{
  "id": 517018,
  "title": "Fun Competition",
  "url": "/competitions/leash-BELKA/discussion/517018",
  "author_name": "",
  "post_date": "2024-07-04T14:20:46.473645100Z",
  "votes": 18,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi Everyone,</p>\n<p>I have to say, this was a really fun competition for me. Probably the most fun one I have competed in yet! Unfortunately, I had to take a break from it before the metric was updated. I thought this break was going to be temporary, but work and life kept getting in the way and I wasn't able to get back to it.</p>\n<p>That said, even though my submissions will likely not generalize well to the private leaderboard, I had a ton of fun in this competition and I learned a lot. This was also the first competition where I had enough domain knowledge to contribute to the public discussion (at least early on) and I really enjoyed making public notebooks and discussing the problem at hand with you all. The kaggle community is an awesome place to be and I'm always so thankful that everyone is so welcoming and forthcoming with their ideas.</p>\n<p>I like to write up competition summaries for myself to recap the things I learned, so here are a few things I learned or experienced for the first time.</p>\n<p>1) I had never worked with such a large dataset before. This was a ton of fun and put some real constraints on the kinds of experiments I could do either locally or on the kaggle servers. I enjoyed the engineering part of this challenge more than I expected and learning how to manage data types to be as efficient as possible was more enjoyable than I expected.</p>\n<p>2) I also had no hands on experience working with chemical language models before, though I had read plenty about them, and this was a really fun chance to try them out. I was honestly impressed at how well even fairly simple language models did when compared to lightGBM based models built on fingerprints. I really wanted to do more experiments with graph neural networks, but I unfortunately didn't have time.</p>\n<p>3) This competition opened my eyes to all of the wonderful open source tools we have to work with 3D chemical structures. I'm not a computational chemist by formal training or vocation, but I really enjoy learning more about the tools we can use to handle 3D structures and it was nice to learn about some of the open source tools available to do this. Specifically, I really enjoyed the practice getting more acquainted with conformer objects in RDKit.</p>\n<p>I'm really looking forward to seeing what sort of solutions end up getting discussed at the end of this competition. I can't wait to learn more from all of you!</p>",
  "messages": [
    {
      "id": "2904685",
      "postDate": "07/04/2024 14:20:46",
      "content": "<p>Hi Everyone,</p>\n<p>I have to say, this was a really fun competition for me. Probably the most fun one I have competed in yet! Unfortunately, I had to take a break from it before the metric was updated. I thought this break was going to be temporary, but work and life kept getting in the way and I wasn't able to get back to it.</p>\n<p>That said, even though my submissions will likely not generalize well to the private leaderboard, I had a ton of fun in this competition and I learned a lot. This was also the first competition where I had enough domain knowledge to contribute to the public discussion (at least early on) and I really enjoyed making public notebooks and discussing the problem at hand with you all. The kaggle community is an awesome place to be and I'm always so thankful that everyone is so welcoming and forthcoming with their ideas.</p>\n<p>I like to write up competition summaries for myself to recap the things I learned, so here are a few things I learned or experienced for the first time.</p>\n<p>1) I had never worked with such a large dataset before. This was a ton of fun and put some real constraints on the kinds of experiments I could do either locally or on the kaggle servers. I enjoyed the engineering part of this challenge more than I expected and learning how to manage data types to be as efficient as possible was more enjoyable than I expected.</p>\n<p>2) I also had no hands on experience working with chemical language models before, though I had read plenty about them, and this was a really fun chance to try them out. I was honestly impressed at how well even fairly simple language models did when compared to lightGBM based models built on fingerprints. I really wanted to do more experiments with graph neural networks, but I unfortunately didn't have time.</p>\n<p>3) This competition opened my eyes to all of the wonderful open source tools we have to work with 3D chemical structures. I'm not a computational chemist by formal training or vocation, but I really enjoy learning more about the tools we can use to handle 3D structures and it was nice to learn about some of the open source tools available to do this. Specifically, I really enjoyed the practice getting more acquainted with conformer objects in RDKit.</p>\n<p>I'm really looking forward to seeing what sort of solutions end up getting discussed at the end of this competition. I can't wait to learn more from all of you!</p>",
