{
  "id": 83596,
  "title": "Kaggle Days Paris by Grandmaster- \"How to win competitions, spending the minimum amount of time”",
  "url": "/competitions/vsb-power-line-fault-detection/discussion/83596",
  "author_name": "",
  "post_date": "2019-03-11T12:13:49.814856300Z",
  "votes": 14,
  "comment_count": 8,
  "views": 0,
  "content": "<p>Recently a top notch Grandmaster Pavel Ostyakov shared some tricks about winning Kaggle competitions. I found it must a watch, hope you will agree (main tip: stay away from jupyter notebooks). <a href=\"https://www.youtube.com/watch?v=oxikDxBzKUU\">https://www.youtube.com/watch?v=oxikDxBzKUU</a></p>\n\n<p><strong>Edit</strong></p>\n\n<p>Following are the timestamps with topics TLDR\n1:20 Speedup convergence (Reduce Learning Rate at Plateaus, Cyclic Learning Rate, Warmup)\n10:39 How to improve training of NN, speedup convergence, Data Augmentation\n19:16 Example of a well organised pipeline\n21:06 7 tips on winning at Kaggle\n24:48 Main problem of competition, goals</p>",
  "messages": [
    {
      "id": "487741",
      "postDate": "03/11/2019 12:13:49",
      "content": "<p>Recently a top notch Grandmaster Pavel Ostyakov shared some tricks about winning Kaggle competitions. I found it must a watch, hope you will agree (main tip: stay away from jupyter notebooks). <a href=\"https://www.youtube.com/watch?v=oxikDxBzKUU\">https://www.youtube.com/watch?v=oxikDxBzKUU</a></p>\n\n<p><strong>Edit</strong></p>\n\n<p>Following are the timestamps with topics TLDR\n1:20 Speedup convergence (Reduce Learning Rate at Plateaus, Cyclic Learning Rate, Warmup)\n10:39 How to improve training of NN, speedup convergence, Data Augmentation\n19:16 Example of a well organised pipeline\n21:06 7 tips on winning at Kaggle\n24:48 Main problem of competition, goals</p>",
      "rawMarkdown": "Recently a top notch Grandmaster Pavel Ostyakov shared some tricks about winning Kaggle competitions. I found it must a watch, hope you will agree (main tip: stay away from jupyter notebooks). https://www.youtube.com/watch?v=oxikDxBzKUU\n\n**Edit**\n\nFollowing are the timestamps with topics TLDR\n1:20 Speedup convergence (Reduce Learning Rate at Plateaus, Cyclic Learning Rate, Warmup)\n10:39 How to improve training of NN, speedup convergence, Data Augmentation\n19:16 Example of a well organised pipeline\n21:06 7 tips on winning at Kaggle\n24:48 Main problem of competition, goals",
      "votes": null
    },
    {
      "id": "487750",
      "postDate": "03/11/2019 12:28:58",
      "content": "<p>Hi <a href=\"/cyberia\">@cyberia</a> !</p>\n\n<p>Thanks for sharing this!</p>\n\n<p>Lots of interesting stuff is available on <a href=\"https://kaggledays.com/\">https://kaggledays.com/</a></p>",
      "rawMarkdown": "Hi @cyberia !\n\nThanks for sharing this!\n\nLots of interesting stuff is available on https://kaggledays.com/",
      "votes": null
    },
    {
      "id": "487785",
      "postDate": "03/11/2019 13:16:28",
      "content": "<blockquote>\n  <p>main tip: stay away from jupyter notebooks</p>\n</blockquote>\n\n<p>Well, if this is the main takeaway you get, then it may be worth viewing it again ;)</p>",
      "rawMarkdown": "&gt; main tip: stay away from jupyter notebooks\n\nWell, if this is the main takeaway you get, then it may be worth viewing it again ;)",
      "votes": null
    },
    {
      "id": "487812",
      "postDate": "03/11/2019 13:46:29",
      "content": "<p>JFP, main tip is from extended video description I just wrote it as it is. Of course there are so many things to learn from Pavel's talk. I closely follow your work, it is inspiring and awesome. Ah now I see your game :)))), here is the link to another awesome talk by CPMP at Kaggle Days Paris <a href=\"https://www.youtube.com/watch?v=fH_FiquKhiI\">https://www.youtube.com/watch?v=fH_FiquKhiI</a>\nIts good to hear your talk which includes Gaussian processes and sampling bias. I always appreciate the beauty of gaussian models, myself being from bayesian camp.</p>",
