There have been a few recent posts on extravasation_injury and weighting -
\nif extravasation_injury is scaled up, then this will help contribute to the any_injury predictions
\n\n\nJust wondered if the metric using an overall any_injury is best or should there be any_injury per label group? or should any_injury be predicted and included in the submissions?
\nPerhaps the effect of extravasation_injury in predictions is just an artifact of the public test set and LB. But if there is some bias/issue in any_injury might be good to know sooner rather than later.
", "messages": [ { "id": 2430203, "postDate": "2023-09-09T06:23:32.460Z", "content": "There have been a few recent posts on extravasation_injury and weighting -
\nif extravasation_injury is scaled up, then this will help contribute to the any_injury predictions
\n\n\nJust wondered if the metric using an overall any_injury is best or should there be any_injury per label group? or should any_injury be predicted and included in the submissions?
\nPerhaps the effect of extravasation_injury in predictions is just an artifact of the public test set and LB. But if there is some bias/issue in any_injury might be good to know sooner rather than later.
", "rawMarkdown": "There have been a few recent posts on extravasation_injury and weighting - \n\n[if extravasation_injury is scaled up, then this will help contribute to the any_injury predictions](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/436501#2428891)\n\n[setting the prediction value for a specific label 'extravasation_injury' to a constant 0.99 yielded outstanding results ](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/437510)\n\n[overfitting,suspect that the test concentration in lb contains more of the extravasation_injury' class,](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/437303)\n\nJust wondered if the metric using an overall any_injury is best or should there be any_injury per label group? or should any_injury be predicted and included in the submissions? \n\nPerhaps the effect of extravasation_injury in predictions is just an artifact of the public test set and LB. But if there is some bias/issue in any_injury might be good to know sooner rather than later. ", "votes": 2 }, { "id": 2430266, "postDate": "2023-09-09T07:15:50.817Z", "content": "any_injury may not be included into predictions. It's calculated automaticallymax(1 - p(healthy|organ))
but the loss function should include this
", "rawMarkdown": "but the loss function should include this" }, { "id": 2430310, "postDate": "2023-09-09T07:40:53.380Z", "content": "This post is not about including it in predictions/submissions. It is about whether deriving it overall is best for how the metric calculates considering posts about how to get better scores by increasing the extravasation_injury predictions.
", "rawMarkdown": "This post is not about including it in predictions/submissions. It is about whether deriving it overall is best for how the metric calculates considering posts about how to get better scores by increasing the extravasation_injury predictions. " } ] } ], "comments": [ { "id": 2430266, "author_name": "JanGlinko2", "author_url": "", "post_date": "2023-09-09T07:15:50.817000", "content": "any_injury may not be included into predictions. It's calculated automaticallymax(1 - p(healthy|organ))
but the loss function should include this
", "votes": 0, "replies": [] }, { "id": 2430310, "author_name": "something4kag", "author_url": "", "post_date": "2023-09-09T07:40:53.380000", "content": "This post is not about including it in predictions/submissions. It is about whether deriving it overall is best for how the metric calculates considering posts about how to get better scores by increasing the extravasation_injury predictions.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2430203": "There have been a few recent posts on extravasation_injury and weighting - \n\n[if extravasation_injury is scaled up, then this will help contribute to the any_injury predictions](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/436501#2428891)\n\n[setting the prediction value for a specific label 'extravasation_injury' to a constant 0.99 yielded outstanding results ](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/437510)\n\n[overfitting,suspect that the test concentration in lb contains more of the extravasation_injury' class,](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/437303)\n\nJust wondered if the metric using an overall any_injury is best or should there be any_injury per label group? or should any_injury be predicted and included in the submissions? \n\nPerhaps the effect of extravasation_injury in predictions is just an artifact of the public test set and LB. But if there is some bias/issue in any_injury might be good to know sooner rather than later. ", "2430266": "any_injury may not be included into predictions. It's calculated automatically` max(1 - p(healthy|organ)) `" } }