![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F405318%2F2ff7154e5edd58268c507ccec67e0578%2F1.2.826.0.1.3680043.8.498.54838912855754621411231816825017074111.jpg?generation=1609748950729761&alt=media\")
This is a 3rd series of potentially useful external data, please make sure to check previous ones:
\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183
\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207228
I am releasing yet another one, which has potential to boost CVC category AUC's for your models. The dataset with its description can be found at:
\nhttps://www.kaggle.com/raddar/ranzcr-clip-abnormal-cvc-categories
Exploratory notebook:
\nhttps://www.kaggle.com/raddar/cvc-categories-explained-with-examples
I strongly believe it is going to be useful, as CVC is the hardest problem in this competition - and in my experience, classification models struggle with CVC position classification - it needs some extra help :)
\nI hope someone finds this useful!
", "messages": [ { "id": "1132559", "postDate": "12/30/2020 13:50:23", "content": "This is a 3rd series of potentially useful external data, please make sure to check previous ones:
\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183
\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207228
I am releasing yet another one, which has potential to boost CVC category AUC's for your models. The dataset with its description can be found at:
\nhttps://www.kaggle.com/raddar/ranzcr-clip-abnormal-cvc-categories
Exploratory notebook:
\nhttps://www.kaggle.com/raddar/cvc-categories-explained-with-examples
I strongly believe it is going to be useful, as CVC is the hardest problem in this competition - and in my experience, classification models struggle with CVC position classification - it needs some extra help :)
\nI hope someone finds this useful!
", "rawMarkdown": "This is a 3rd series of potentially useful external data, please make sure to check previous ones:\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207228\n\nI am releasing yet another one, which has potential to boost CVC category AUC's for your models. The dataset with its description can be found at:\nhttps://www.kaggle.com/raddar/ranzcr-clip-abnormal-cvc-categories\n\nExploratory notebook:\nhttps://www.kaggle.com/raddar/cvc-categories-explained-with-examples\n\nI strongly believe it is going to be useful, as CVC is the hardest problem in this competition - and in my experience, classification models struggle with CVC position classification - it needs some extra help :)\n\nI hope someone finds this useful!", "votes": null }, { "id": "1132567", "postDate": "12/30/2020 13:52:49", "content": "On the side note, if anyone interested in teaming up, please PM.
\nI can only spend 1-2 hrs a day, but have previous experience and domain knowledge with the competition problem.
\n\nI am releasing yet another one, which has potential to boost CVC category AUC's for your models.
\n
Maybe i misunderstood it but, did you made AUC's for every label?
", "rawMarkdown": "> I am releasing yet another one, which has potential to boost CVC category AUC's for your models. \n\nMaybe i misunderstood it but, did you made AUC's for every label?", "votes": null }, { "id": "1132589", "postDate": "12/30/2020 14:16:14", "content": "If you train model, and look at validation AUC, you will see that CVC categories have worse AUC than ETT and NGT categories. so biggest uplift woud come from improving CVC category AUC scores.
", "rawMarkdown": "If you train model, and look at validation AUC, you will see that CVC categories have worse AUC than ETT and NGT categories. so biggest uplift woud come from improving CVC category AUC scores.", "votes": null }, { "id": "1134048", "postDate": "12/31/2020 19:31:17", "content": "Yes, I noticed that too, may I ask what's your AUC on the CVC targets? For me, with a EfficientNetB6, the AUC for CVC seems to be about 89~ish.
", "rawMarkdown": "Yes, I noticed that too, may I ask what's your AUC on the CVC targets? For me, with a EfficientNetB6, the AUC for CVC seems to be about 89~ish.", "votes": null }, { "id": "1137403", "postDate": "01/03/2021 22:28:50", "content": "Interesting, thank you for sharing!
", "rawMarkdown": "Interesting, thank you for sharing!", "votes": null }, { "id": "1137595", "postDate": "01/04/2021 04:50:56", "content": "Thanks @raddar for all your posts, notebooks and explanations. Was thinking there should be a discussion thread with all of them! Will add this one here as well just in case for others -
\nhttps://www.kaggle.com/raddar/errors-in-ett-abnormal-label
Do you think in CVC - Borderline there are similar issues? Or just issues with subjectivity in labelling - one person's Borderline is another's Abnormal. For me CVCs AUC are not the best but so far CVC - Borderline is the worst.
", "rawMarkdown": "Thanks @raddar for all your posts, notebooks and explanations. Was thinking there should be a discussion thread with all of them! Will add this one here as well just in case for others -\nhttps://www.kaggle.com/raddar/errors-in-ett-abnormal-label\n\nDo you think in CVC - Borderline there are similar issues? Or just issues with subjectivity in labelling - one person's Borderline is another's Abnormal. For me CVCs AUC are not the best but so far CVC - Borderline is the worst.", "votes": null }, { "id": "1137871", "postDate": "01/04/2021 08:32:44", "content": "Yeah Borderline is hard, look at these cases:
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F405318%2F8f8ed7453776c3586768db5ebb5ee445%2F1.2.826.0.1.3680043.8.498.13352422433167291378082369226432905451.jpg?generation=1609749057818143&alt=media\")
Light blue is CVC-Normal, dark blue box - CVC-Borderline. As you can see the CVC ends in the same spot, but is treated differentl!y.
