In the competition, each sample corresponds to a specific target name. Is it recommended to train a separate model for each biomass component (Dry_Green_g, Dry_Dead_g, Dry_Clover_g, GDM_g, Dry_Total_g), or is it more efficient to train a single model capable of predicting all five components simultaneously?
", "messages": [ { "id": 3308979, "postDate": "2025-10-30T16:36:06.050Z", "content": "In the competition, each sample corresponds to a specific target name. Is it recommended to train a separate model for each biomass component (Dry_Green_g, Dry_Dead_g, Dry_Clover_g, GDM_g, Dry_Total_g), or is it more efficient to train a single model capable of predicting all five components simultaneously?
", "rawMarkdown": "In the competition, each sample corresponds to a specific target name. Is it recommended to train a separate model for each biomass component (Dry_Green_g, Dry_Dead_g, Dry_Clover_g, GDM_g, Dry_Total_g), or is it more efficient to train a single model capable of predicting all five components simultaneously?", "votes": 2 }, { "id": 3309095, "postDate": "2025-10-31T00:04:59.627Z", "content": "Training time is not going to be that long - so I think you can go with the process that you believe gives the best result.
", "rawMarkdown": "Training time is not going to be that long - so I think you can go with the process that you believe gives the best result.", "votes": 1 }, { "id": 3311806, "postDate": "2025-11-05T20:12:31.653Z", "content": "Aren’t GDM_g and Dry_Total_g strictly the addition of other labels?
\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g
\nGDM_g = Dry_Green_g + Dry_Clover_g
\nSo I think we will only need to predict for Dry_Green_g, Dry_Dead_g and Dry_Clover_g?
", "rawMarkdown": "Aren’t GDM_g and Dry_Total_g strictly the addition of other labels?\n\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g\n\nGDM_g = Dry_Green_g + Dry_Clover_g \n\nSo I think we will only need to predict for Dry_Green_g, Dry_Dead_g and Dry_Clover_g?\n", "replies": [ { "id": 3311852, "postDate": "2025-11-05T23:26:36.187Z", "content": "I assumed predicting the total amount directly would be more accurate than summing the predictions of the three components. The accuracy of the total amount is more important than that of the individual components (as reflected in the evaluation metrics), so if my assumption is correct, predicting the total amount is also necessary.
", "rawMarkdown": "I assumed predicting the total amount directly would be more accurate than summing the predictions of the three components. The accuracy of the total amount is more important than that of the individual components (as reflected in the evaluation metrics), so if my assumption is correct, predicting the total amount is also necessary.", "replies": [ { "id": 3311862, "postDate": "2025-11-06T00:17:56.850Z", "content": "I see. But wouldn’t it not make sense if the sum of the individual components doesn’t equal the total amount?
", "rawMarkdown": "I see. But wouldn’t it not make sense if the sum of the individual components doesn’t equal the total amount?" }, { "id": 3311919, "postDate": "2025-11-06T03:42:13.840Z", "content": "They should be the same.
", "rawMarkdown": "They should be the same. " }, { "id": 3312890, "postDate": "2025-11-08T07:01:46.797Z", "content": "Okay but there is no way to guarantee that if you are predicting each label with a different model or treating them as different targets.
", "rawMarkdown": "Okay but there is no way to guarantee that if you are predicting each label with a different model or treating them as different targets." }, { "id": 3312892, "postDate": "2025-11-08T07:14:59.970Z", "content": "Yes you're right, I don't think predicting independently can be accurate enough. There has been some discussion on this topic, e.g., https://www.kaggle.com/competitions/csiro-biomass/discussion/614912 , you can refer to them.
", "rawMarkdown": "Yes you're right, I don't think predicting independently can be accurate enough. There has been some discussion on this topic, e.g., https://www.kaggle.com/competitions/csiro-biomass/discussion/614912 , you can refer to them." }, { "id": 3312899, "postDate": "2025-11-08T07:31:09.953Z", "content": "I see. That is a pretty interesting trick. But even then, the relationships
\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g
\nGDM_g = Dry_Green_g + Dry_Clover_g
\nare not enforced. Does that mean we can just emphasize Dry_Total_g even if that comes at the loss of… logical consistency?
", "rawMarkdown": "I see. That is a pretty interesting trick. But even then, the relationships\n\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g\n\nGDM_g = Dry_Green_g + Dry_Clover_g\n\nare not enforced. Does that mean we can just emphasize Dry_Total_g even if that comes at the loss of... logical consistency?" }, { "id": 3313423, "postDate": "2025-11-09T08:43:36.127Z", "content": "In model evaluation, only the weighted R-squared metric is considered; logical consistency is not an evaluation criterion.
