I verified all the target values of train.csv and found that:
\nBut they are not exactly equal (such as ID1035947949), I would like to know the reason for this problem and whether there is a corresponding solution (how many decimal places should be kept in the submission?)
", "messages": [ { "id": "3309671", "postDate": "11/01/2025 06:25:42", "content": "I verified all the target values of train.csv and found that:
\nBut they are not exactly equal (such as ID1035947949), I would like to know the reason for this problem and whether there is a corresponding solution (how many decimal places should be kept in the submission?)
", "rawMarkdown": "I verified all the target values of train.csv and found that:\n1. Total = Green + Clover + Dead\n2. GDM = Green + Clover\n\n[Target Check](https://www.kaggle.com/code/unsayin9/target-check?scriptVersionId=272560832)\n\nBut they are not exactly equal (such as ID1035947949), I would like to know the reason for this problem and whether there is a corresponding solution (how many decimal places should be kept in the submission?)", "votes": null }, { "id": "3309708", "postDate": "11/01/2025 07:54:42", "content": "round(a + b) is not necessarily strictly equal to round(a) + round(b), which should be the cause of this error.
\nFor example, ID1035947949, Total = 34.1844, Green + Clover + Dead = 34.1843 (-e-4);
\nGDM = 10.9605, Green + Clover = 10.9604 (-e-4).
\nThis has a negligible impact on the evaluation.
\nThere are no specific requirements for precision when submitting. You can retain the original precision of the model output without rounding it.
", "rawMarkdown": "round(a + b) is not necessarily strictly equal to round(a) + round(b), which should be the cause of this error.\n\nFor example, ID1035947949, Total = 34.1844, Green + Clover + Dead = 34.1843 (-e-4); \n\nGDM = 10.9605, Green + Clover = 10.9604 (-e-4).\n\nThis has a negligible impact on the evaluation.\n\nThere are no specific requirements for precision when submitting. You can retain the original precision of the model output without rounding it.", "votes": null }, { "id": "3380846", "postDate": "12/23/2025 07:22:42", "content": "\"Total = Green + Clover + Dead. What is the maximum value of Total? Were the shooting equipment and camera angles kept consistent?
\nI wonder if the value of Total might be affected by the distance of the camera. For instance, when the camera zooms out, the captured area is larger compared to zooming in. Does this mean that Total increases as the camera moves further away?\"
", "rawMarkdown": "\"Total = Green + Clover + Dead. What is the maximum value of Total? Were the shooting equipment and camera angles kept consistent? \n\nI wonder if the value of Total might be affected by the distance of the camera. For instance, when the camera zooms out, the captured area is larger compared to zooming in. Does this mean that Total increases as the camera moves further away?\"", "votes": null }, { "id": "3381030", "postDate": "12/23/2025 15:58:47", "content": "Reading the paper it would seem that the area measured was bounded by a frame. You can often see it's shadow or one or more of the actual edges in the images. Everything inside the frame was harvested.
\nIMO - the images were never intended to be used for modeling - they were a quality control action. When the experiment was designed visual models were barely in existence and cats from dogs was a real challenge. The initial model creation was tabular.
\nIMO - one of the issues that is keeping R2 values in the range we are seeing is that the image might in fact NOT show all the harvested material.
", "rawMarkdown": "Reading the [paper](https://www.kaggle.com/competitions/csiro-biomass/discussion/613941) it would seem that the area measured was bounded by a frame. You can often see it's shadow or one or more of the actual edges in the images. Everything inside the frame was harvested.\n\nIMO - the images were never intended to be used for modeling - they were a quality control action. When the experiment was designed visual models were barely in existence and cats from dogs was a real challenge. The initial model creation was tabular.\n\nIMO - one of the issues that is keeping R2 values in the range we are seeing is that the image might in fact NOT show all the harvested material.", "votes": null }, { "id": "3381041", "postDate": "12/23/2025 16:15:31", "content": "I wish my models were sensitive to the 3 or 4th decimal place :)
\nRight now I am still trying to get my models to predict 190 grams when they want to tell me the answer is 90.
