{
  "id": 220734,
  "title": "Takeaway from this (my) first competition!!",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/220734",
  "author_name": "",
  "post_date": "2021-02-19T10:35:29.754231700Z",
  "votes": 6,
  "comment_count": 1,
  "views": 0,
  "content": "<p>During the Christmas vacation, I decided to spend my free time at the Kaggle competition. I chose cassava-leaf-disease-classification as I thought visuals will more fun to learn and compete. <br>\nMy only focus was to learn and I'm glad after ending the competition, I can proudly say that I learned a lot(more than any other platform). Though I'm not on the medal-list but I'm taking my learning very seriously 😃 Well, here is the summary</p>\n<blockquote>\n  <ol>\n  <li>Write a generic code and keep experimenting with different models</li>\n  <li>Log all metrics of the models and also prepare a spreadsheet (it will help in final submission)</li>\n  <li>Trust on your CV but also don't underestimate public LB. In other word keep checking the correlation of your models score and public LB</li>\n  <li>If you are stuck somewhere or not getting the concepts, don't hesitate to ask the question. Kagglers never judge you and that's the beauty of this platform.</li>\n  <li>Be kind and appreciate other works by upvoting especially if you have used their ideas or code.</li>\n  <li>Form a team if you have less time to devote to the competition. It will help productivity and more experiments(I committed this mistake but honestly it is also hard to find a like-minded partner 😏)</li>\n  <li>Most important, don't let go of all of your experiments go, and here points 2 come handly because it's time for ensembling them all and squeezing every drop of the model's accuracy.</li>\n  <li>Make a wise submission else it will cost you to take your medals home 😃 I also missed it to recognize (the right submission could have given me at least landing in the top 12% rather than 32%)</li>\n  <li>Trust in your work. It doesn't matter you lose or win but you always learn.</li>\n  <li>Finally, organize your code and experiment and write a blog to document them. Next task for me 😃 </li>\n  </ol>\n</blockquote>\n<p>In this journey, the first time I used PyTorch, experimented with different SOTA models, image feature techniques, read many papers (including SnapMix) and learned a new way of thinking to solve a problem. </p>\n<p><strong>Congratulations to all winners, participants and once again thanks to all of you for this journey.</strong></p>",
  "messages": [
    {
      "id": "1210303",
      "postDate": "02/19/2021 10:35:29",
      "content": "<p>During the Christmas vacation, I decided to spend my free time at the Kaggle competition. I chose cassava-leaf-disease-classification as I thought visuals will more fun to learn and compete. <br>\nMy only focus was to learn and I'm glad after ending the competition, I can proudly say that I learned a lot(more than any other platform). Though I'm not on the medal-list but I'm taking my learning very seriously 😃 Well, here is the summary</p>\n<blockquote>\n  <ol>\n  <li>Write a generic code and keep experimenting with different models</li>\n  <li>Log all metrics of the models and also prepare a spreadsheet (it will help in final submission)</li>\n  <li>Trust on your CV but also don't underestimate public LB. In other word keep checking the correlation of your models score and public LB</li>\n  <li>If you are stuck somewhere or not getting the concepts, don't hesitate to ask the question. Kagglers never judge you and that's the beauty of this platform.</li>\n  <li>Be kind and appreciate other works by upvoting especially if you have used their ideas or code.</li>\n  <li>Form a team if you have less time to devote to the competition. It will help productivity and more experiments(I committed this mistake but honestly it is also hard to find a like-minded partner 😏)</li>\n  <li>Most important, don't let go of all of your experiments go, and here points 2 come handly because it's time for ensembling them all and squeezing every drop of the model's accuracy.</li>\n  <li>Make a wise submission else it will cost you to take your medals home 😃 I also missed it to recognize (the right submission could have given me at least landing in the top 12% rather than 32%)</li>\n  <li>Trust in your work. It doesn't matter you lose or win but you always learn.</li>\n  <li>Finally, organize your code and experiment and write a blog to document them. Next task for me 😃 </li>\n  </ol>\n</blockquote>\n<p>In this journey, the first time I used PyTorch, experimented with different SOTA models, image feature techniques, read many papers (including SnapMix) and learned a new way of thinking to solve a problem. </p>\n<p><strong>Congratulations to all winners, participants and once again thanks to all of you for this journey.</strong></p>",
