=> Any approach to fix the zero submission scores? ( except trail and error sigma approach )
", "messages": [ { "id": 2986173, "postDate": "2024-09-11T13:02:45.970Z", "content": "I just fix my sigma to be 1.5e-4, and I never have any zero submission score after fixing the value.
", "rawMarkdown": "I just fix my sigma to be 1.5e-4, and I never have any zero submission score after fixing the value.", "votes": 4, "replies": [ { "id": 2986179, "postDate": "2024-09-11T13:05:44.063Z", "content": "Thanks, i will try with 1.5e-4 @chingyinng
", "rawMarkdown": "Thanks, i will try with **1.5e-4** @chingyinng " }, { "id": 2993669, "postDate": "2024-09-20T05:23:29.430Z", "content": "Thanks @seshurajup for raising this issue and thanks @chingyinng for the answer, the problem is indeed frustrating.
\nDo you guys mean sigma defined at the starter's notebook https://www.kaggle.com/code/gordonyip/update-calibrating-and-binning-astronomical-data and used for astropy.stats.sigma_clip?
@andrey885 Not really. We are talking about a different sigma.
\nThe sigma in sigma clipping is used to identify hot pixels, that has a pixel value outside 5 standard deviations. The sigma that we are talking about is the uncertainty estimation, that you have to submit with submission.csv.
Hi ChingYinNg,
\nThis is my first try on kaggle and I have 3 tries with score 0.000.
\nI changed my sigma for constant 1.5e-4 and still have the same score. I am just wondering if it because of error in my sbmission file or because results are very sensitive. I think that my error should not be more than 10% to 20%. Is this leading to 0.000 score?
@mjz1977 Its probably because your model’s error is >> 1e-4. Try to increase the sigma to match your model’s uncertainty. For my first attempts I actually used 8e-4 as the sigma. I changed it to 1.5e-4 after improving the model.
", "rawMarkdown": "@mjz1977 Its probably because your model’s error is >> 1e-4. Try to increase the sigma to match your model’s uncertainty. For my first attempts I actually used 8e-4 as the sigma. I changed it to 1.5e-4 after improving the model.", "votes": 1 }, { "id": 2998137, "postDate": "2024-09-25T09:37:35.800Z", "content": "I found the error. I put the answer here because it can helps kaggle beginners like me 😅.
\nThe real (and hidden) test database is more than 1 planet. My error was that I write the code as if the there is only one planet in the test database.
=> Any approach to fix the zero submission scores? ( except trail and error sigma approach )
", "rawMarkdown": "=> Any approach to fix the zero submission scores? ( except trail and error sigma approach )", "votes": 4 } ], "comments": [ { "id": 2986173, "author_name": "ChingYinNg", "author_url": "", "post_date": "2024-09-11T13:02:45.970000", "content": "I just fix my sigma to be 1.5e-4, and I never have any zero submission score after fixing the value.
", "votes": 4, "replies": [ { "id": 2986179, "author_name": "SeshuRaju 🧘♂️", "author_url": "", "post_date": "2024-09-11T13:05:44.063000", "content": "Thanks, i will try with 1.5e-4 @chingyinng
", "votes": 0, "replies": [] }, { "id": 2993669, "author_name": "Andrey", "author_url": "", "post_date": "2024-09-20T05:23:29.430000", "content": "Thanks @seshurajup for raising this issue and thanks @chingyinng for the answer, the problem is indeed frustrating.
\nDo you guys mean sigma defined at the starter's notebook https://www.kaggle.com/code/gordonyip/update-calibrating-and-binning-astronomical-data and used for astropy.stats.sigma_clip?
@andrey885 Not really. We are talking about a different sigma.
\nThe sigma in sigma clipping is used to identify hot pixels, that has a pixel value outside 5 standard deviations. The sigma that we are talking about is the uncertainty estimation, that you have to submit with submission.csv.
Hi ChingYinNg,
\nThis is my first try on kaggle and I have 3 tries with score 0.000.
\nI changed my sigma for constant 1.5e-4 and still have the same score. I am just wondering if it because of error in my sbmission file or because results are very sensitive. I think that my error should not be more than 10% to 20%. Is this leading to 0.000 score?
@mjz1977 Its probably because your model’s error is >> 1e-4. Try to increase the sigma to match your model’s uncertainty. For my first attempts I actually used 8e-4 as the sigma. I changed it to 1.5e-4 after improving the model.
", "votes": 1, "replies": [] }, { "id": 2998137, "author_name": "MJZ1977", "author_url": "", "post_date": "2024-09-25T09:37:35.800000", "content": "I found the error. I put the answer here because it can helps kaggle beginners like me 😅.
\nThe real (and hidden) test database is more than 1 planet. My error was that I write the code as if the there is only one planet in the test database.