As many people has pointed out, there are lots of NaNs in train events data. I used to think that it means no meaningful events detected at that night. However, there are also series like 31011ade7c0a and a596ad0b82aa, that have no NaNs yet a long label-less period.
\nSo what does NaN really means? what is the difference between NaNs and label-less period?
\nps: there are also series with NaNs and a long label-less period, like 05e1944c3818 and 10469f6765bf.
", "messages": [ { "id": "2452401", "postDate": "09/23/2023 08:59:22", "content": "As many people has pointed out, there are lots of NaNs in train events data. I used to think that it means no meaningful events detected at that night. However, there are also series like 31011ade7c0a and a596ad0b82aa, that have no NaNs yet a long label-less period.
\nSo what does NaN really means? what is the difference between NaNs and label-less period?
\nps: there are also series with NaNs and a long label-less period, like 05e1944c3818 and 10469f6765bf.
", "rawMarkdown": "As many people has pointed out, there are lots of NaNs in train events data. I used to think that it means no meaningful events detected at that night. However, there are also series like 31011ade7c0a and a596ad0b82aa, that have no NaNs yet a long label-less period. \n\nSo what does NaN really means? what is the difference between NaNs and label-less period?\n\nps: there are also series with NaNs and a long label-less period, like 05e1944c3818 and 10469f6765bf.", "votes": null }, { "id": "2452678", "postDate": "09/23/2023 13:55:46", "content": "Yes, NaNs mean there were no events for that night (most likely, that the watch was taken off for the night). Also yes, there are some (what I think are) mislabelled series like 31011ade7c0a and a596ad0b82aa.
", "rawMarkdown": "Yes, NaNs mean there were no events for that night (most likely, that the watch was taken off for the night). Also yes, there are some (what I think are) mislabelled series like 31011ade7c0a and a596ad0b82aa.", "votes": null }, { "id": "2452762", "postDate": "09/23/2023 15:02:46", "content": "I think that make some sense if NaN and label-less period have no co-occurrence. However, there are also series with both present, like 05e1944c3818 and 10469f6765bf.
", "rawMarkdown": "I think that make some sense if NaN and label-less period have no co-occurrence. However, there are also series with both present, like 05e1944c3818 and 10469f6765bf.", "votes": null }, { "id": "2452850", "postDate": "09/23/2023 16:05:06", "content": "Hm. It may come down to this line:
\n\"No sleep windows can be detected unless the watch is deemed to be worn for the duration\"
\nIn 31011ade7c0a in particular, that may explain some of what I thought was mislabelled… (explains some of the others as well perhaps)
", "rawMarkdown": "Hm. It may come down to this line:\n\n\"No sleep windows can be detected unless the watch is deemed to be worn for the duration\"\n\nIn 31011ade7c0a in particular, that may explain some of what I thought was mislabelled... (explains some of the others as well perhaps)", "votes": null }, { "id": "2452971", "postDate": "09/23/2023 17:23:21", "content": "The NaN periods are within the consecutive labeled nights starting from night one -- these are given in the events file. It seems that any label-less periods come after these and so are at the end of the series, as if the rater-person just got tired of labeling the events. There are other posts describing unlabeled regions of series.
\nWhat I don't think happens is having an unlabeled night (or nights) with labeled nights before and after.
", "rawMarkdown": "The NaN periods are within the consecutive labeled nights starting from night one -- these are given in the events file. It seems that any label-less periods come after these and so are at the end of the series, as if the rater-person just got tired of labeling the events. There are other posts describing unlabeled regions of series.\n\nWhat I don't think happens is having an unlabeled night (or nights) with labeled nights before and after.", "votes": null }, { "id": "2453038", "postDate": "09/23/2023 18:47:58", "content": "It seems that unlabeled nights do appear at the end of the series, maybe some inconsistence in labeling I guess.
", "rawMarkdown": "It seems that unlabeled nights do appear at the end of the series, maybe some inconsistence in labeling I guess.", "votes": null }, { "id": "2459005", "postDate": "09/28/2023 01:34:40", "content": "@aphysict i made some analysis on these events , it's stated that a sleep period must be at least 30 minutes long. After checking, it seems that these missing sleep periods meet this requirement, as they appear to be longer than 30 minutes.
\nI dont think so its an issue of mislabelling see my updated notebook section Simple Analysis For Missing Steps
", "rawMarkdown": "aphysict i made some analysis on these events , it's stated that a sleep period must be at least 30 minutes long. After checking, it seems that these missing sleep periods meet this requirement, as they appear to be longer than 30 minutes. \n\nI dont think so its an issue of mislabelling see my updated notebook section [Simple Analysis For Missing Steps](https://www.kaggle.com/code/oraware/dask-data-processing-with-train-series)", "votes": null }, { "id": "2460715", "postDate": "09/29/2023 04:15:26", "content": "Hello, do you know (or have the host noticed) how the evaluation code works when real test events(label) contain NAN like train dataset?
\nThey simply ignore these samples or set these samples as awake?
\nThanks.
No idea, I guess they drop nans. Even for the train data, I don't know how to get score=1 with nan included. Hope @ryanholbrook can share some information.
