{
  "id": 252991,
  "title": "Competition Wrap Up and CVPPA Submissions",
  "url": "/competitions/sorghum-biomass-prediction/discussion/252991",
  "author_name": "",
  "post_date": "2021-07-14T14:45:33.403878400Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello all, and congratulations to our top finishers! This was an extremely challenging task, and as some competitors noted, it was extremely sensitive to overfitting to the training set. To that end, I'm sharing the final private leaderboard, which differs in its ordering from the public leaderboard:</p>\n<p><img src=\"https://cs.slu.edu/~astylianou/images/sorghum_biomass_predictions.png\" alt=\"Final private leaderboard\"></p>\n<p>The example submission simply predicted the mean end of season dry biomass for every single plot -- an extremely simple, but not egregiously bad baseline. The fact that several competitors significantly outperformed this is very exciting, and indicates that there are approaches that can yield meaningful predictive capabilities for end of season biomass relatively early in the season!</p>\n<p>We hope that competitors will be open to sharing about the approaches they took! At minimum, please consider posting a write up of your approach on this discussion board. But more importantly, I'd like to suggest submitting a write up of your approach to the CVPPA workshop! The deadline is this Friday for both full papers and (perhaps more realistically in this time frame if you've not already started your writeup) extended abstracts which describe your approach: <a href=\"https://cvppa2021.github.io/dates/\" target=\"_blank\">https://cvppa2021.github.io/dates/</a></p>\n<p>I'm looking forward to hearing more about the approaches folks took, and look forward to running versions of this competition again in the future!</p>",
  "messages": [
    {
      "id": "1387964",
      "postDate": "07/14/2021 14:45:33",
      "content": "<p>Hello all, and congratulations to our top finishers! This was an extremely challenging task, and as some competitors noted, it was extremely sensitive to overfitting to the training set. To that end, I'm sharing the final private leaderboard, which differs in its ordering from the public leaderboard:</p>\n<p><img src=\"https://cs.slu.edu/~astylianou/images/sorghum_biomass_predictions.png\" alt=\"Final private leaderboard\"></p>\n<p>The example submission simply predicted the mean end of season dry biomass for every single plot -- an extremely simple, but not egregiously bad baseline. The fact that several competitors significantly outperformed this is very exciting, and indicates that there are approaches that can yield meaningful predictive capabilities for end of season biomass relatively early in the season!</p>\n<p>We hope that competitors will be open to sharing about the approaches they took! At minimum, please consider posting a write up of your approach on this discussion board. But more importantly, I'd like to suggest submitting a write up of your approach to the CVPPA workshop! The deadline is this Friday for both full papers and (perhaps more realistically in this time frame if you've not already started your writeup) extended abstracts which describe your approach: <a href=\"https://cvppa2021.github.io/dates/\" target=\"_blank\">https://cvppa2021.github.io/dates/</a></p>\n<p>I'm looking forward to hearing more about the approaches folks took, and look forward to running versions of this competition again in the future!</p>",
      "rawMarkdown": "Hello all, and congratulations to our top finishers! This was an extremely challenging task, and as some competitors noted, it was extremely sensitive to overfitting to the training set. To that end, I'm sharing the final private leaderboard, which differs in its ordering from the public leaderboard:\n\n![Final private leaderboard](https://cs.slu.edu/~astylianou/images/sorghum_biomass_predictions.png)\n\nThe example submission simply predicted the mean end of season dry biomass for every single plot -- an extremely simple, but not egregiously bad baseline. The fact that several competitors significantly outperformed this is very exciting, and indicates that there are approaches that can yield meaningful predictive capabilities for end of season biomass relatively early in the season!\n\nWe hope that competitors will be open to sharing about the approaches they took! At minimum, please consider posting a write up of your approach on this discussion board. But more importantly, I'd like to suggest submitting a write up of your approach to the CVPPA workshop! The deadline is this Friday for both full papers and (perhaps more realistically in this time frame if you've not already started your writeup) extended abstracts which describe your approach: [https://cvppa2021.github.io/dates/](https://cvppa2021.github.io/dates/)\n\nI'm looking forward to hearing more about the approaches folks took, and look forward to running versions of this competition again in the future!",
      "votes": null
    },
    {
      "id": "1388045",
      "postDate": "07/14/2021 16:00:51",
      "content": "<p>Thanks again for putting this on Abby!</p>\n<p>The private leaderboard is really interesting. Since the example submission is just predicting the population mean (coefficient of determination = 0.0), I would assume that randomly sampled models would be normally distributed with the mean at that point. Eyeballing the private leaderboard, it looks like that's exactly what happened so it's tough to say whether any of the models have any predictive power at all.</p>",
      "rawMarkdown": "Thanks again for putting this on Abby!\n\nThe private leaderboard is really interesting. Since the example submission is just predicting the population mean (coefficient of determination = 0.0), I would assume that randomly sampled models would be normally distributed with the mean at that point. Eyeballing the private leaderboard, it looks like that's exactly what happened so it's tough to say whether any of the models have any predictive power at all.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1388045,
      "author_name": "jubbens",
      "author_url": "",
      "post_date": "07/14/2021 16:00:51",
      "content": "<p>Thanks again for putting this on Abby!</p>\n<p>The private leaderboard is really interesting. Since the example submission is just predicting the population mean (coefficient of determination = 0.0), I would assume that randomly sampled models would be normally distributed with the mean at that point. Eyeballing the private leaderboard, it looks like that's exactly what happened so it's tough to say whether any of the models have any predictive power at all.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1387964": "Hello all, and congratulations to our top finishers! This was an extremely challenging task, and as some competitors noted, it was extremely sensitive to overfitting to the training set. To that end, I'm sharing the final private leaderboard, which differs in its ordering from the public leaderboard:\n\n![Final private leaderboard](https://cs.slu.edu/~astylianou/images/sorghum_biomass_predictions.png)\n\nThe example submission simply predicted the mean end of season dry biomass for every single plot -- an extremely simple, but not egregiously bad baseline. The fact that several competitors significantly outperformed this is very exciting, and indicates that there are approaches that can yield meaningful predictive capabilities for end of season biomass relatively early in the season!\n\nWe hope that competitors will be open to sharing about the approaches they took! At minimum, please consider posting a write up of your approach on this discussion board. But more importantly, I'd like to suggest submitting a write up of your approach to the CVPPA workshop! The deadline is this Friday for both full papers and (perhaps more realistically in this time frame if you've not already started your writeup) extended abstracts which describe your approach: [https://cvppa2021.github.io/dates/](https://cvppa2021.github.io/dates/)\n\nI'm looking forward to hearing more about the approaches folks took, and look forward to running versions of this competition again in the future!",
    "1388045": "Thanks again for putting this on Abby!\n\nThe private leaderboard is really interesting. Since the example submission is just predicting the population mean (coefficient of determination = 0.0), I would assume that randomly sampled models would be normally distributed with the mean at that point. Eyeballing the private leaderboard, it looks like that's exactly what happened so it's tough to say whether any of the models have any predictive power at all."
  },
  "source": "meta"
}