{ "id": 444786, "title": "Binary classification or ternary classification？", "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/444786", "author_name": "", "post_date": "2023-10-03T15:44:26.473289400Z", "votes": 9, "comment_count": 7, "views": 0, "content": "

As we can see from the description:

\n
\n

Though each series is a continuous recording, there may be periods in the series when the accelerometer device was removed. These period are determined as those where suspiciously little variation in the accelerometer signals occur over an extended period of time, which is unrealistic for typical human participants. Events are not annotated for these periods, and you should attempt to refrain from making event predictions during these periods: an event prediction will be scored as false positive.

\n
\n

There are some 'invalid' periods. we shouldn't make our onset or wakaup in these periods.
\nSo is this a binary classification, or a ternary classification? After I try to plot a series, like this one:
\n\"\"
\nwe can see the black zones might be the 'invalid' periods. should we label them to another class? If so, we can handle it like a ternary classification, with onset=0, wakaup=1, and invalid=2.

", "messages": [ { "id": "2466152", "postDate": "10/03/2023 15:44:26", "content": "

As we can see from the description:

\n
\n

Though each series is a continuous recording, there may be periods in the series when the accelerometer device was removed. These period are determined as those where suspiciously little variation in the accelerometer signals occur over an extended period of time, which is unrealistic for typical human participants. Events are not annotated for these periods, and you should attempt to refrain from making event predictions during these periods: an event prediction will be scored as false positive.

\n
\n

There are some 'invalid' periods. we shouldn't make our onset or wakaup in these periods.
\nSo is this a binary classification, or a ternary classification? After I try to plot a series, like this one:
\n\"\"
\nwe can see the black zones might be the 'invalid' periods. should we label them to another class? If so, we can handle it like a ternary classification, with onset=0, wakaup=1, and invalid=2.

", "rawMarkdown": "As we can see from the description:\n> Though each series is a continuous recording, there may be periods in the series when the accelerometer device was removed. These period are determined as those where suspiciously little variation in the accelerometer signals occur over an extended period of time, which is unrealistic for typical human participants. Events are not annotated for these periods, and you should attempt to refrain from making event predictions during these periods: an event prediction will be scored as false positive.\n\nThere are some 'invalid' periods. we shouldn't make our `onset` or `wakaup` in these periods.\nSo is this a binary classification, or a ternary classification? After I try to plot a series, like this one:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3110858%2Fee2d67e0c949e276280cab8a17db961b%2FWX20231003-2336512x.png?generation=1696347716631248&alt=media)\nwe can see the **black** zones might be the 'invalid' periods. should we label them to another class? If so, we can handle it like a ternary classification, with `onset=0`, `wakaup=1`, and `invalid=2`.", "votes": null }, { "id": "2466191", "postDate": "10/03/2023 16:19:45", "content": "

Hey Joseph,

\n

I think you have to experiment with this and decide which way gives you better results in the end.
\nProbably no right or wrong if it generates good outcomes! If feel, this is what makes this competition interesting… :)

\n

Best,
\nJan

", "rawMarkdown": "Hey Joseph,\n\nI think you have to experiment with this and decide which way gives you better results in the end.\nProbably no right or wrong if it generates good outcomes! If feel, this is what makes this competition interesting... :) \n\nBest,\nJan", "votes": null }, { "id": "2466193", "postDate": "10/03/2023 16:21:00", "content": "

I mean, have you already compared the two approaches? And if so, can you report on your findings ? :)

", "rawMarkdown": "I mean, have you already compared the two approaches? And if so, can you report on your findings ? :)", "votes": null }, { "id": "2466196", "postDate": "10/03/2023 16:26:46", "content": "

Have not yet. But I will have a try ASAP.

", "rawMarkdown": "Have not yet. But I will have a try ASAP.", "votes": null }, { "id": "2466260", "postDate": "10/03/2023 17:14:12", "content": "

Would really appreciate you reporting on this!

\n

Best,
\nJan

", "rawMarkdown": "Would really appreciate you reporting on this!\n\nBest,\nJan", "votes": null }, { "id": "2466902", "postDate": "10/04/2023 06:39:42", "content": "

Hi @takanashihumbert. Have you seen this post? The best thing to do is to try it.

