{
  "id": 73646,
  "title": "Why don't some kernals use GoogleNews for embedding?",
  "url": "/competitions/quora-insincere-questions-classification/discussion/73646",
  "author_name": "",
  "post_date": "2018-12-04T14:58:54.381618500Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>They use glove, fasttext and paragram instead of GoogleNews </p>",
  "messages": [
    {
      "id": "433006",
      "postDate": "12/04/2018 14:58:54",
      "content": "<p>They use glove, fasttext and paragram instead of GoogleNews </p>",
      "rawMarkdown": "They use glove, fasttext and paragram instead of GoogleNews",
      "votes": null
    },
    {
      "id": "433016",
      "postDate": "12/04/2018 15:08:43",
      "content": "<p><a href=\"https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/72507\">https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/72507</a></p>",
      "rawMarkdown": "https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/72507",
      "votes": null
    },
    {
      "id": "435349",
      "postDate": "12/07/2018 22:14:56",
      "content": "<p>Personally, the time and memory requirements to load it are simply impossible. I'm writing and testing my code on my own machine, not trying to code and test and code and test using the web interface. I \"only\" have 8 gig of ram and 8 gig of swap. I couldn't even test LOADING Google's creation 'cause instead of a .CSV file, they chose to do things rather differently... loading it with gensim used WAY too much memory and would have taken WAY too long (I gave up after about a half hour or so). I'm sure Kaggle's machines are a BIT faster than my far from new laptop, and do provide more ram... but it was clear that just loading Google's pretrained vectors would take way too much of the runtime.</p>",
      "rawMarkdown": "Personally, the time and memory requirements to load it are simply impossible. I'm writing and testing my code on my own machine, not trying to code and test and code and test using the web interface. I \"only\" have 8 gig of ram and 8 gig of swap. I couldn't even test LOADING Google's creation 'cause instead of a .CSV file, they chose to do things rather differently... loading it with gensim used WAY too much memory and would have taken WAY too long (I gave up after about a half hour or so). I'm sure Kaggle's machines are a BIT faster than my far from new laptop, and do provide more ram... but it was clear that just loading Google's pretrained vectors would take way too much of the runtime.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 433016,
      "author_name": "shujian",
      "author_url": "",
      "post_date": "12/04/2018 15:08:43",
      "content": "<p><a href=\"https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/72507\">https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/72507</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 435349,
      "author_name": "stevenaleach",
      "author_url": "",
      "post_date": "12/07/2018 22:14:56",
      "content": "<p>Personally, the time and memory requirements to load it are simply impossible. I'm writing and testing my code on my own machine, not trying to code and test and code and test using the web interface. I \"only\" have 8 gig of ram and 8 gig of swap. I couldn't even test LOADING Google's creation 'cause instead of a .CSV file, they chose to do things rather differently... loading it with gensim used WAY too much memory and would have taken WAY too long (I gave up after about a half hour or so). I'm sure Kaggle's machines are a BIT faster than my far from new laptop, and do provide more ram... but it was clear that just loading Google's pretrained vectors would take way too much of the runtime.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "433006": "They use glove, fasttext and paragram instead of GoogleNews",
    "433016": "https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/72507",
    "435349": "Personally, the time and memory requirements to load it are simply impossible. I'm writing and testing my code on my own machine, not trying to code and test and code and test using the web interface. I \"only\" have 8 gig of ram and 8 gig of swap. I couldn't even test LOADING Google's creation 'cause instead of a .CSV file, they chose to do things rather differently... loading it with gensim used WAY too much memory and would have taken WAY too long (I gave up after about a half hour or so). I'm sure Kaggle's machines are a BIT faster than my far from new laptop, and do provide more ram... but it was clear that just loading Google's pretrained vectors would take way too much of the runtime."
  },
  "source": "meta"
}