      "rawMarkdown": "JFP, main tip is from extended video description I just wrote it as it is. Of course there are so many things to learn from Pavel's talk. I closely follow your work, it is inspiring and awesome. Ah now I see your game :)))), here is the link to another awesome talk by CPMP at Kaggle Days Paris https://www.youtube.com/watch?v=fH_FiquKhiI\nIts good to hear your talk which includes Gaussian processes and sampling bias. I always appreciate the beauty of gaussian models, myself being from bayesian camp.",
      "votes": null
    },
    {
      "id": "487858",
      "postDate": "03/11/2019 14:53:08",
      "content": "<p>If you think I posted to get ad, you're wrong.  I hesitated posting my comment as I knew it could be misinterpreted that way.</p>\n\n<p>Anyway, my point was that Pavel's talk is way more interesting than just his opinion about Jupyter notebooks, and that you did not give him justice.</p>",
      "rawMarkdown": "If you think I posted to get ad, you're wrong.  I hesitated posting my comment as I knew it could be misinterpreted that way.\n\nAnyway, my point was that Pavel's talk is way more interesting than just his opinion about Jupyter notebooks, and that you did not give him justice.",
      "votes": null
    },
    {
      "id": "488198",
      "postDate": "03/12/2019 05:00:24",
      "content": "<p>Thanks for sharing, happy to listen to Grand Masters sharing their wisdom.</p>",
      "rawMarkdown": "Thanks for sharing, happy to listen to Grand Masters sharing their wisdom.",
      "votes": null
    },
    {
      "id": "488237",
      "postDate": "03/12/2019 06:34:59",
      "content": "<p>always a pleasure to see the flesh and bone behind these usernames :) Thanks for posting</p>",
      "rawMarkdown": "always a pleasure to see the flesh and bone behind these usernames :) Thanks for posting",
      "votes": null
    },
    {
      "id": "490182",
      "postDate": "03/14/2019 10:55:41",
      "content": "<p>I watched your talk at kaggle days (<a href=\"https://www.youtube.com/watch?v=fH_FiquKhiI\">https://www.youtube.com/watch?v=fH_FiquKhiI</a>) and I learnt so much from you. Now, I have some idea about how to survive a private LB shakeup ! Thanks CPMP !</p>",
      "rawMarkdown": "I watched your talk at kaggle days (https://www.youtube.com/watch?v=fH_FiquKhiI) and I learnt so much from you. Now, I have some idea about how to survive a private LB shakeup ! Thanks CPMP !",
      "votes": null
    },
    {
      "id": "491933",
      "postDate": "03/16/2019 12:05:51",
      "content": "<p><a href=\"/nyleve\">@nyleve</a> You're welcome. You can see related videos and hands on practise sessions <a href=\"https://www.youtube.com/user/kaggledotcom\">https://www.youtube.com/user/kaggledotcom</a></p>",
      "rawMarkdown": "nyleve You're welcome. You can see related videos and hands on practise sessions https://www.youtube.com/user/kaggledotcom",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 487750,
      "author_name": "vjcalling",
      "author_url": "",
      "post_date": "03/11/2019 12:28:58",
      "content": "<p>Hi <a href=\"/cyberia\">@cyberia</a> !</p>\n\n<p>Thanks for sharing this!</p>\n\n<p>Lots of interesting stuff is available on <a href=\"https://kaggledays.com/\">https://kaggledays.com/</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 487785,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "03/11/2019 13:16:28",
      "content": "<blockquote>\n  <p>main tip: stay away from jupyter notebooks</p>\n</blockquote>\n\n<p>Well, if this is the main takeaway you get, then it may be worth viewing it again ;)</p>",
      "votes": null,
      "replies": [
        {
          "id": 487812,
          "author_name": "cyberia",
          "author_url": "",
          "post_date": "03/11/2019 13:46:29",
          "content": "<p>JFP, main tip is from extended video description I just wrote it as it is. Of course there are so many things to learn from Pavel's talk. I closely follow your work, it is inspiring and awesome. Ah now I see your game :)))), here is the link to another awesome talk by CPMP at Kaggle Days Paris <a href=\"https://www.youtube.com/watch?v=fH_FiquKhiI\">https://www.youtube.com/watch?v=fH_FiquKhiI</a>\nIts good to hear your talk which includes Gaussian processes and sampling bias. I always appreciate the beauty of gaussian models, myself being from bayesian camp.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 487858,
