{ "id": 223788, "title": "Independent validation - links to dataset", "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/223788", "author_name": "Dr. Amritpal Singh", "post_date": "2021-03-05T15:08:56.722000", "votes": 0, "comment_count": 2, "views": 0, "content": "

Hii Kagglers,
\nI am Dr. Amrit. Currently, there is a discussion going on discussion forms of this competition regarding over-fitting on public datasets. I have curated a validation dataset for this competition which you can use to evaluate your models.
\nSince labelling Test-set isnt allowed, so I came up with some smart ways to use other data.

\n

Link to the dataset - https://www.kaggle.com/amritpal333/ranzcrindependentvalidation2

\n

As per competition guidelines, I am sharing the dataset here. (I hope I am not missing anything, if I am, please let me know below).

\n

Like if this helps you.
\nI hope this helps the community. May the best one win.

\n

P.S. - Some parts of the self-labeled Test subset might be wrong. If you find something, Kindly write to me ap4.singh@gmail.com

\n

\"\"

", "messages": [ { "id": 1227474, "postDate": "2021-03-05T15:18:27.903Z", "content": "

Since re-labeling the test set isn't allowed, we will explore three possibilities.

\n

Description of the Dataset -

\n
    \n

  1. 3_class.csv - Subset of Support_devices from Chexpert dataset. (contains labels for 4 classes - ETT, NGT, CVC, Swan).
    \n images in folder - Chex_pert_subset

    2. \n

  2. Train_subset_equally_divided.csv A reliable label by a radiologist, with equal distribution from each class. 10-15 images per class. images - Train set of Ranzcr competition

    3. \n

  3. Far_north+south_600.csv Far_north and Far_south are cases that the model(one of mine) was very confident(90+ for present or below 10% meaning absent), but actually model was wrong about those predictions. images - Train set of Ranzcr competition. To understand this Far North and Far south concept - Read the paper - Link to paper

    4. \n
", "rawMarkdown": "Since re-labeling the test set isn't allowed, we will explore three possibilities. \n \nDescription of the Dataset - \n1. 3_class.csv - Subset of Support_devices from Chexpert dataset. (contains labels for 4 classes - ETT, NGT, CVC, Swan).\n images in folder - Chex_pert_subset\n\n2. Train_subset_equally_divided.csv A reliable label by a radiologist, with equal distribution from each class. 10-15 images per class. images - Train set of Ranzcr competition\n\n3. Far_north+south_600.csv Far_north and Far_south are cases that the model(one of mine) was very confident(90+ for present or below 10% meaning absent), but actually model was wrong about those predictions. images - Train set of Ranzcr competition. To understand this Far North and Far south concept - Read the paper - [Link to paper](https://pubmed.ncbi.nlm.nih.gov/31818381/)\n\n", "votes": 1 }, { "id": 1227466, "postDate": "2021-03-05T15:08:56.723Z", "content": "

Hii Kagglers,
\nI am Dr. Amrit. Currently, there is a discussion going on discussion forms of this competition regarding over-fitting on public datasets. I have curated a validation dataset for this competition which you can use to evaluate your models.
\nSince labelling Test-set isnt allowed, so I came up with some smart ways to use other data.

\n

Link to the dataset - https://www.kaggle.com/amritpal333/ranzcrindependentvalidation2

\n

As per competition guidelines, I am sharing the dataset here. (I hope I am not missing anything, if I am, please let me know below).

\n

Like if this helps you.
\nI hope this helps the community. May the best one win.

