when my own GPU is blocked by training), but have restrictions on RAM and space. I am afraid including the images directly will be hard to organize due to disk space restrictions. Some shared extracted image features in kernels. So you could start incorporating those. Hope that helps.</p>",
      "votes": null,
      "replies": [
        {
          "id": 343584,
          "author_name": "kaggler2018",
          "author_url": "",
          "post_date": "06/15/2018 16:06:54",
          "content": "<p>Thanks a lot!  That sure helped, especially for pointing out the extracted images features shared by some.  I will check it out and most likely start from there.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "340888": "I have no experience of running ML algorithms on hundred of G data including photos.  It seems to me that deep NN is needed as one ingredient.   Is it OK I can just use a mac (no GPU) for the computation ?",
    "341718": "You can get pretty far using kaggle kernel or google colab which both offer free GPU power (I use them for running tests and tuning the network when my own GPU is blocked by training), but have restrictions on RAM and space. I am afraid including the images directly will be hard to organize due to disk space restrictions. Some shared extracted image features in kernels. So you could start incorporating those. Hope that helps.",
    "343584": "Thanks a lot!  That sure helped, especially for pointing out the extracted images features shared by some.  I will check it out and most likely start from there."
  },
  "source": "meta"
}