The CSIRO dataset seems to be pretty small in comparison to other competitions. Any suggestions on how to make best of this, or any other datasets that have proven useful.\nAny help is appreciated.\nThank you :p
", "messages": [ { "id": "3368604", "postDate": "12/09/2025 11:28:11", "content": "The CSIRO dataset seems to be pretty small in comparison to other competitions. Any suggestions on how to make best of this, or any other datasets that have proven useful.\nAny help is appreciated.\nThank you :p
", "rawMarkdown": "The CSIRO dataset seems to be pretty small in comparison to other competitions. Any suggestions on how to make best of this, or any other datasets that have proven useful.\nAny help is appreciated.\nThank you :p", "votes": null }, { "id": "3368930", "postDate": "12/09/2025 16:14:18", "content": "I have an augmented crop dataset: https://www.kaggle.com/competitions/csiro-biomass/discussion/654028
", "rawMarkdown": "I have an augmented crop dataset: https://www.kaggle.com/competitions/csiro-biomass/discussion/654028", "votes": null }, { "id": "3369020", "postDate": "12/09/2025 17:19:46", "content": "There are two similar datasets that folks have mentioned in discussion topics. They were conducted in similar fashion but you need to do a decent amount of work to get them in the same format and content as our data. I am working on that this week. They did not include dead material in the measurements, which happens to be the poorest performing label for my models. 99% chance that the measurements they made are correlated but not exactly the same as our data. I assume someone has already tested one or both of these data sets - maybe they will share the experience.
\nWith some success you can get one of the LLM's to generate decent looking images, Google Pro will even give you an estimate of the biomass :) It was painful to do and not sure I found value yet but it's still on my todo list. Same story about measurements correlated but not exactly the same.
\nBiggest issue I see with use of external datasets is that we are not attempting to model the TRUTH, but rather attempting to model the Test data set. In Kaggle competitions these two are often very different :)
", "rawMarkdown": "There are two similar datasets that folks have mentioned in discussion topics. They were conducted in similar fashion but you need to do a decent amount of work to get them in the same format and content as our data. I am working on that this week. They did not include dead material in the measurements, which happens to be the poorest performing label for my models. 99% chance that the measurements they made are correlated but not exactly the same as our data. I assume someone has already tested one or both of these data sets - maybe they will share the experience.\n\nWith some success you can get one of the LLM's to generate decent looking images, Google Pro will even give you an estimate of the biomass :) It was painful to do and not sure I found value yet but it's still on my todo list. Same story about measurements correlated but not exactly the same.\n\nBiggest issue I see with use of external datasets is that we are not attempting to model the TRUTH, but rather attempting to model the Test data set. In Kaggle competitions these two are often very different :)", "votes": null }, { "id": "3369028", "postDate": "12/09/2025 17:23:52", "content": "Thanks for the response. \nI’ve found some leaks in my pipeline and hopefully after that’s sorted, I hope I can achieve what I want without additional datasets.
", "rawMarkdown": "Thanks for the response. \nI’ve found some leaks in my pipeline and hopefully after that’s sorted, I hope I can achieve what I want without additional datasets.", "votes": null }, { "id": "3369030", "postDate": "12/09/2025 17:24:19", "content": "What LB and CV did you get with that?
", "rawMarkdown": "What LB and CV did you get with that?", "votes": null } ], "comments": [ { "id": 3368930, "author_name": "laodriverayu", "author_url": "", "post_date": "12/09/2025 16:14:18", "content": "I have an augmented crop dataset: https://www.kaggle.com/competitions/csiro-biomass/discussion/654028
", "votes": null, "replies": [ { "id": 3369030, "author_name": "agastyasabarad", "author_url": "", "post_date": "12/09/2025 17:24:19", "content": "What LB and CV did you get with that?
", "votes": null, "replies": [] } ] }, { "id": 3369020, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/09/2025 17:19:46", "content": "There are two similar datasets that folks have mentioned in discussion topics. They were conducted in similar fashion but you need to do a decent amount of work to get them in the same format and content as our data. I am working on that this week. They did not include dead material in the measurements, which happens to be the poorest performing label for my models. 99% chance that the measurements they made are correlated but not exactly the same as our data. I assume someone has already tested one or both of these data sets - maybe they will share the experience.
\nWith some success you can get one of the LLM's to generate decent looking images, Google Pro will even give you an estimate of the biomass :) It was painful to do and not sure I found value yet but it's still on my todo list. Same story about measurements correlated but not exactly the same.
\nBiggest issue I see with use of external datasets is that we are not attempting to model the TRUTH, but rather attempting to model the Test data set. In Kaggle competitions these two are often very different :)
", "votes": null, "replies": [ { "id": 3369028, "author_name": "agastyasabarad", "author_url": "", "post_date": "12/09/2025 17:23:52", "content": "Thanks for the response. \nI’ve found some leaks in my pipeline and hopefully after that’s sorted, I hope I can achieve what I want without additional datasets.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3368604": "The CSIRO dataset seems to be pretty small in comparison to other competitions. Any suggestions on how to make best of this, or any other datasets that have proven useful.\nAny help is appreciated.\nThank you :p", "3368930": "I have an augmented crop dataset: https://www.kaggle.com/competitions/csiro-biomass/discussion/654028", "3369020": "There are two similar datasets that folks have mentioned in discussion topics. They were conducted in similar fashion but you need to do a decent amount of work to get them in the same format and content as our data. I am working on that this week. They did not include dead material in the measurements, which happens to be the poorest performing label for my models. 99% chance that the measurements they made are correlated but not exactly the same as our data. I assume someone has already tested one or both of these data sets - maybe they will share the experience.\n\nWith some success you can get one of the LLM's to generate decent looking images, Google Pro will even give you an estimate of the biomass :) It was painful to do and not sure I found value yet but it's still on my todo list. Same story about measurements correlated but not exactly the same.\n\nBiggest issue I see with use of external datasets is that we are not attempting to model the TRUTH, but rather attempting to model the Test data set. In Kaggle competitions these two are often very different :)", "3369028": "Thanks for the response. \nI’ve found some leaks in my pipeline and hopefully after that’s sorted, I hope I can achieve what I want without additional datasets.", "3369030": "What LB and CV did you get with that?" }, "source": "meta" }