\nThe problem with CVC-Borderline - it has many definitions in this competition. There are hard borderline cases (heart region) and easy borderline cases (i.e. CVC ends in the arm)
", "rawMarkdown": "Yeah Borderline is hard, look at these cases:\n\n\n\n\n\nLight blue is CVC-Normal, dark blue box - CVC-Borderline. As you can see the CVC ends in the same spot, but is treated differentl!y.\n\n\nThe problem with CVC-Borderline - it has many definitions in this competition. There are hard borderline cases (heart region) and easy borderline cases (i.e. CVC ends in the arm)", "votes": null }, { "id": "1138497", "postDate": "01/04/2021 18:15:50", "content": "great work but i have a question how can i figure out auc score for CVC targets
", "rawMarkdown": "great work but i have a question how can i figure out auc score for CVC targets", "votes": null }, { "id": "1138853", "postDate": "01/05/2021 02:40:26", "content": "Hi,how can I use this data to train model?I need to add another class like'CVC-abnormal-right-atrium'?Thanks!
", "rawMarkdown": "Hi,how can I use this data to train model?I need to add another class like'CVC-abnormal-right-atrium'?Thanks!", "votes": null }, { "id": "1166047", "postDate": "01/23/2021 11:11:12", "content": "Hi,I use this extra data to test the resnet200d using @yasufuminakama 's 3 steps method,but I find the local cv is 0.949 is lower than origin 0.955.I change the labels to 13. Don't know if I did something wrong. Thanks! @raddar
", "rawMarkdown": "Hi,I use this extra data to test the resnet200d using @yasufuminakama 's 3 steps method,but I find the local cv is 0.949 is lower than origin 0.955.I change the labels to 13. Don't know if I did something wrong. Thanks! @raddar", "votes": null }, { "id": "1185298", "postDate": "02/04/2021 05:25:37", "content": "@raddar - I only realised there is the ability to attach a dataset to a competition via the tab along the top Overview Data… so I have added this Dataset of yours just for ease of finding. Hope that is OK. The link is the same just makes it show up here in case people are looking for it and it is still has your name. If you want/don't mind I will add other Datasets you have made for this competition too.
", "rawMarkdown": "raddar - I only realised there is the ability to attach a dataset to a competition via the tab along the top Overview Data... so I have added this Dataset of yours just for ease of finding. Hope that is OK. The link is the same just makes it show up here in case people are looking for it and it is still has your name. If you want/don't mind I will add other Datasets you have made for this competition too.", "votes": null } ], "comments": [ { "id": 1132567, "author_name": "raddar", "author_url": "", "post_date": "12/30/2020 13:52:49", "content": "On the side note, if anyone interested in teaming up, please PM.
\nI can only spend 1-2 hrs a day, but have previous experience and domain knowledge with the competition problem.
\n\nI am releasing yet another one, which has potential to boost CVC category AUC's for your models.
\n
Maybe i misunderstood it but, did you made AUC's for every label?
", "votes": null, "replies": [ { "id": 1132589, "author_name": "raddar", "author_url": "", "post_date": "12/30/2020 14:16:14", "content": "If you train model, and look at validation AUC, you will see that CVC categories have worse AUC than ETT and NGT categories. so biggest uplift woud come from improving CVC category AUC scores.
", "votes": null, "replies": [] }, { "id": 1134048, "author_name": "andy1010", "author_url": "", "post_date": "12/31/2020 19:31:17", "content": "Yes, I noticed that too, may I ask what's your AUC on the CVC targets? For me, with a EfficientNetB6, the AUC for CVC seems to be about 89~ish.
", "votes": null, "replies": [] } ] }, { "id": 1137403, "author_name": "raymondlzhou", "author_url": "", "post_date": "01/03/2021 22:28:50", "content": "Interesting, thank you for sharing!
", "votes": null, "replies": [] }, { "id": 1137595, "author_name": "something4kag", "author_url": "", "post_date": "01/04/2021 04:50:56", "content": "Thanks @raddar for all your posts, notebooks and explanations. Was thinking there should be a discussion thread with all of them! Will add this one here as well just in case for others -
\nhttps://www.kaggle.com/raddar/errors-in-ett-abnormal-label
Do you think in CVC - Borderline there are similar issues? Or just issues with subjectivity in labelling - one person's Borderline is another's Abnormal. For me CVCs AUC are not the best but so far CVC - Borderline is the worst.
