I understand that there is a period of time when the child is off the clock, how should I process the data for that period of time?
", "messages": [ { "id": "2469470", "postDate": "10/06/2023 11:21:57", "content": "I understand that there is a period of time when the child is off the clock, how should I process the data for that period of time?
", "rawMarkdown": "I understand that there is a period of time when the child is off the clock, how should I process the data for that period of time?", "votes": null }, { "id": "2473739", "postDate": "10/08/2023 14:36:34", "content": "@daichi1412 : Deleting such data observations could be one option to go. However, it should be justified against the testing set - if the testing records have the series fragments with the child(s) to be off the clock, it would mean the trained models may mis-score the events under such a condition.
\nAnother option, as per the suggestion of @carlmcbrideellis , it to turn the contest problem into a multi-class tabular classification problem. You can see Carl's rationale behind it in the respective discussion thread. On a separate note, Carl prepared and the publicly contributed a new preprocessed dataset for the multi-class classification setup.
\nAs for me, I have not tried the multi-class prediction yet. However, it sounds promising to me to give it a try down the road.
\nI hope it is helpful.
", "rawMarkdown": "daichi1412 : Deleting such data observations could be one option to go. However, it should be justified against the testing set - if the testing records have the series fragments with the child(s) to be off the clock, it would mean the trained models may mis-score the events under such a condition.\n\nAnother option, as per the suggestion of @carlmcbrideellis , it to turn the contest problem into a multi-class tabular classification problem. You can see Carl's rationale behind it in the [respective discussion thread](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/438438). On a separate note, Carl prepared and the publicly contributed a new preprocessed [dataset](https://www.kaggle.com/datasets/carlmcbrideellis/zzzs-lightweight-training-dataset-target) for the multi-class classification setup.\n\nAs for me, I have not tried the multi-class prediction yet. However, it sounds promising to me to give it a try down the road.\n\nI hope it is helpful.", "votes": null } ], "comments": [ { "id": 2473739, "author_name": "gvyshnya", "author_url": "", "post_date": "10/08/2023 14:36:34", "content": "@daichi1412 : Deleting such data observations could be one option to go. However, it should be justified against the testing set - if the testing records have the series fragments with the child(s) to be off the clock, it would mean the trained models may mis-score the events under such a condition.
\nAnother option, as per the suggestion of @carlmcbrideellis , it to turn the contest problem into a multi-class tabular classification problem. You can see Carl's rationale behind it in the respective discussion thread. On a separate note, Carl prepared and the publicly contributed a new preprocessed dataset for the multi-class classification setup.
\nAs for me, I have not tried the multi-class prediction yet. However, it sounds promising to me to give it a try down the road.
\nI hope it is helpful.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2469470": "I understand that there is a period of time when the child is off the clock, how should I process the data for that period of time?", "2473739": "daichi1412 : Deleting such data observations could be one option to go. However, it should be justified against the testing set - if the testing records have the series fragments with the child(s) to be off the clock, it would mean the trained models may mis-score the events under such a condition.\n\nAnother option, as per the suggestion of @carlmcbrideellis , it to turn the contest problem into a multi-class tabular classification problem. You can see Carl's rationale behind it in the [respective discussion thread](https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/438438). On a separate note, Carl prepared and the publicly contributed a new preprocessed [dataset](https://www.kaggle.com/datasets/carlmcbrideellis/zzzs-lightweight-training-dataset-target) for the multi-class classification setup.\n\nAs for me, I have not tried the multi-class prediction yet. However, it sounds promising to me to give it a try down the road.\n\nI hope it is helpful." }, "source": "meta" }