", "rawMarkdown": "Hi @takanashihumbert. Have you seen [this post](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/438438)? The best thing to do is to try it.", "votes": null }, { "id": "2467226", "postDate": "10/04/2023 11:10:11", "content": "

False negatives will be punished (If a model makes a prediction for the invalid nights, it will hurt the score).

", "rawMarkdown": "False negatives will be punished (If a model makes a prediction for the invalid nights, it will hurt the score).", "votes": null }, { "id": "2467273", "postDate": "10/04/2023 12:02:54", "content": "

Thank you for pointing to this!

\n

Best,
\nJan

", "rawMarkdown": "Thank you for pointing to this!\n\nBest,\nJan", "votes": null } ], "comments": [ { "id": 2466191, "author_name": "janbrederecke", "author_url": "", "post_date": "10/03/2023 16:19:45", "content": "

Hey Joseph,

\n

I think you have to experiment with this and decide which way gives you better results in the end.
\nProbably no right or wrong if it generates good outcomes! If feel, this is what makes this competition interesting… :)

\n

Best,
\nJan

", "votes": null, "replies": [ { "id": 2466193, "author_name": "janbrederecke", "author_url": "", "post_date": "10/03/2023 16:21:00", "content": "

I mean, have you already compared the two approaches? And if so, can you report on your findings ? :)

", "votes": null, "replies": [ { "id": 2466196, "author_name": "takanashihumbert", "author_url": "", "post_date": "10/03/2023 16:26:46", "content": "

Have not yet. But I will have a try ASAP.

", "votes": null, "replies": [ { "id": 2466260, "author_name": "janbrederecke", "author_url": "", "post_date": "10/03/2023 17:14:12", "content": "

Would really appreciate you reporting on this!

\n

Best,
\nJan

", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2466902, "author_name": "mitsuyasuhoshino", "author_url": "", "post_date": "10/04/2023 06:39:42", "content": "

Hi @takanashihumbert. Have you seen this post? The best thing to do is to try it.

", "votes": null, "replies": [ { "id": 2467273, "author_name": "janbrederecke", "author_url": "", "post_date": "10/04/2023 12:02:54", "content": "

Thank you for pointing to this!

\n

Best,
\nJan

", "votes": null, "replies": [] } ] }, { "id": 2467226, "author_name": "tolgadincer", "author_url": "", "post_date": "10/04/2023 11:10:11", "content": "

False negatives will be punished (If a model makes a prediction for the invalid nights, it will hurt the score).

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2466152": "As we can see from the description:\n> Though each series is a continuous recording, there may be periods in the series when the accelerometer device was removed. These period are determined as those where suspiciously little variation in the accelerometer signals occur over an extended period of time, which is unrealistic for typical human participants. Events are not annotated for these periods, and you should attempt to refrain from making event predictions during these periods: an event prediction will be scored as false positive.\n\nThere are some 'invalid' periods. we shouldn't make our `onset` or `wakaup` in these periods.\nSo is this a binary classification, or a ternary classification? After I try to plot a series, like this one:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3110858%2Fee2d67e0c949e276280cab8a17db961b%2FWX20231003-2336512x.png?generation=1696347716631248&alt=media)\nwe can see the **black** zones might be the 'invalid' periods. should we label them to another class? If so, we can handle it like a ternary classification, with `onset=0`, `wakaup=1`, and `invalid=2`.", "2466191": "Hey Joseph,\n\nI think you have to experiment with this and decide which way gives you better results in the end.\nProbably no right or wrong if it generates good outcomes! If feel, this is what makes this competition interesting... :) \n\nBest,\nJan", "2466193": "I mean, have you already compared the two approaches? And if so, can you report on your findings ? :)", "2466196": "Have not yet. But I will have a try ASAP.", "2466260": "Would really appreciate you reporting on this!\n\nBest,\nJan", "2466902": "Hi @takanashihumbert. Have you seen [this post](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/438438)? The best thing to do is to try it.", "2467226": "False negatives will be punished (If a model makes a prediction for the invalid nights, it will hurt the score).", "2467273": "Thank you for pointing to this!\n\nBest,\nJan" }, "source": "meta" }