I read many topics about periods, when the user did not wear the device. Therefore, it is not a sleep state, nor the wakeup. But I still do not understand, how should we deal with them in the test data. How do I name these events in my submission file? Should I use NaNs or should I delete these rows?
\n\n", "messages": [ { "id": "2505866", "postDate": "10/30/2023 22:27:59", "content": "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
I read many topics about periods, when the user did not wear the device. Therefore, it is not a sleep state, nor the wakeup. But I still do not understand, how should we deal with them in the test data. How do I name these events in my submission file? Should I use NaNs or should I delete these rows?
\n\n", "rawMarkdown": "I read many topics about periods, when the user did not wear the device. Therefore, it is not a sleep state, nor the wakeup. But I still do not understand, how should we deal with them in the test data. How do I name these events in my submission file? Should I use NaNs or should I delete these rows?\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.", "votes": null }, { "id": "2505896", "postDate": "10/31/2023 00:16:23", "content": "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
Hi @manwithaflower , I posted about it here. I made this notebook to analyze some patterns.
", "rawMarkdown": "Hi @manwithaflower , I posted about it [here](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/451403). I made this [notebook](https://www.kaggle.com/code/renatoreggiani/zzzs-viz-for-non-wear-events) to analyze some patterns.", "votes": null }, { "id": "2505925", "postDate": "10/31/2023 01:21:01", "content": "Don't worry about naming those events in the submission file. They are only interested in \"onset\" and \"wakeup\" events, so if there is a time that you detect the device wasn't being worn, you don't have to report anything for that time.
", "rawMarkdown": "Don't worry about naming those events in the submission file. They are only interested in \"onset\" and \"wakeup\" events, so if there is a time that you detect the device wasn't being worn, you don't have to report anything for that time.", "votes": null }, { "id": "2508020", "postDate": "11/01/2023 12:03:29", "content": "It'd be really interesting to try classifying these as a separate event with your model to remove any false negatives or false positives in your predictions, but you'd need to construc the labels for these events yourself and like @chris62 said they're only interested in the onset and wakeup events
", "rawMarkdown": "It'd be really interesting to try classifying these as a separate event with your model to remove any false negatives or false positives in your predictions, but you'd need to construc the labels for these events yourself and like @chris62 said they're only interested in the onset and wakeup events", "votes": null }, { "id": "2510191", "postDate": "11/02/2023 19:24:05", "content": "Hello @manwithaflower! My solution was just to add an another extra-class (the third class), which indicates whether the child wears the device or not. And then, in submission, we indicate that extra-class as wakeup state (Or copy the states from the previous day). We added that extra-class to make the model not to confuse whether it's wakup or onset state on the samples where the child doesn't wear the device
", "rawMarkdown": "Hello @manwithaflower! My solution was just to add an another extra-class (the third class), which indicates whether the child wears the device or not. And then, in submission, we indicate that extra-class as wakeup state (Or copy the states from the previous day). We added that extra-class to make the model not to confuse whether it's wakup or onset state on the samples where the child doesn't wear the device", "votes": null } ], "comments": [ { "id": 2505896, "author_name": "renatoreggiani", "author_url": "", "post_date": "10/31/2023 00:16:23", "content": "Hi @manwithaflower , I posted about it here. I made this notebook to analyze some patterns.
", "votes": null, "replies": [] }, { "id": 2505925, "author_name": "chris62", "author_url": "", "post_date": "10/31/2023 01:21:01", "content": "Don't worry about naming those events in the submission file. They are only interested in \"onset\" and \"wakeup\" events, so if there is a time that you detect the device wasn't being worn, you don't have to report anything for that time.
", "votes": null, "replies": [] }, { "id": 2508020, "author_name": "judehunt23", "author_url": "", "post_date": "11/01/2023 12:03:29", "content": "It'd be really interesting to try classifying these as a separate event with your model to remove any false negatives or false positives in your predictions, but you'd need to construc the labels for these events yourself and like @chris62 said they're only interested in the onset and wakeup events
", "votes": null, "replies": [] }, { "id": 2510191, "author_name": "demko1", "author_url": "", "post_date": "11/02/2023 19:24:05", "content": "Hello @manwithaflower! My solution was just to add an another extra-class (the third class), which indicates whether the child wears the device or not. And then, in submission, we indicate that extra-class as wakeup state (Or copy the states from the previous day). We added that extra-class to make the model not to confuse whether it's wakup or onset state on the samples where the child doesn't wear the device
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2505866": "I read many topics about periods, when the user did not wear the device. Therefore, it is not a sleep state, nor the wakeup. But I still do not understand, how should we deal with them in the test data. How do I name these events in my submission file? Should I use NaNs or should I delete these rows?\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.", "2505896": "Hi @manwithaflower , I posted about it [here](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/451403). I made this [notebook](https://www.kaggle.com/code/renatoreggiani/zzzs-viz-for-non-wear-events) to analyze some patterns.", "2505925": "Don't worry about naming those events in the submission file. They are only interested in \"onset\" and \"wakeup\" events, so if there is a time that you detect the device wasn't being worn, you don't have to report anything for that time.", "2508020": "It'd be really interesting to try classifying these as a separate event with your model to remove any false negatives or false positives in your predictions, but you'd need to construc the labels for these events yourself and like @chris62 said they're only interested in the onset and wakeup events", "2510191": "Hello @manwithaflower! My solution was just to add an another extra-class (the third class), which indicates whether the child wears the device or not. And then, in submission, we indicate that extra-class as wakeup state (Or copy the states from the previous day). We added that extra-class to make the model not to confuse whether it's wakup or onset state on the samples where the child doesn't wear the device" }, "source": "meta" }