          "author_name": "cpmpml",
          "author_url": "",
          "post_date": "03/11/2019 14:53:08",
          "content": "<p>If you think I posted to get ad, you're wrong.  I hesitated posting my comment as I knew it could be misinterpreted that way.</p>\n\n<p>Anyway, my point was that Pavel's talk is way more interesting than just his opinion about Jupyter notebooks, and that you did not give him justice.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 488198,
          "author_name": "subrahmanyamv",
          "author_url": "",
          "post_date": "03/12/2019 05:00:24",
          "content": "<p>Thanks for sharing, happy to listen to Grand Masters sharing their wisdom.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 490182,
          "author_name": "tarunpaparaju",
          "author_url": "",
          "post_date": "03/14/2019 10:55:41",
          "content": "<p>I watched your talk at kaggle days (<a href=\"https://www.youtube.com/watch?v=fH_FiquKhiI\">https://www.youtube.com/watch?v=fH_FiquKhiI</a>) and I learnt so much from you. Now, I have some idea about how to survive a private LB shakeup ! Thanks CPMP !</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 488237,
      "author_name": "nyleve",
      "author_url": "",
      "post_date": "03/12/2019 06:34:59",
      "content": "<p>always a pleasure to see the flesh and bone behind these usernames :) Thanks for posting</p>",
      "votes": null,
      "replies": [
        {
          "id": 491933,
          "author_name": "cyberia",
          "author_url": "",
          "post_date": "03/16/2019 12:05:51",
          "content": "<p><a href=\"/nyleve\">@nyleve</a> You're welcome. You can see related videos and hands on practise sessions <a href=\"https://www.youtube.com/user/kaggledotcom\">https://www.youtube.com/user/kaggledotcom</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "487741": "Recently a top notch Grandmaster Pavel Ostyakov shared some tricks about winning Kaggle competitions. I found it must a watch, hope you will agree (main tip: stay away from jupyter notebooks). https://www.youtube.com/watch?v=oxikDxBzKUU\n\n**Edit**\n\nFollowing are the timestamps with topics TLDR\n1:20 Speedup convergence (Reduce Learning Rate at Plateaus, Cyclic Learning Rate, Warmup)\n10:39 How to improve training of NN, speedup convergence, Data Augmentation\n19:16 Example of a well organised pipeline\n21:06 7 tips on winning at Kaggle\n24:48 Main problem of competition, goals",
    "487750": "Hi @cyberia !\n\nThanks for sharing this!\n\nLots of interesting stuff is available on https://kaggledays.com/",
    "487785": "&gt; main tip: stay away from jupyter notebooks\n\nWell, if this is the main takeaway you get, then it may be worth viewing it again ;)",
    "487812": "JFP, main tip is from extended video description I just wrote it as it is. Of course there are so many things to learn from Pavel's talk. I closely follow your work, it is inspiring and awesome. Ah now I see your game :)))), here is the link to another awesome talk by CPMP at Kaggle Days Paris https://www.youtube.com/watch?v=fH_FiquKhiI\nIts good to hear your talk which includes Gaussian processes and sampling bias. I always appreciate the beauty of gaussian models, myself being from bayesian camp.",
    "487858": "If you think I posted to get ad, you're wrong.  I hesitated posting my comment as I knew it could be misinterpreted that way.\n\nAnyway, my point was that Pavel's talk is way more interesting than just his opinion about Jupyter notebooks, and that you did not give him justice.",
    "488198": "Thanks for sharing, happy to listen to Grand Masters sharing their wisdom.",
    "488237": "always a pleasure to see the flesh and bone behind these usernames :) Thanks for posting",
    "490182": "I watched your talk at kaggle days (https://www.youtube.com/watch?v=fH_FiquKhiI) and I learnt so much from you. Now, I have some idea about how to survive a private LB shakeup ! Thanks CPMP !",
    "491933": "nyleve You're welcome. You can see related videos and hands on practise sessions https://www.youtube.com/user/kaggledotcom"
  },
  "source": "meta"
}