In the prizes section, it is mentioned that with context and without context submissions will have separate prizes.\nDoes an ensemble of EfficientNet and XG Boost come under with context category?\nWhat is the best way to select the final submission files
", "messages": [ { "id": "958509", "postDate": "08/05/2020 03:07:09", "content": "In the prizes section, it is mentioned that with context and without context submissions will have separate prizes.\nDoes an ensemble of EfficientNet and XG Boost come under with context category?\nWhat is the best way to select the final submission files
", "rawMarkdown": "In the prizes section, it is mentioned that with context and without context submissions will have separate prizes.\nDoes an ensemble of EfficientNet and XG Boost come under with context category?\nWhat is the best way to select the final submission files", "votes": null }, { "id": "958514", "postDate": "08/05/2020 03:12:26", "content": "\"with context\" means using the column patient_id. If you don't use patient_id then it doesn't matter if you use CNN or XGB.
Thanks Chris.\nWhat will be a good strategy to select the final 3 submissions?\nOne will be the ensemble of our best private submissions.
", "rawMarkdown": "Thanks Chris.\nWhat will be a good strategy to select the final 3 submissions?\nOne will be the ensemble of our best private submissions.", "votes": null }, { "id": "958524", "postDate": "08/05/2020 03:22:22", "content": "It depends if you want to attempt the special prizes. If you don't care about the special prizes, then you have 3 chances for the main prize. Because leaderboard score has a lot of variance I would pick 3 models (or ensembles) that are very different from each other.
", "rawMarkdown": "It depends if you want to attempt the special prizes. If you don't care about the special prizes, then you have 3 chances for the main prize. Because leaderboard score has a lot of variance I would pick 3 models (or ensembles) that are very different from each other.", "votes": null }, { "id": "958638", "postDate": "08/05/2020 05:10:58", "content": "@cdeotte A small mistake i think.\nWe are using patient_id to create folds..So that wont become a problem for that prize
@msharuk589 Good point. We are allowed to use patient_id to create folds and still win the the \"without context\" special prize:
\n\n\n\"Without Context\" refers to approaches which do not in any way consider the variable patient_id to cluster images for inference on training, validation, or test set. If Patient ID is used only for splitting the training data into validation subsets, this is still considered “without context.”
\n
This quote is from prizes description here
", "rawMarkdown": "msharuk589 Good point. We are allowed to use `patient_id` to create folds and still win the the \"without context\" special prize:\n \n> \"Without Context\" refers to approaches which do not in any way consider the variable patient_id to cluster images for inference on training, validation, or test set. **If Patient ID is used only for splitting the training data into validation subsets, this is still considered “without context.**”\n\nThis quote is from prizes description [here][1]\n\n[1]: https://www.kaggle.com/c/siim-isic-melanoma-classification/overview/prizes", "votes": null }, { "id": "959491", "postDate": "08/05/2020 16:12:48", "content": "Good to know!
", "rawMarkdown": "Good to know!", "votes": null }, { "id": "959511", "postDate": "08/05/2020 16:29:46", "content": "I think the Special Prize should be given straight to @cdeotte for all his amazing contributions.
", "rawMarkdown": "I think the Special Prize should be given straight to @cdeotte for all his amazing contributions.", "votes": null } ], "comments": [ { "id": 958514, "author_name": "cdeotte", "author_url": "", "post_date": "08/05/2020 03:12:26", "content": "\"with context\" means using the column patient_id. If you don't use patient_id then it doesn't matter if you use CNN or XGB.
Thanks Chris.\nWhat will be a good strategy to select the final 3 submissions?\nOne will be the ensemble of our best private submissions.
", "votes": null, "replies": [] }, { "id": 958524, "author_name": "cdeotte", "author_url": "", "post_date": "08/05/2020 03:22:22", "content": "It depends if you want to attempt the special prizes. If you don't care about the special prizes, then you have 3 chances for the main prize. Because leaderboard score has a lot of variance I would pick 3 models (or ensembles) that are very different from each other.
", "votes": null, "replies": [] }, { "id": 958638, "author_name": "msharuk589", "author_url": "", "post_date": "08/05/2020 05:10:58", "content": "@cdeotte A small mistake i think.\nWe are using patient_id to create folds..So that wont become a problem for that prize
@msharuk589 Good point. We are allowed to use patient_id to create folds and still win the the \"without context\" special prize:
\n\n\n\"Without Context\" refers to approaches which do not in any way consider the variable patient_id to cluster images for inference on training, validation, or test set. If Patient ID is used only for splitting the training data into validation subsets, this is still considered “without context.”
\n
This quote is from prizes description here
", "votes": null, "replies": [] }, { "id": 959491, "author_name": "santiviquez", "author_url": "", "post_date": "08/05/2020 16:12:48", "content": "Good to know!
", "votes": null, "replies": [] } ] }, { "id": 959511, "author_name": "serigne", "author_url": "", "post_date": "08/05/2020 16:29:46", "content": "I think the Special Prize should be given straight to @cdeotte for all his amazing contributions.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "958509": "In the prizes section, it is mentioned that with context and without context submissions will have separate prizes.\nDoes an ensemble of EfficientNet and XG Boost come under with context category?\nWhat is the best way to select the final submission files", "958514": "\"with context\" means using the column `patient_id`. If you don't use `patient_id` then it doesn't matter if you use CNN or XGB.", "958520": "Thanks Chris.\nWhat will be a good strategy to select the final 3 submissions?\nOne will be the ensemble of our best private submissions.", "958524": "It depends if you want to attempt the special prizes. If you don't care about the special prizes, then you have 3 chances for the main prize. Because leaderboard score has a lot of variance I would pick 3 models (or ensembles) that are very different from each other.", "958638": "cdeotte A small mistake i think.\nWe are using `patient_id` to create folds..So that wont become a problem for that prize", "958648": "msharuk589 Good point. We are allowed to use `patient_id` to create folds and still win the the \"without context\" special prize:\n \n> \"Without Context\" refers to approaches which do not in any way consider the variable patient_id to cluster images for inference on training, validation, or test set. **If Patient ID is used only for splitting the training data into validation subsets, this is still considered “without context.**”\n\nThis quote is from prizes description [here][1]\n\n[1]: https://www.kaggle.com/c/siim-isic-melanoma-classification/overview/prizes", "959491": "Good to know!", "959511": "I think the Special Prize should be given straight to @cdeotte for all his amazing contributions." }, "source": "meta" }