{ "id": 466876, "title": "Predictions need to sum to 1 for every row", "url": "/competitions/hms-harmful-brain-activity-classification/discussion/466876", "author_name": "narsil (jobs-in-data.com)", "post_date": "2024-01-10T08:32:23.718000", "votes": 18, "comment_count": 8, "views": 0, "content": "

According to the metric implementation: https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook:

\n
\n

The predictions in each row must add to one

\n
\n

It is not mentioned in the data description, but any submission.csv that does not fulfill this condition throws an error. It means that predictions are not rescaled automatically.

", "messages": [ { "id": 2595087, "postDate": "2024-01-10T08:32:23.720Z", "content": "

According to the metric implementation: https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook:

\n
\n

The predictions in each row must add to one

\n
\n

It is not mentioned in the data description, but any submission.csv that does not fulfill this condition throws an error. It means that predictions are not rescaled automatically.

", "rawMarkdown": "According to the metric implementation: `https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook`:\n\n>The predictions in each row must add to one\n\nIt is not mentioned in the data description, but any `submission.csv` that does not fulfill this condition throws an error. It means that predictions are not rescaled automatically.", "votes": 17 }, { "id": 2595914, "postDate": "2024-01-10T18:09:23.950Z", "content": "

@narsil Thanks for the feedback. I've updated the description to make this constraint more explicit.

", "rawMarkdown": "@narsil Thanks for the feedback. I've updated the description to make this constraint more explicit.", "votes": 6 }, { "id": 2595518, "postDate": "2024-01-10T13:36:00.770Z", "content": "

Nice discovery Narsil. I think the Evaluation Tab needs to be updated because the sample_submission.csv example (here) that Kaggle provides does not add to one:

\n
,seizure_vote,lpd_vote,gpd_vote,lrda_vote,grda_vote,other_vote\n,.,.,.,.,.,.\n,.,.,.,.,.,.\n.\n
", "rawMarkdown": "Nice discovery Narsil. I think the Evaluation Tab needs to be updated because the `sample_submission.csv` example ([here][1]) that Kaggle provides does not add to one:\n\n eeg_id,seizure_vote,lpd_vote,gpd_vote,lrda_vote,grda_vote,other_vote\n 0,0.166,0.166,0.166,0.166,0.166,0.166\n 1,0.166,0.166,0.166,0.166,0.166,0.166\n etc.\n\n[1]: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/overview/evaluation", "votes": 4, "replies": [ { "id": 2595917, "postDate": "2024-01-10T18:09:52.633Z", "content": "

Fair enough, done.

", "rawMarkdown": "Fair enough, done.", "votes": 2, "replies": [ { "id": 2595928, "postDate": "2024-01-10T18:17:24.387Z", "rawMarkdown": "", "isDeleted": true } ] } ] }, { "id": 2623379, "postDate": "2024-01-28T07:37:09.797Z", "content": "

The total value is 1, but an error also occurs if the predicted value contains a negative value. Since we are predicting probabilities, it is not inherently possible for a negative value to be included, but I tried it and it was an error.

", "rawMarkdown": "The total value is 1, but an error also occurs if the predicted value contains a negative value. Since we are predicting probabilities, it is not inherently possible for a negative value to be included, but I tried it and it was an error.", "votes": 1 }, { "id": 2595122, "postDate": "2024-01-10T08:54:52.020Z", "content": "

So this is multiclass classification then. Can someone with domain knowledge confirm that labels are mutually exclusive?

", "rawMarkdown": "So this is multiclass classification then. Can someone with domain knowledge confirm that labels are mutually exclusive?", "replies": [ { "id": 2595139, "postDate": "2024-01-10T09:09:57.803Z", "content": "

Also,

\n
\n

The KL divergence is technically undefined/infinite where the target equals zero.

\n
\n

So it means that all the patients in the dataset have some type of a harmful brain activity. Our objective is to classify which one.