", "rawMarkdown": "In model evaluation, only the weighted R-squared metric is considered; logical consistency is not an evaluation criterion." }, { "id": 3313435, "postDate": "2025-11-09T09:07:17.350Z", "content": "I see. Thank you for the clarification
", "rawMarkdown": "I see. Thank you for the clarification\n" } ] } ] }, { "id": 3309166, "postDate": "2025-10-31T05:01:30.817Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 3309236, "postDate": "2025-10-31T09:24:42.457Z", "content": "What makes you think so? \nThese are strongly correlated targets, i.e. dry total should be roughly the sum of the separate dry components
", "rawMarkdown": "What makes you think so? \nThese are strongly correlated targets, i.e. dry total should be roughly the sum of the separate dry components", "replies": [ { "id": 3310592, "postDate": "2025-11-03T08:13:02.083Z", "content": "I guess we’ll have to find out through experiments.
", "rawMarkdown": "I guess we’ll have to find out through experiments.", "votes": 1 } ] } ] } ], "comments": [ { "id": 3309095, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2025-10-31T00:04:59.627000", "content": "Training time is not going to be that long - so I think you can go with the process that you believe gives the best result.
", "votes": 1, "replies": [] }, { "id": 3311806, "author_name": "porpita", "author_url": "", "post_date": "2025-11-05T20:12:31.653000", "content": "Aren’t GDM_g and Dry_Total_g strictly the addition of other labels?
\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g
\nGDM_g = Dry_Green_g + Dry_Clover_g
\nSo I think we will only need to predict for Dry_Green_g, Dry_Dead_g and Dry_Clover_g?
", "votes": 0, "replies": [ { "id": 3311852, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-05T23:26:36.187000", "content": "I assumed predicting the total amount directly would be more accurate than summing the predictions of the three components. The accuracy of the total amount is more important than that of the individual components (as reflected in the evaluation metrics), so if my assumption is correct, predicting the total amount is also necessary.
", "votes": 0, "replies": [ { "id": 3311862, "author_name": "porpita", "author_url": "", "post_date": "2025-11-06T00:17:56.850000", "content": "I see. But wouldn’t it not make sense if the sum of the individual components doesn’t equal the total amount?
", "votes": 0, "replies": [] }, { "id": 3311919, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-06T03:42:13.840000", "content": "They should be the same.
", "votes": 0, "replies": [] }, { "id": 3312890, "author_name": "porpita", "author_url": "", "post_date": "2025-11-08T07:01:46.797000", "content": "Okay but there is no way to guarantee that if you are predicting each label with a different model or treating them as different targets.
", "votes": 0, "replies": [] }, { "id": 3312892, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-08T07:14:59.970000", "content": "Yes you're right, I don't think predicting independently can be accurate enough. There has been some discussion on this topic, e.g., https://www.kaggle.com/competitions/csiro-biomass/discussion/614912 , you can refer to them.
", "votes": 0, "replies": [] }, { "id": 3312899, "author_name": "porpita", "author_url": "", "post_date": "2025-11-08T07:31:09.953000", "content": "I see. That is a pretty interesting trick. But even then, the relationships
\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g
\nGDM_g = Dry_Green_g + Dry_Clover_g
\nare not enforced. Does that mean we can just emphasize Dry_Total_g even if that comes at the loss of… logical consistency?
", "votes": 0, "replies": [] }, { "id": 3313423, "author_name": "Qiyu Liao", "author_url": "", "post_date": "2025-11-09T08:43:36.127000", "content": "In model evaluation, only the weighted R-squared metric is considered; logical consistency is not an evaluation criterion.
", "votes": 0, "replies": [] }, { "id": 3313435, "author_name": "porpita", "author_url": "", "post_date": "2025-11-09T09:07:17.350000", "content": "I see. Thank you for the clarification
", "votes": 0, "replies": [] } ] } ] }, { "id": 3309166, "author_name": "", "author_url": "", "post_date": "2025-10-31T05:01:30.817000", "content": "", "votes": 0, "replies": [ { "id": 3309236, "author_name": "fnands", "author_url": "", "post_date": "2025-10-31T09:24:42.457000", "content": "What makes you think so? \nThese are strongly correlated targets, i.e. dry total should be roughly the sum of the separate dry components
", "votes": 0, "replies": [ { "id": 3310592, "author_name": "Aic", "author_url": "", "post_date": "2025-11-03T08:13:02.083000", "content": "I guess we’ll have to find out through experiments.
", "votes": 1, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3308979": "In the competition, each sample corresponds to a specific target name. Is it recommended to train a separate model for each biomass component (Dry_Green_g, Dry_Dead_g, Dry_Clover_g, GDM_g, Dry_Total_g), or is it more efficient to train a single model capable of predicting all five components simultaneously?", "3309095": "Training time is not going to be that long - so I think you can go with the process that you believe gives the best result.", "3311806": "Aren’t GDM_g and Dry_Total_g strictly the addition of other labels?\n\nDry_Total_g = Dry_Green_g + Dry_Clover_g + Dry_Dead_g\n\nGDM_g = Dry_Green_g + Dry_Clover_g \n\nSo I think we will only need to predict for Dry_Green_g, Dry_Dead_g and Dry_Clover_g?\n", "3309166": "" } }