", "rawMarkdown": "No idea, I guess they drop nans. Even for the train data, I don't know how to get score=1 with nan included. Hope @ryanholbrook can share some information.", "votes": null } ], "comments": [ { "id": 2452678, "author_name": "chris62", "author_url": "", "post_date": "09/23/2023 13:55:46", "content": "Yes, NaNs mean there were no events for that night (most likely, that the watch was taken off for the night). Also yes, there are some (what I think are) mislabelled series like 31011ade7c0a and a596ad0b82aa.
", "votes": null, "replies": [ { "id": 2452762, "author_name": "aphysict", "author_url": "", "post_date": "09/23/2023 15:02:46", "content": "I think that make some sense if NaN and label-less period have no co-occurrence. However, there are also series with both present, like 05e1944c3818 and 10469f6765bf.
", "votes": null, "replies": [ { "id": 2452850, "author_name": "chris62", "author_url": "", "post_date": "09/23/2023 16:05:06", "content": "Hm. It may come down to this line:
\n\"No sleep windows can be detected unless the watch is deemed to be worn for the duration\"
\nIn 31011ade7c0a in particular, that may explain some of what I thought was mislabelled… (explains some of the others as well perhaps)
", "votes": null, "replies": [] }, { "id": 2452971, "author_name": "dan3dewey", "author_url": "", "post_date": "09/23/2023 17:23:21", "content": "The NaN periods are within the consecutive labeled nights starting from night one -- these are given in the events file. It seems that any label-less periods come after these and so are at the end of the series, as if the rater-person just got tired of labeling the events. There are other posts describing unlabeled regions of series.
\nWhat I don't think happens is having an unlabeled night (or nights) with labeled nights before and after.
", "votes": null, "replies": [] }, { "id": 2453038, "author_name": "aphysict", "author_url": "", "post_date": "09/23/2023 18:47:58", "content": "It seems that unlabeled nights do appear at the end of the series, maybe some inconsistence in labeling I guess.
", "votes": null, "replies": [] } ] } ] }, { "id": 2459005, "author_name": "oraware", "author_url": "", "post_date": "09/28/2023 01:34:40", "content": "@aphysict i made some analysis on these events , it's stated that a sleep period must be at least 30 minutes long. After checking, it seems that these missing sleep periods meet this requirement, as they appear to be longer than 30 minutes.
\nI dont think so its an issue of mislabelling see my updated notebook section Simple Analysis For Missing Steps
", "votes": null, "replies": [] }, { "id": 2460715, "author_name": "zui0711", "author_url": "", "post_date": "09/29/2023 04:15:26", "content": "Hello, do you know (or have the host noticed) how the evaluation code works when real test events(label) contain NAN like train dataset?
\nThey simply ignore these samples or set these samples as awake?
\nThanks.
No idea, I guess they drop nans. Even for the train data, I don't know how to get score=1 with nan included. Hope @ryanholbrook can share some information.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2452401": "As many people has pointed out, there are lots of NaNs in train events data. I used to think that it means no meaningful events detected at that night. However, there are also series like 31011ade7c0a and a596ad0b82aa, that have no NaNs yet a long label-less period. \n\nSo what does NaN really means? what is the difference between NaNs and label-less period?\n\nps: there are also series with NaNs and a long label-less period, like 05e1944c3818 and 10469f6765bf.", "2452678": "Yes, NaNs mean there were no events for that night (most likely, that the watch was taken off for the night). Also yes, there are some (what I think are) mislabelled series like 31011ade7c0a and a596ad0b82aa.", "2452762": "I think that make some sense if NaN and label-less period have no co-occurrence. However, there are also series with both present, like 05e1944c3818 and 10469f6765bf.", "2452850": "Hm. It may come down to this line:\n\n\"No sleep windows can be detected unless the watch is deemed to be worn for the duration\"\n\nIn 31011ade7c0a in particular, that may explain some of what I thought was mislabelled... (explains some of the others as well perhaps)", "2452971": "The NaN periods are within the consecutive labeled nights starting from night one -- these are given in the events file. It seems that any label-less periods come after these and so are at the end of the series, as if the rater-person just got tired of labeling the events. There are other posts describing unlabeled regions of series.\n\nWhat I don't think happens is having an unlabeled night (or nights) with labeled nights before and after.", "2453038": "It seems that unlabeled nights do appear at the end of the series, maybe some inconsistence in labeling I guess.", "2459005": "aphysict i made some analysis on these events , it's stated that a sleep period must be at least 30 minutes long. After checking, it seems that these missing sleep periods meet this requirement, as they appear to be longer than 30 minutes. \n\nI dont think so its an issue of mislabelling see my updated notebook section [Simple Analysis For Missing Steps](https://www.kaggle.com/code/oraware/dask-data-processing-with-train-series)", "2460715": "Hello, do you know (or have the host noticed) how the evaluation code works when real test events(label) contain NAN like train dataset?\nThey simply ignore these samples or set these samples as awake?\nThanks.", "2461133": "No idea, I guess they drop nans. Even for the train data, I don't know how to get score=1 with nan included. Hope @ryanholbrook can share some information." }, "source": "meta" }