", "rawMarkdown": "Also, \n>The KL divergence is technically undefined/infinite where the target equals zero.\n\nSo it means that all the patients in the dataset have some type of a harmful brain activity. Our objective is to classify which one.", "votes": 4 } ] }, { "id": 2600670, "postDate": "2024-01-13T19:17:31.577Z", "content": "

Thanks for the info!

", "rawMarkdown": "Thanks for the info!" } ], "comments": [ { "id": 2595914, "author_name": "Sohier Dane", "author_url": "", "post_date": "2024-01-10T18:09:23.950000", "content": "

@narsil Thanks for the feedback. I've updated the description to make this constraint more explicit.

", "votes": 6, "replies": [] }, { "id": 2595518, "author_name": "Chris Deotte", "author_url": "", "post_date": "2024-01-10T13:36:00.770000", "content": "

Nice discovery Narsil. I think the Evaluation Tab needs to be updated because the sample_submission.csv example (here) that Kaggle provides does not add to one:

\n
,seizure_vote,lpd_vote,gpd_vote,lrda_vote,grda_vote,other_vote\n,.,.,.,.,.,.\n,.,.,.,.,.,.\n.\n
", "votes": 4, "replies": [ { "id": 2595917, "author_name": "Sohier Dane", "author_url": "", "post_date": "2024-01-10T18:09:52.633000", "content": "

Fair enough, done.

", "votes": 2, "replies": [ { "id": 2595928, "author_name": "", "author_url": "", "post_date": "2024-01-10T18:17:24.387000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 2623379, "author_name": "tc0000", "author_url": "", "post_date": "2024-01-28T07:37:09.797000", "content": "

The total value is 1, but an error also occurs if the predicted value contains a negative value. Since we are predicting probabilities, it is not inherently possible for a negative value to be included, but I tried it and it was an error.

", "votes": 1, "replies": [] }, { "id": 2595122, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2024-01-10T08:54:52.020000", "content": "

So this is multiclass classification then. Can someone with domain knowledge confirm that labels are mutually exclusive?

", "votes": 0, "replies": [ { "id": 2595139, "author_name": "narsil (jobs-in-data.com)", "author_url": "", "post_date": "2024-01-10T09:09:57.803000", "content": "

Also,

\n
\n

The KL divergence is technically undefined/infinite where the target equals zero.

\n
\n

So it means that all the patients in the dataset have some type of a harmful brain activity. Our objective is to classify which one.

", "votes": 4, "replies": [] } ] }, { "id": 2600670, "author_name": "Ethan S. Lee", "author_url": "", "post_date": "2024-01-13T19:17:31.577000", "content": "

Thanks for the info!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2595087": "According to the metric implementation: `https://www.kaggle.com/code/metric/kullback-leibler-divergence/notebook`:\n\n>The predictions in each row must add to one\n\nIt is not mentioned in the data description, but any `submission.csv` that does not fulfill this condition throws an error. It means that predictions are not rescaled automatically.", "2595914": "@narsil Thanks for the feedback. I've updated the description to make this constraint more explicit.", "2595518": "Nice discovery Narsil. I think the Evaluation Tab needs to be updated because the `sample_submission.csv` example ([here][1]) that Kaggle provides does not add to one:\n\n eeg_id,seizure_vote,lpd_vote,gpd_vote,lrda_vote,grda_vote,other_vote\n 0,0.166,0.166,0.166,0.166,0.166,0.166\n 1,0.166,0.166,0.166,0.166,0.166,0.166\n etc.\n\n[1]: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/overview/evaluation", "2623379": "The total value is 1, but an error also occurs if the predicted value contains a negative value. Since we are predicting probabilities, it is not inherently possible for a negative value to be included, but I tried it and it was an error.", "2595122": "So this is multiclass classification then. Can someone with domain knowledge confirm that labels are mutually exclusive?", "2600670": "Thanks for the info!" } }