      "rawMarkdown": "Hi Everyone,\n\nI have to say, this was a really fun competition for me. Probably the most fun one I have competed in yet! Unfortunately, I had to take a break from it before the metric was updated. I thought this break was going to be temporary, but work and life kept getting in the way and I wasn't able to get back to it.\n\nThat said, even though my submissions will likely not generalize well to the private leaderboard, I had a ton of fun in this competition and I learned a lot. This was also the first competition where I had enough domain knowledge to contribute to the public discussion (at least early on) and I really enjoyed making public notebooks and discussing the problem at hand with you all. The kaggle community is an awesome place to be and I'm always so thankful that everyone is so welcoming and forthcoming with their ideas.\n\nI like to write up competition summaries for myself to recap the things I learned, so here are a few things I learned or experienced for the first time.\n\n1) I had never worked with such a large dataset before. This was a ton of fun and put some real constraints on the kinds of experiments I could do either locally or on the kaggle servers. I enjoyed the engineering part of this challenge more than I expected and learning how to manage data types to be as efficient as possible was more enjoyable than I expected.\n\n2) I also had no hands on experience working with chemical language models before, though I had read plenty about them, and this was a really fun chance to try them out. I was honestly impressed at how well even fairly simple language models did when compared to lightGBM based models built on fingerprints. I really wanted to do more experiments with graph neural networks, but I unfortunately didn't have time.\n\n3) This competition opened my eyes to all of the wonderful open source tools we have to work with 3D chemical structures. I'm not a computational chemist by formal training or vocation, but I really enjoy learning more about the tools we can use to handle 3D structures and it was nice to learn about some of the open source tools available to do this. Specifically, I really enjoyed the practice getting more acquainted with conformer objects in RDKit.\n\nI'm really looking forward to seeing what sort of solutions end up getting discussed at the end of this competition. I can't wait to learn more from all of you!",
      "votes": null
    },
    {
      "id": "2905422",
      "postDate": "07/04/2024 23:48:23",
      "content": "<p>I think so too.<br>\nThis competition is my first one.<br>\nI learned a lot by other codes and discussions. I appreciate them and rich resources.</p>",
      "rawMarkdown": "I think so too.\nThis competition is my first one.\nI learned a lot by other codes and discussions. I appreciate them and rich resources.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2905422,
      "author_name": "takanorisugita",
      "author_url": "",
      "post_date": "07/04/2024 23:48:23",
      "content": "<p>I think so too.<br>\nThis competition is my first one.<br>\nI learned a lot by other codes and discussions. I appreciate them and rich resources.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2904685": "Hi Everyone,\n\nI have to say, this was a really fun competition for me. Probably the most fun one I have competed in yet! Unfortunately, I had to take a break from it before the metric was updated. I thought this break was going to be temporary, but work and life kept getting in the way and I wasn't able to get back to it.\n\nThat said, even though my submissions will likely not generalize well to the private leaderboard, I had a ton of fun in this competition and I learned a lot. This was also the first competition where I had enough domain knowledge to contribute to the public discussion (at least early on) and I really enjoyed making public notebooks and discussing the problem at hand with you all. The kaggle community is an awesome place to be and I'm always so thankful that everyone is so welcoming and forthcoming with their ideas.\n\nI like to write up competition summaries for myself to recap the things I learned, so here are a few things I learned or experienced for the first time.\n\n1) I had never worked with such a large dataset before. This was a ton of fun and put some real constraints on the kinds of experiments I could do either locally or on the kaggle servers. I enjoyed the engineering part of this challenge more than I expected and learning how to manage data types to be as efficient as possible was more enjoyable than I expected.\n\n2) I also had no hands on experience working with chemical language models before, though I had read plenty about them, and this was a really fun chance to try them out. I was honestly impressed at how well even fairly simple language models did when compared to lightGBM based models built on fingerprints. I really wanted to do more experiments with graph neural networks, but I unfortunately didn't have time.\n\n3) This competition opened my eyes to all of the wonderful open source tools we have to work with 3D chemical structures. I'm not a computational chemist by formal training or vocation, but I really enjoy learning more about the tools we can use to handle 3D structures and it was nice to learn about some of the open source tools available to do this. Specifically, I really enjoyed the practice getting more acquainted with conformer objects in RDKit.\n\nI'm really looking forward to seeing what sort of solutions end up getting discussed at the end of this competition. I can't wait to learn more from all of you!",
    "2905422": "I think so too.\nThis competition is my first one.\nI learned a lot by other codes and discussions. I appreciate them and rich resources."
  },
  "source": "meta"
}