{
  "id": 522168,
  "title": "Anyone used a genetic algorithm succesfully?",
  "url": "/competitions/uspto-explainable-ai/discussion/522168",
  "author_name": "",
  "post_date": "2024-07-24T19:04:06.532413900Z",
  "votes": 3,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hey guys! Did anyone implement succesfully a genetic algorithm? In our case, we tried hard but it was timing out all time, even if we tuned it a lot. We got local scores around 0.74-0.75 but could never confirm in the public LB. Curious to know!</p>",
  "messages": [
    {
      "id": "2934844",
      "postDate": "07/24/2024 19:04:06",
      "content": "<p>Hey guys! Did anyone implement succesfully a genetic algorithm? In our case, we tried hard but it was timing out all time, even if we tuned it a lot. We got local scores around 0.74-0.75 but could never confirm in the public LB. Curious to know!</p>",
      "rawMarkdown": "Hey guys! Did anyone implement succesfully a genetic algorithm? In our case, we tried hard but it was timing out all time, even if we tuned it a lot. We got local scores around 0.74-0.75 but could never confirm in the public LB. Curious to know!",
      "votes": null
    },
    {
      "id": "2934918",
      "postDate": "07/24/2024 19:52:31",
      "content": "<p>Yeah, I did at one point. Max LB 0.38. My local score was around the same as yours. I set a time limit per query.</p>",
      "rawMarkdown": "Yeah, I did at one point. Max LB 0.38. My local score was around the same as yours. I set a time limit per query.",
      "votes": null
    },
    {
      "id": "2935029",
      "postDate": "07/24/2024 22:45:56",
      "content": "<p>I wasn’t succesful with the time limit, if tou can share when it finishes, highly appreciated! <a href=\"https://www.kaggle.com/keakohv\" target=\"_blank\">@keakohv</a> </p>",
      "rawMarkdown": "I wasn’t succesful with the time limit, if tou can share when it finishes, highly appreciated! @keakohv",
      "votes": null
    },
    {
      "id": "2935059",
      "postDate": "07/25/2024 00:15:43",
      "content": "<p>Yeah, I find it better than annealing, especially if there are more keywords than the limit<br>\nfeel free to check the implementation: <a href=\"https://www.kaggle.com/code/huanligong/uspto-genetic-algorithm-for-keyword-selection\" target=\"_blank\">code</a></p>",
      "rawMarkdown": "Yeah, I find it better than annealing, especially if there are more keywords than the limit\nfeel free to check the implementation: [code](https://www.kaggle.com/code/huanligong/uspto-genetic-algorithm-for-keyword-selection)",
      "votes": null
    },
    {
      "id": "2935146",
      "postDate": "07/25/2024 02:55:26",
      "content": "<p>I used the pymoo package which allows to set a termination criterion based on time. E.g.:</p>\n<pre><code> = get_termination(, )\n</code></pre>\n<p><a href=\"https://pymoo.org/interface/termination.html\" target=\"_blank\">https://pymoo.org/interface/termination.html</a></p>",
      "rawMarkdown": "I used the pymoo package which allows to set a termination criterion based on time. E.g.:\n\n```\ntermination = get_termination(\"time\", \"00:00:10\")\n```\n\nhttps://pymoo.org/interface/termination.html",
      "votes": null
    },
    {
      "id": "2935297",
      "postDate": "07/25/2024 06:26:19",
      "content": "<p>I also tried genetic algorithm and could not improve score.  But very interesting indeed. </p>",
      "rawMarkdown": "I also tried genetic algorithm and could not improve score.  But very interesting indeed.",
      "votes": null
    },
    {
      "id": "2940820",
      "postDate": "07/30/2024 13:40:18",
      "content": "<p><a href=\"https://www.kaggle.com/octaviograu\" target=\"_blank\">@octaviograu</a>, it's great to see your effort with genetic algorithms in the USPTO competition. They can be tricky, especially with timing issues and tuning challenges. It might be worth revisiting your algorithm's parameters or considering alternative approaches to optimization. For instance, exploring different crossover and mutation strategies, or using parallel processing, could potentially improve performance. Also, sharing more details on your setup might help others provide more targeted advice. <br>\nKeep pushing the boundaries, and best of luck with your ongoing experimentation.</p>",
      "rawMarkdown": "octaviograu, it's great to see your effort with genetic algorithms in the USPTO competition. They can be tricky, especially with timing issues and tuning challenges. It might be worth revisiting your algorithm's parameters or considering alternative approaches to optimization. For instance, exploring different crossover and mutation strategies, or using parallel processing, could potentially improve performance. Also, sharing more details on your setup might help others provide more targeted advice. \nKeep pushing the boundaries, and best of luck with your ongoing experimentation.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2934918,