", "rawMarkdown": "I wish my models were sensitive to the 3 or 4th decimal place :) \n\nRight now I am still trying to get my models to predict 190 grams when they want to tell me the answer is 90.", "votes": null } ], "comments": [ { "id": 3309708, "author_name": "qiyuliaocsiro", "author_url": "", "post_date": "11/01/2025 07:54:42", "content": "round(a + b) is not necessarily strictly equal to round(a) + round(b), which should be the cause of this error.
\nFor example, ID1035947949, Total = 34.1844, Green + Clover + Dead = 34.1843 (-e-4);
\nGDM = 10.9605, Green + Clover = 10.9604 (-e-4).
\nThis has a negligible impact on the evaluation.
\nThere are no specific requirements for precision when submitting. You can retain the original precision of the model output without rounding it.
", "votes": null, "replies": [ { "id": 3380846, "author_name": "chenbaoying", "author_url": "", "post_date": "12/23/2025 07:22:42", "content": "\"Total = Green + Clover + Dead. What is the maximum value of Total? Were the shooting equipment and camera angles kept consistent?
\nI wonder if the value of Total might be affected by the distance of the camera. For instance, when the camera zooms out, the captured area is larger compared to zooming in. Does this mean that Total increases as the camera moves further away?\"
", "votes": null, "replies": [ { "id": 3381030, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/23/2025 15:58:47", "content": "Reading the paper it would seem that the area measured was bounded by a frame. You can often see it's shadow or one or more of the actual edges in the images. Everything inside the frame was harvested.
\nIMO - the images were never intended to be used for modeling - they were a quality control action. When the experiment was designed visual models were barely in existence and cats from dogs was a real challenge. The initial model creation was tabular.
\nIMO - one of the issues that is keeping R2 values in the range we are seeing is that the image might in fact NOT show all the harvested material.
", "votes": null, "replies": [] } ] } ] }, { "id": 3381041, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/23/2025 16:15:31", "content": "I wish my models were sensitive to the 3 or 4th decimal place :)
\nRight now I am still trying to get my models to predict 190 grams when they want to tell me the answer is 90.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3309671": "I verified all the target values of train.csv and found that:\n1. Total = Green + Clover + Dead\n2. GDM = Green + Clover\n\n[Target Check](https://www.kaggle.com/code/unsayin9/target-check?scriptVersionId=272560832)\n\nBut they are not exactly equal (such as ID1035947949), I would like to know the reason for this problem and whether there is a corresponding solution (how many decimal places should be kept in the submission?)", "3309708": "round(a + b) is not necessarily strictly equal to round(a) + round(b), which should be the cause of this error.\n\nFor example, ID1035947949, Total = 34.1844, Green + Clover + Dead = 34.1843 (-e-4); \n\nGDM = 10.9605, Green + Clover = 10.9604 (-e-4).\n\nThis has a negligible impact on the evaluation.\n\nThere are no specific requirements for precision when submitting. You can retain the original precision of the model output without rounding it.", "3380846": "\"Total = Green + Clover + Dead. What is the maximum value of Total? Were the shooting equipment and camera angles kept consistent? \n\nI wonder if the value of Total might be affected by the distance of the camera. For instance, when the camera zooms out, the captured area is larger compared to zooming in. Does this mean that Total increases as the camera moves further away?\"", "3381030": "Reading the [paper](https://www.kaggle.com/competitions/csiro-biomass/discussion/613941) it would seem that the area measured was bounded by a frame. You can often see it's shadow or one or more of the actual edges in the images. Everything inside the frame was harvested.\n\nIMO - the images were never intended to be used for modeling - they were a quality control action. When the experiment was designed visual models were barely in existence and cats from dogs was a real challenge. The initial model creation was tabular.\n\nIMO - one of the issues that is keeping R2 values in the range we are seeing is that the image might in fact NOT show all the harvested material.", "3381041": "I wish my models were sensitive to the 3 or 4th decimal place :) \n\nRight now I am still trying to get my models to predict 190 grams when they want to tell me the answer is 90." }, "source": "meta" }