![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2Fad8d8c0b230547a214cc7399f483845e%2FScreenshot%202024-01-16%20at%203.25.43AM.png?generation=1705355775031469&alt=media\")
Which transformation is used for the label during official evaluation? label/sum(label)?softmax(label)?or?
", "messages": [ { "id": 2606497, "postDate": "2024-01-17T17:21:29.730Z", "content": "Which transformation is used for the label during official evaluation? label/sum(label)?softmax(label)?or?
", "rawMarkdown": "Which transformation is used for the label during official evaluation? label/sum(label)?softmax(label)?or?", "votes": 21 }, { "id": 2606507, "postDate": "2024-01-17T17:27:21.093Z", "content": "No transformation. If you rows don't add up to 1, then your submission will throw an error
", "rawMarkdown": "No transformation. If you rows don't add up to 1, then your submission will throw an error", "votes": 3, "replies": [ { "id": 2606512, "postDate": "2024-01-17T17:32:49.473Z", "content": "What I mean is label, not pred.
", "rawMarkdown": "What I mean is label, not pred.", "votes": 1 }, { "id": 2607129, "postDate": "2024-01-18T04:20:09.880Z", "content": "Yep that explains a lot hahaha I have been losing my mind on what has caused these thrown exception errors
", "rawMarkdown": "Yep that explains a lot hahaha I have been losing my mind on what has caused these thrown exception errors", "votes": 1, "replies": [ { "id": 2607182, "postDate": "2024-01-18T05:20:24.133Z", "rawMarkdown": "", "isDeleted": true } ] } ] }, { "id": 2606531, "postDate": "2024-01-17T17:54:45.123Z", "content": "The metric code is posted here: https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook
", "rawMarkdown": "The metric code is posted here: https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook", "votes": 1, "replies": [ { "id": 2606565, "postDate": "2024-01-17T18:10:49.273Z", "content": "I know about metrics, but still can't solve my question. We only know that the sum of labels is 1, but I have ten thousand ways to make the sum of labels be 1.
", "rawMarkdown": "I know about metrics, but still can't solve my question. We only know that the sum of labels is 1, but I have ten thousand ways to make the sum of labels be 1.", "votes": 7, "replies": [ { "id": 2606567, "postDate": "2024-01-17T18:12:34.753Z", "content": "Good point Rib. @sohier I think the question is how does Kaggle make the ground truth probabilities from the ground truth vote counts (during LB evaluation)?
", "rawMarkdown": "Good point Rib. @sohier I think the question is how does Kaggle make the ground truth probabilities from the ground truth vote counts (during LB evaluation)?", "votes": 3 }, { "id": 2606605, "postDate": "2024-01-17T18:38:08.323Z", "content": "If there were 7 votes, 4 for seizure and 3 for other, the target values would be 4/7 and 3/7 respectively.
", "rawMarkdown": "If there were 7 votes, 4 for seizure and 3 for other, the target values would be 4/7 and 3/7 respectively.", "votes": 11 }, { "id": 2606636, "postDate": "2024-01-17T19:00:06.943Z", "content": "@sohier for egg_id = 11127485 as discussed here
\nseizure_vote = 28/29 and gpd_vote = 1/29 ( as we are aggregating all votes for each eeg_id from all its eeg_sub_ids) is this assumption right ? since test set have only eeg_id, spectrogram_id, patient_id
\n
The test data only has 1 of each eeg_id in test.csv. So your question does not apply to the test data. Nor Kaggle's computation of our LB score.
@cdeotte Understood, i miss these words test_eegs: Exactly 50 seconds of EEG data.. Thanks, my question is not relevant. @sohier ignore my question please.
", "rawMarkdown": "@cdeotte Understood, i miss these words *test_eegs*: **Exactly 50 seconds of EEG data.**. Thanks, my question is not relevant. @sohier ignore my question please.", "votes": 3 } ] }, { "id": 2611253, "postDate": "2024-01-20T16:34:41.233Z", "content": "@sohier, as KL is not symmetric, the P will be expert probability and Q will be model predicted probability. Is that correct? Thank you,
", "rawMarkdown": "@sohier, as [KL](https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence) is not symmetric, the P will be expert probability and Q will be model predicted probability. Is that correct? Thank you," } ] } ], "comments": [ { "id": 2606507, "author_name": "Chris Deotte", "author_url": "", "post_date": "2024-01-17T17:27:21.093000", "content": "No transformation. If you rows don't add up to 1, then your submission will throw an error
", "votes": 3, "replies": [ { "id": 2606512, "author_name": "Rib~", "author_url": "", "post_date": "2024-01-17T17:32:49.473000", "content": "What I mean is label, not pred.
", "votes": 1, "replies": [] }, { "id": 2607129, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-01-18T04:20:09.880000", "content": "Yep that explains a lot hahaha I have been losing my mind on what has caused these thrown exception errors
", "votes": 1, "replies": [ { "id": 2607182, "author_name": "", "author_url": "", "post_date": "2024-01-18T05:20:24.133000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 2606531, "author_name": "Sohier Dane", "author_url": "", "post_date": "2024-01-17T17:54:45.123000", "content": "The metric code is posted here: https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook
", "votes": 1, "replies": [ { "id": 2606565, "author_name": "Rib~", "author_url": "", "post_date": "2024-01-17T18:10:49.273000", "content": "I know about metrics, but still can't solve my question. We only know that the sum of labels is 1, but I have ten thousand ways to make the sum of labels be 1.
", "votes": 7, "replies": [ { "id": 2606567, "author_name": "Chris Deotte", "author_url": "", "post_date": "2024-01-17T18:12:34.753000", "content": "Good point Rib. @sohier I think the question is how does Kaggle make the ground truth probabilities from the ground truth vote counts (during LB evaluation)?
", "votes": 3, "replies": [] }, { "id": 2606605, "author_name": "Sohier Dane", "author_url": "", "post_date": "2024-01-17T18:38:08.323000", "content": "If there were 7 votes, 4 for seizure and 3 for other, the target values would be 4/7 and 3/7 respectively.
", "votes": 11, "replies": [] }, { "id": 2606636, "author_name": "SeshuRaju 🧘♂️", "author_url": "", "post_date": "2024-01-17T19:00:06.943000", "content": "@sohier for egg_id = 11127485 as discussed here
\nseizure_vote = 28/29 and gpd_vote = 1/29 ( as we are aggregating all votes for each eeg_id from all its eeg_sub_ids) is this assumption right ? since test set have only eeg_id, spectrogram_id, patient_id
\n
The test data only has 1 of each eeg_id in test.csv. So your question does not apply to the test data. Nor Kaggle's computation of our LB score.
@cdeotte Understood, i miss these words test_eegs: Exactly 50 seconds of EEG data.. Thanks, my question is not relevant. @sohier ignore my question please.
", "votes": 3, "replies": [] } ] }, { "id": 2611253, "author_name": "Zhenlan", "author_url": "", "post_date": "2024-01-20T16:34:41.233000", "content": "@sohier, as KL is not symmetric, the P will be expert probability and Q will be model predicted probability. Is that correct? Thank you,
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2606497": "Which transformation is used for the label during official evaluation? label/sum(label)?softmax(label)?or?", "2606507": "No transformation. If you rows don't add up to 1, then your submission will throw an error", "2606531": "The metric code is posted here: https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook" } }