To Kaggle/Organisers,
\n\nEvaluation is stated as follows:
\n\n\n\n\nSubmissions are evaluated using a weighted multi-label logarithmic loss. Each hemorrhage sub-type is its own row for every image, and you are expected to predict a probability for that sub-type of hemorrhage. There is also an any label, which indicates that a hemorrhage of ANY kind exists in the image. The any label is weighted more highly than specific hemorrhage sub-types.
\n
Are you able to tell us the weighting of the \"any\" class? If not, any reason why?
", "messages": [ { "id": 631570, "postDate": "2019-09-22T08:48:36.773Z", "content": "To Kaggle/Organisers,
\n\nEvaluation is stated as follows:
\n\n\n\n\nSubmissions are evaluated using a weighted multi-label logarithmic loss. Each hemorrhage sub-type is its own row for every image, and you are expected to predict a probability for that sub-type of hemorrhage. There is also an any label, which indicates that a hemorrhage of ANY kind exists in the image. The any label is weighted more highly than specific hemorrhage sub-types.
\n
Are you able to tell us the weighting of the \"any\" class? If not, any reason why?
", "rawMarkdown": "To Kaggle/Organisers,\n\nEvaluation is stated as follows:\n\n> Submissions are evaluated using a weighted multi-label logarithmic loss. Each hemorrhage sub-type is its own row for every image, and you are expected to predict a probability for that sub-type of hemorrhage. There is also an any label, which indicates that a hemorrhage of ANY kind exists in the image. **The any label is weighted more highly than specific hemorrhage sub-types.**\n\nAre you able to tell us the weighting of the \"any\" class? If not, any reason why?", "votes": 5 }, { "id": 631885, "postDate": "2019-09-22T21:49:27.110Z", "content": "I found the 'any' weight is x2 of others.\nhttps://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109526#latest-630190
\n\nBut any == max(epidural, intraparenchymal, intraventricular, subarachnoid, subdural) in stage_1_train.csv, so I think training model without any label and fill any as post-processing is good.
", "rawMarkdown": "I found the 'any' weight is x2 of others.\nhttps://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109526#latest-630190\n\nBut any == max(epidural, intraparenchymal, intraventricular, subarachnoid, subdural) in stage_1_train.csv, so I think training model without any label and fill any as post-processing is good.", "votes": 3 }, { "id": 631830, "postDate": "2019-09-22T19:04:32.547Z", "content": "I think it's possible to calculate weights by LB probing. At least that was done so in PLAsTiCC competition to calculate \"secret\" weights for classes.
", "rawMarkdown": "I think it's possible to calculate weights by LB probing. At least that was done so in PLAsTiCC competition to calculate \"secret\" weights for classes.", "votes": 1, "replies": [ { "id": 631864, "postDate": "2019-09-22T20:42:55.533Z", "content": "Yeah, there is the notebook with that:\nhttps://www.kaggle.com/kambarakun/maybe-any-weight-is-x2-of-others
", "rawMarkdown": "Yeah, there is the notebook with that:\nhttps://www.kaggle.com/kambarakun/maybe-any-weight-is-x2-of-others\n" } ] } ], "comments": [ { "id": 631885, "author_name": "kambarakun", "author_url": "", "post_date": "2019-09-22T21:49:27.110000", "content": "I found the 'any' weight is x2 of others.\nhttps://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109526#latest-630190
\n\nBut any == max(epidural, intraparenchymal, intraventricular, subarachnoid, subdural) in stage_1_train.csv, so I think training model without any label and fill any as post-processing is good.
", "votes": 3, "replies": [] }, { "id": 631830, "author_name": "Sergey Zlobin", "author_url": "", "post_date": "2019-09-22T19:04:32.547000", "content": "I think it's possible to calculate weights by LB probing. At least that was done so in PLAsTiCC competition to calculate \"secret\" weights for classes.
", "votes": 1, "replies": [ { "id": 631864, "author_name": "Sergey Zlobin", "author_url": "", "post_date": "2019-09-22T20:42:55.533000", "content": "Yeah, there is the notebook with that:\nhttps://www.kaggle.com/kambarakun/maybe-any-weight-is-x2-of-others
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "631570": "To Kaggle/Organisers,\n\nEvaluation is stated as follows:\n\n> Submissions are evaluated using a weighted multi-label logarithmic loss. Each hemorrhage sub-type is its own row for every image, and you are expected to predict a probability for that sub-type of hemorrhage. There is also an any label, which indicates that a hemorrhage of ANY kind exists in the image. **The any label is weighted more highly than specific hemorrhage sub-types.**\n\nAre you able to tell us the weighting of the \"any\" class? If not, any reason why?", "631885": "I found the 'any' weight is x2 of others.\nhttps://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109526#latest-630190\n\nBut any == max(epidural, intraparenchymal, intraventricular, subarachnoid, subdural) in stage_1_train.csv, so I think training model without any label and fill any as post-processing is good.", "631830": "I think it's possible to calculate weights by LB probing. At least that was done so in PLAsTiCC competition to calculate \"secret\" weights for classes." } }