A lot of smart people have spent a lot of time studying the strengths and weaknesses of this dataset. Maybe the leaks can be solved with quick fixes. Maybe not. This seems like as good a time as ever for members to weigh in on possible improvements.
\nTo try to get the ball rolling, I'll make a point of my own. The task is to find interesting, novel signatures (a,k,a \"needles\") that show up in the on channels, but not the off channels. The organizers created numerous positive examples by adding needles to the on-channels. I don't think they created negative examples by adding needles to both on and off channels. (This notebook https://www.kaggle.com/glazed/eda-looking-for-needles-in-all-the-wrong-places shows my reasoning.) The SETI problem needs models that know to reject needles that also show up in off channels, but I question whether this dataset will lead to them learning that.
", "messages": [ { "id": 1354498, "postDate": "2021-06-17T16:07:45.740Z", "content": "A lot of smart people have spent a lot of time studying the strengths and weaknesses of this dataset. Maybe the leaks can be solved with quick fixes. Maybe not. This seems like as good a time as ever for members to weigh in on possible improvements.
\nTo try to get the ball rolling, I'll make a point of my own. The task is to find interesting, novel signatures (a,k,a \"needles\") that show up in the on channels, but not the off channels. The organizers created numerous positive examples by adding needles to the on-channels. I don't think they created negative examples by adding needles to both on and off channels. (This notebook https://www.kaggle.com/glazed/eda-looking-for-needles-in-all-the-wrong-places shows my reasoning.) The SETI problem needs models that know to reject needles that also show up in off channels, but I question whether this dataset will lead to them learning that.
", "rawMarkdown": "A lot of smart people have spent a lot of time studying the strengths and weaknesses of this dataset. Maybe the leaks can be solved with quick fixes. Maybe not. This seems like as good a time as ever for members to weigh in on possible improvements.\n\nTo try to get the ball rolling, I'll make a point of my own. The task is to find interesting, novel signatures (a,k,a \"needles\") that show up in the on channels, but not the off channels. The organizers created numerous positive examples by adding needles to the on-channels. I don't think they created negative examples by adding needles to both on and off channels. (This notebook https://www.kaggle.com/glazed/eda-looking-for-needles-in-all-the-wrong-places shows my reasoning.) The SETI problem needs models that know to reject needles that also show up in off channels, but I question whether this dataset will lead to them learning that.", "votes": 7 }, { "id": 1358707, "postDate": "2021-06-20T17:15:19.780Z", "content": "If only every tenth or so found needle-like sample is a false positive in real life, then it's not big deal to check them manually, at least for now, I think.
\nBut if there are a lot of such examples that can be identified based on off-samples then yes, it could be great value for researchers.
", "rawMarkdown": "If only every tenth or so found needle-like sample is a false positive in real life, then it's not big deal to check them manually, at least for now, I think.\n\nBut if there are a lot of such examples that can be identified based on off-samples then yes, it could be great value for researchers.", "votes": 1 }, { "id": 1389394, "postDate": "2021-07-15T17:06:48.420Z", "content": "I updated my notebook to take a quick and dirty look at the new dataset.
", "rawMarkdown": "I updated my notebook to take a quick and dirty look at the new dataset." } ], "comments": [ { "id": 1358707, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-06-20T17:15:19.780000", "content": "If only every tenth or so found needle-like sample is a false positive in real life, then it's not big deal to check them manually, at least for now, I think.
\nBut if there are a lot of such examples that can be identified based on off-samples then yes, it could be great value for researchers.
", "votes": 1, "replies": [] }, { "id": 1389394, "author_name": "glazed", "author_url": "", "post_date": "2021-07-15T17:06:48.420000", "content": "I updated my notebook to take a quick and dirty look at the new dataset.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1354498": "A lot of smart people have spent a lot of time studying the strengths and weaknesses of this dataset. Maybe the leaks can be solved with quick fixes. Maybe not. This seems like as good a time as ever for members to weigh in on possible improvements.\n\nTo try to get the ball rolling, I'll make a point of my own. The task is to find interesting, novel signatures (a,k,a \"needles\") that show up in the on channels, but not the off channels. The organizers created numerous positive examples by adding needles to the on-channels. I don't think they created negative examples by adding needles to both on and off channels. (This notebook https://www.kaggle.com/glazed/eda-looking-for-needles-in-all-the-wrong-places shows my reasoning.) The SETI problem needs models that know to reject needles that also show up in off channels, but I question whether this dataset will lead to them learning that.", "1358707": "If only every tenth or so found needle-like sample is a false positive in real life, then it's not big deal to check them manually, at least for now, I think.\n\nBut if there are a lot of such examples that can be identified based on off-samples then yes, it could be great value for researchers.", "1389394": "I updated my notebook to take a quick and dirty look at the new dataset." } }