", "votes": null, "replies": [ { "id": 1137871, "author_name": "raddar", "author_url": "", "post_date": "01/04/2021 08:32:44", "content": "Yeah Borderline is hard, look at these cases:
\n
Light blue is CVC-Normal, dark blue box - CVC-Borderline. As you can see the CVC ends in the same spot, but is treated differentl!y.
\nThe problem with CVC-Borderline - it has many definitions in this competition. There are hard borderline cases (heart region) and easy borderline cases (i.e. CVC ends in the arm)
", "votes": null, "replies": [] } ] }, { "id": 1138497, "author_name": "mohamed3abdelrazik", "author_url": "", "post_date": "01/04/2021 18:15:50", "content": "great work but i have a question how can i figure out auc score for CVC targets
", "votes": null, "replies": [] }, { "id": 1138853, "author_name": "bcwang", "author_url": "", "post_date": "01/05/2021 02:40:26", "content": "Hi,how can I use this data to train model?I need to add another class like'CVC-abnormal-right-atrium'?Thanks!
", "votes": null, "replies": [] }, { "id": 1166047, "author_name": "bcwang", "author_url": "", "post_date": "01/23/2021 11:11:12", "content": "Hi,I use this extra data to test the resnet200d using @yasufuminakama 's 3 steps method,but I find the local cv is 0.949 is lower than origin 0.955.I change the labels to 13. Don't know if I did something wrong. Thanks! @raddar
", "votes": null, "replies": [] }, { "id": 1185298, "author_name": "something4kag", "author_url": "", "post_date": "02/04/2021 05:25:37", "content": "@raddar - I only realised there is the ability to attach a dataset to a competition via the tab along the top Overview Data… so I have added this Dataset of yours just for ease of finding. Hope that is OK. The link is the same just makes it show up here in case people are looking for it and it is still has your name. If you want/don't mind I will add other Datasets you have made for this competition too.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1132559": "This is a 3rd series of potentially useful external data, please make sure to check previous ones:\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183\nhttps://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207228\n\nI am releasing yet another one, which has potential to boost CVC category AUC's for your models. The dataset with its description can be found at:\nhttps://www.kaggle.com/raddar/ranzcr-clip-abnormal-cvc-categories\n\nExploratory notebook:\nhttps://www.kaggle.com/raddar/cvc-categories-explained-with-examples\n\nI strongly believe it is going to be useful, as CVC is the hardest problem in this competition - and in my experience, classification models struggle with CVC position classification - it needs some extra help :)\n\nI hope someone finds this useful!", "1132567": "On the side note, if anyone interested in teaming up, please PM.\nI can only spend 1-2 hrs a day, but have previous experience and domain knowledge with the competition problem.", "1132585": "> I am releasing yet another one, which has potential to boost CVC category AUC's for your models. \n\nMaybe i misunderstood it but, did you made AUC's for every label?", "1132589": "If you train model, and look at validation AUC, you will see that CVC categories have worse AUC than ETT and NGT categories. so biggest uplift woud come from improving CVC category AUC scores.", "1134048": "Yes, I noticed that too, may I ask what's your AUC on the CVC targets? For me, with a EfficientNetB6, the AUC for CVC seems to be about 89~ish.", "1137403": "Interesting, thank you for sharing!", "1137595": "Thanks @raddar for all your posts, notebooks and explanations. Was thinking there should be a discussion thread with all of them! Will add this one here as well just in case for others -\nhttps://www.kaggle.com/raddar/errors-in-ett-abnormal-label\n\nDo you think in CVC - Borderline there are similar issues? Or just issues with subjectivity in labelling - one person's Borderline is another's Abnormal. For me CVCs AUC are not the best but so far CVC - Borderline is the worst.", "1137871": "Yeah Borderline is hard, look at these cases:\n\n\n\n\n\nLight blue is CVC-Normal, dark blue box - CVC-Borderline. As you can see the CVC ends in the same spot, but is treated differentl!y.\n\n\nThe problem with CVC-Borderline - it has many definitions in this competition. There are hard borderline cases (heart region) and easy borderline cases (i.e. CVC ends in the arm)", "1138497": "great work but i have a question how can i figure out auc score for CVC targets", "1138853": "Hi,how can I use this data to train model?I need to add another class like'CVC-abnormal-right-atrium'?Thanks!", "1166047": "Hi,I use this extra data to test the resnet200d using @yasufuminakama 's 3 steps method,but I find the local cv is 0.949 is lower than origin 0.955.I change the labels to 13. Don't know if I did something wrong. Thanks! @raddar", "1185298": "raddar - I only realised there is the ability to attach a dataset to a competition via the tab along the top Overview Data... so I have added this Dataset of yours just for ease of finding. Hope that is OK. The link is the same just makes it show up here in case people are looking for it and it is still has your name. If you want/don't mind I will add other Datasets you have made for this competition too." }, "source": "meta" }