      "rawMarkdown": "During the Christmas vacation, I decided to spend my free time at the Kaggle competition. I chose cassava-leaf-disease-classification as I thought visuals will more fun to learn and compete. \nMy only focus was to learn and I'm glad after ending the competition, I can proudly say that I learned a lot(more than any other platform). Though I'm not on the medal-list but I'm taking my learning very seriously 😃 Well, here is the summary\n\n> 1. \t Write a generic code and keep experimenting with different models\n2. \tLog all metrics of the models and also prepare a spreadsheet (it will help in final submission)\n3.\tTrust on your CV but also don't underestimate public LB. In other word keep checking the correlation of your models score and public LB\n4.\tIf you are stuck somewhere or not getting the concepts, don't hesitate to ask the question. Kagglers never judge you and that's the beauty of this platform.\n5.\tBe kind and appreciate other works by upvoting especially if you have used their ideas or code.\n6.\tForm a team if you have less time to devote to the competition. It will help productivity and more experiments(I committed this mistake but honestly it is also hard to find a like-minded partner 😏)\n7.\tMost important, don't let go of all of your experiments go, and here points 2 come handly because it's time for ensembling them all and squeezing every drop of the model's accuracy.\n8.\tMake a wise submission else it will cost you to take your medals home 😃 I also missed it to recognize (the right submission could have given me at least landing in the top 12% rather than 32%)\n9.\tTrust in your work. It doesn't matter you lose or win but you always learn.\n10.\t Finally, organize your code and experiment and write a blog to document them. Next task for me 😃 \n\t\nIn this journey, the first time I used PyTorch, experimented with different SOTA models, image feature techniques, read many papers (including SnapMix) and learned a new way of thinking to solve a problem. \n\n**Congratulations to all winners, participants and once again thanks to all of you for this journey.**",
      "votes": null
    },
    {
      "id": "1210412",
      "postDate": "02/19/2021 12:14:21",
      "content": "<p>nice idea man, I should also do a write-up of everything I have learned from this competetion.</p>",
      "rawMarkdown": "nice idea man, I should also do a write-up of everything I have learned from this competetion.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1210412,
      "author_name": "mohneesh7",
      "author_url": "",
      "post_date": "02/19/2021 12:14:21",
      "content": "<p>nice idea man, I should also do a write-up of everything I have learned from this competetion.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1210303": "During the Christmas vacation, I decided to spend my free time at the Kaggle competition. I chose cassava-leaf-disease-classification as I thought visuals will more fun to learn and compete. \nMy only focus was to learn and I'm glad after ending the competition, I can proudly say that I learned a lot(more than any other platform). Though I'm not on the medal-list but I'm taking my learning very seriously 😃 Well, here is the summary\n\n> 1. \t Write a generic code and keep experimenting with different models\n2. \tLog all metrics of the models and also prepare a spreadsheet (it will help in final submission)\n3.\tTrust on your CV but also don't underestimate public LB. In other word keep checking the correlation of your models score and public LB\n4.\tIf you are stuck somewhere or not getting the concepts, don't hesitate to ask the question. Kagglers never judge you and that's the beauty of this platform.\n5.\tBe kind and appreciate other works by upvoting especially if you have used their ideas or code.\n6.\tForm a team if you have less time to devote to the competition. It will help productivity and more experiments(I committed this mistake but honestly it is also hard to find a like-minded partner 😏)\n7.\tMost important, don't let go of all of your experiments go, and here points 2 come handly because it's time for ensembling them all and squeezing every drop of the model's accuracy.\n8.\tMake a wise submission else it will cost you to take your medals home 😃 I also missed it to recognize (the right submission could have given me at least landing in the top 12% rather than 32%)\n9.\tTrust in your work. It doesn't matter you lose or win but you always learn.\n10.\t Finally, organize your code and experiment and write a blog to document them. Next task for me 😃 \n\t\nIn this journey, the first time I used PyTorch, experimented with different SOTA models, image feature techniques, read many papers (including SnapMix) and learned a new way of thinking to solve a problem. \n\n**Congratulations to all winners, participants and once again thanks to all of you for this journey.**",
    "1210412": "nice idea man, I should also do a write-up of everything I have learned from this competetion."
  },
  "source": "meta"
}