{
  "id": 111289,
  "title": "How high on the leaderboard could one get by just using scikit-learn?",
  "url": "/competitions/Kannada-MNIST/discussion/111289",
  "author_name": "",
  "post_date": "2019-10-04T14:49:42.604638600Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Scanning the leaderboard and looking at a few notebooks makes it clear that to be in the top few, you need to use keras/tensorflow and other methodologies. Not straight sklearn. </p>\n\n<p>My question is if you just use sklearn, how high can one get in terms of score? I am still learning and have not gotten to the fancier stuff, so I am trying to see what my ceiling is.</p>\n\n<p>If anyone has used sklearn and has scored high, please share.</p>",
  "messages": [
    {
      "id": "641269",
      "postDate": "10/04/2019 14:49:42",
      "content": "<p>Scanning the leaderboard and looking at a few notebooks makes it clear that to be in the top few, you need to use keras/tensorflow and other methodologies. Not straight sklearn. </p>\n\n<p>My question is if you just use sklearn, how high can one get in terms of score? I am still learning and have not gotten to the fancier stuff, so I am trying to see what my ceiling is.</p>\n\n<p>If anyone has used sklearn and has scored high, please share.</p>",
      "rawMarkdown": "Scanning the leaderboard and looking at a few notebooks makes it clear that to be in the top few, you need to use keras/tensorflow and other methodologies. Not straight sklearn. \n\nMy question is if you just use sklearn, how high can one get in terms of score? I am still learning and have not gotten to the fancier stuff, so I am trying to see what my ceiling is.\n\nIf anyone has used sklearn and has scored high, please share.",
      "votes": null
    },
    {
      "id": "641410",
      "postDate": "10/04/2019 17:00:22",
      "content": "<p>I think the answer is no. To score high, you need deep learning and CNN (convolutional neural networks). Both are absent in scikit-learn.</p>\n\n<p>You can use tensorflow, keras, pytorch, fast.ai. All these frameworks will be working. But first you should read about CNN and image recognition.</p>\n\n<p>Hope this helps.</p>",
      "rawMarkdown": "I think the answer is no. To score high, you need deep learning and CNN (convolutional neural networks). Both are absent in scikit-learn.\n\nYou can use tensorflow, keras, pytorch, fast.ai. All these frameworks will be working. But first you should read about CNN and image recognition.\n\nHope this helps.",
      "votes": null
    },
    {
      "id": "641434",
      "postDate": "10/04/2019 17:27:06",
      "content": "<p>Thank you Anna. I think it makes more sense now. \nAs I am a beginner, I will be forced to stick with scikit-learn for some more time, but it helps to know why.</p>",
      "rawMarkdown": "Thank you Anna. I think it makes more sense now. \nAs I am a beginner, I will be forced to stick with scikit-learn for some more time, but it helps to know why.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 641410,
      "author_name": "demonplus",
      "author_url": "",
      "post_date": "10/04/2019 17:00:22",
      "content": "<p>I think the answer is no. To score high, you need deep learning and CNN (convolutional neural networks). Both are absent in scikit-learn.</p>\n\n<p>You can use tensorflow, keras, pytorch, fast.ai. All these frameworks will be working. But first you should read about CNN and image recognition.</p>\n\n<p>Hope this helps.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 641434,
      "author_name": "sanjos10",
      "author_url": "",
      "post_date": "10/04/2019 17:27:06",
      "content": "<p>Thank you Anna. I think it makes more sense now. \nAs I am a beginner, I will be forced to stick with scikit-learn for some more time, but it helps to know why.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "641269": "Scanning the leaderboard and looking at a few notebooks makes it clear that to be in the top few, you need to use keras/tensorflow and other methodologies. Not straight sklearn. \n\nMy question is if you just use sklearn, how high can one get in terms of score? I am still learning and have not gotten to the fancier stuff, so I am trying to see what my ceiling is.\n\nIf anyone has used sklearn and has scored high, please share.",
    "641410": "I think the answer is no. To score high, you need deep learning and CNN (convolutional neural networks). Both are absent in scikit-learn.\n\nYou can use tensorflow, keras, pytorch, fast.ai. All these frameworks will be working. But first you should read about CNN and image recognition.\n\nHope this helps.",
    "641434": "Thank you Anna. I think it makes more sense now. \nAs I am a beginner, I will be forced to stick with scikit-learn for some more time, but it helps to know why."
  },
  "source": "meta"
}