\n

P.S. - Some parts of the self-labeled Test subset might be wrong. If you find something, Kindly write to me ap4.singh@gmail.com

\n

\"\"

", "rawMarkdown": "Hii Kagglers,\nI am Dr. Amrit. Currently, there is a discussion going on discussion forms of this competition regarding over-fitting on public datasets. I have curated a validation dataset for this competition which you can use to evaluate your models. \nSince labelling Test-set isnt allowed, so I came up with some smart ways to use other data.\n\nLink to the dataset - https://www.kaggle.com/amritpal333/ranzcrindependentvalidation2\n\nAs per competition guidelines, I am sharing the dataset here. (I hope I am not missing anything, if I am, please let me know below).\n\nLike if this helps you.\nI hope this helps the community. May the best one win.\n\nP.S. - Some parts of the self-labeled Test subset might be wrong. If you find something, Kindly write to me ap4.singh@gmail.com \n\n![](https://storage.googleapis.com/kagglesdsdata/datasets/1194466/1996989/Ranzcr%20-%20Frame%206.jpg?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=databundle-worker-v2%40kaggle-161607.iam.gserviceaccount.com%2F20210305%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20210305T161613Z&X-Goog-Expires=172799&X-Goog-SignedHeaders=host&X-Goog-Signature=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)\n" }, { "id": 1227469, "postDate": "2021-03-05T15:11:34.817Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1227474, "author_name": "Dr. Amritpal Singh", "author_url": "", "post_date": "2021-03-05T15:18:27.903000", "content": "

Since re-labeling the test set isn't allowed, we will explore three possibilities.

\n

Description of the Dataset -

\n
    \n

  1. 3_class.csv - Subset of Support_devices from Chexpert dataset. (contains labels for 4 classes - ETT, NGT, CVC, Swan).
    \n images in folder - Chex_pert_subset

    2. \n

  2. Train_subset_equally_divided.csv A reliable label by a radiologist, with equal distribution from each class. 10-15 images per class. images - Train set of Ranzcr competition

    3. \n

  3. Far_north+south_600.csv Far_north and Far_south are cases that the model(one of mine) was very confident(90+ for present or below 10% meaning absent), but actually model was wrong about those predictions. images - Train set of Ranzcr competition. To understand this Far North and Far south concept - Read the paper - Link to paper

    4. \n
", "votes": 1, "replies": [] }, { "id": 1227469, "author_name": "", "author_url": "", "post_date": "2021-03-05T15:11:34.817000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1227474": "Since re-labeling the test set isn't allowed, we will explore three possibilities. \n \nDescription of the Dataset - \n1. 3_class.csv - Subset of Support_devices from Chexpert dataset. (contains labels for 4 classes - ETT, NGT, CVC, Swan).\n images in folder - Chex_pert_subset\n\n2. Train_subset_equally_divided.csv A reliable label by a radiologist, with equal distribution from each class. 10-15 images per class. images - Train set of Ranzcr competition\n\n3. Far_north+south_600.csv Far_north and Far_south are cases that the model(one of mine) was very confident(90+ for present or below 10% meaning absent), but actually model was wrong about those predictions. images - Train set of Ranzcr competition. To understand this Far North and Far south concept - Read the paper - [Link to paper](https://pubmed.ncbi.nlm.nih.gov/31818381/)\n\n", "1227466": "Hii Kagglers,\nI am Dr. Amrit. Currently, there is a discussion going on discussion forms of this competition regarding over-fitting on public datasets. I have curated a validation dataset for this competition which you can use to evaluate your models. \nSince labelling Test-set isnt allowed, so I came up with some smart ways to use other data.\n\nLink to the dataset - https://www.kaggle.com/amritpal333/ranzcrindependentvalidation2\n\nAs per competition guidelines, I am sharing the dataset here. (I hope I am not missing anything, if I am, please let me know below).\n\nLike if this helps you.\nI hope this helps the community. May the best one win.\n\nP.S. - Some parts of the self-labeled Test subset might be wrong. If you find something, Kindly write to me ap4.singh@gmail.com \n\n![](https://storage.googleapis.com/kagglesdsdata/datasets/1194466/1996989/Ranzcr%20-%20Frame%206.jpg?X-Goog-Algorithm=GOOG4-RSA-SHA256&X-Goog-Credential=databundle-worker-v2%40kaggle-161607.iam.gserviceaccount.com%2F20210305%2Fauto%2Fstorage%2Fgoog4_request&X-Goog-Date=20210305T161613Z&X-Goog-Expires=172799&X-Goog-SignedHeaders=host&X-Goog-Signature=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)\n", "1227469": "" } }