      "author_name": "keakohv",
      "author_url": "",
      "post_date": "07/24/2024 19:52:31",
      "content": "<p>Yeah, I did at one point. Max LB 0.38. My local score was around the same as yours. I set a time limit per query.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2935029,
          "author_name": "octaviograu",
          "author_url": "",
          "post_date": "07/24/2024 22:45:56",
          "content": "<p>I wasn’t succesful with the time limit, if tou can share when it finishes, highly appreciated! <a href=\"https://www.kaggle.com/keakohv\" target=\"_blank\">@keakohv</a> </p>",
          "votes": null,
          "replies": [
            {
              "id": 2935146,
              "author_name": "keakohv",
              "author_url": "",
              "post_date": "07/25/2024 02:55:26",
              "content": "<p>I used the pymoo package which allows to set a termination criterion based on time. E.g.:</p>\n<pre><code> = get_termination(, )\n</code></pre>\n<p><a href=\"https://pymoo.org/interface/termination.html\" target=\"_blank\">https://pymoo.org/interface/termination.html</a></p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2935059,
      "author_name": "huanligong",
      "author_url": "",
      "post_date": "07/25/2024 00:15:43",
      "content": "<p>Yeah, I find it better than annealing, especially if there are more keywords than the limit<br>\nfeel free to check the implementation: <a href=\"https://www.kaggle.com/code/huanligong/uspto-genetic-algorithm-for-keyword-selection\" target=\"_blank\">code</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2935297,
      "author_name": "crsuthikshnkumar",
      "author_url": "",
      "post_date": "07/25/2024 06:26:19",
      "content": "<p>I also tried genetic algorithm and could not improve score.  But very interesting indeed. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2940820,
      "author_name": "",
      "author_url": "",
      "post_date": "07/30/2024 13:40:18",
      "content": "<p><a href=\"https://www.kaggle.com/octaviograu\" target=\"_blank\">@octaviograu</a>, it's great to see your effort with genetic algorithms in the USPTO competition. They can be tricky, especially with timing issues and tuning challenges. It might be worth revisiting your algorithm's parameters or considering alternative approaches to optimization. For instance, exploring different crossover and mutation strategies, or using parallel processing, could potentially improve performance. Also, sharing more details on your setup might help others provide more targeted advice. <br>\nKeep pushing the boundaries, and best of luck with your ongoing experimentation.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2934844": "Hey guys! Did anyone implement succesfully a genetic algorithm? In our case, we tried hard but it was timing out all time, even if we tuned it a lot. We got local scores around 0.74-0.75 but could never confirm in the public LB. Curious to know!",
    "2934918": "Yeah, I did at one point. Max LB 0.38. My local score was around the same as yours. I set a time limit per query.",
    "2935029": "I wasn’t succesful with the time limit, if tou can share when it finishes, highly appreciated! @keakohv",
    "2935059": "Yeah, I find it better than annealing, especially if there are more keywords than the limit\nfeel free to check the implementation: [code](https://www.kaggle.com/code/huanligong/uspto-genetic-algorithm-for-keyword-selection)",
    "2935146": "I used the pymoo package which allows to set a termination criterion based on time. E.g.:\n\n```\ntermination = get_termination(\"time\", \"00:00:10\")\n```\n\nhttps://pymoo.org/interface/termination.html",
    "2935297": "I also tried genetic algorithm and could not improve score.  But very interesting indeed.",
    "2940820": "octaviograu, it's great to see your effort with genetic algorithms in the USPTO competition. They can be tricky, especially with timing issues and tuning challenges. It might be worth revisiting your algorithm's parameters or considering alternative approaches to optimization. For instance, exploring different crossover and mutation strategies, or using parallel processing, could potentially improve performance. Also, sharing more details on your setup might help others provide more targeted advice. \nKeep pushing the boundaries, and best of luck with your ongoing experimentation."
  },
  "source": "meta"
}