I’m trying to combine the ideas from two notebooks. I used the ADC_convert function from this notebook within the read_and_preprocess function from this other notebook. However, my notebook ended up with a worse public score. Does this make sense? Theoretically, shouldn’t calibration improve the results? I’d appreciate any help with this. Thanks in advance!
Hi, I wonder if the issue comes from the function load signal before ADC convert during your preprocess. Maybe you should convert the original signal with a load function without any transformation inside ?
", "rawMarkdown": "Hi, I wonder if the issue comes from the function load signal before ADC convert during your preprocess. Maybe you should convert the original signal with a load function without any transformation inside ?", "votes": 2, "replies": [ { "id": 2972095, "postDate": "2024-08-28T01:38:26.713Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2977535, "postDate": "2024-09-03T00:34:19.477Z", "content": "I think the problem with the notebook has been addressed to some degree. Thank you very much for your help.
", "rawMarkdown": "I think the problem with the notebook has been addressed to some degree. Thank you very much for your help." } ] }, { "id": 2971219, "postDate": "2024-08-27T01:05:20.730Z", "content": "Does it make sense? No. I used the normalizations, and the effect was small, and there was definitely no decrease of >0.1. The good news is that the issue is not with test part implementation/test 'surprise'- look at the 'correlation between relative signal reduction and target' plot. For the original notebook, there is a nice ~linear plot. On the other hand, yours is crazy, with some extreme outliers. Clearly, something is not working as expected…I am intrigued; if you find the problem, please share lol.
", "rawMarkdown": "Does it make sense? No. I used the normalizations, and the effect was small, and there was definitely no decrease of >0.1. The good news is that the issue is not with test part implementation/test 'surprise'- look at the 'correlation between relative signal reduction and target' plot. For the original notebook, there is a nice ~linear plot. On the other hand, yours is crazy, with some extreme outliers. Clearly, something is not working as expected...I am intrigued; if you find the problem, please share lol.", "votes": 2, "replies": [ { "id": 2972096, "postDate": "2024-08-28T01:41:00.757Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2977536, "postDate": "2024-09-03T00:34:52.977Z", "content": "I believe the issue with the notebook has been partially resolved. Thank you very much for your help.
", "rawMarkdown": "I believe the issue with the notebook has been partially resolved. Thank you very much for your help." } ] }, { "id": 2971148, "postDate": "2024-08-26T21:55:19.700Z", "content": "I’m trying to combine the ideas from two notebooks. I used the ADC_convert function from this notebook within the read_and_preprocess function from this other notebook. However, my notebook ended up with a worse public score. Does this make sense? Theoretically, shouldn’t calibration improve the results? I’d appreciate any help with this. Thanks in advance!
Hi, I wonder if the issue comes from the function load signal before ADC convert during your preprocess. Maybe you should convert the original signal with a load function without any transformation inside ?
", "votes": 2, "replies": [ { "id": 2972095, "author_name": "", "author_url": "", "post_date": "2024-08-28T01:38:26.713000", "content": "", "votes": 0, "replies": [] }, { "id": 2977535, "author_name": "Regis Vargas", "author_url": "", "post_date": "2024-09-03T00:34:19.477000", "content": "I think the problem with the notebook has been addressed to some degree. Thank you very much for your help.
", "votes": 0, "replies": [] } ] }, { "id": 2971219, "author_name": "greySnow", "author_url": "", "post_date": "2024-08-27T01:05:20.730000", "content": "Does it make sense? No. I used the normalizations, and the effect was small, and there was definitely no decrease of >0.1. The good news is that the issue is not with test part implementation/test 'surprise'- look at the 'correlation between relative signal reduction and target' plot. For the original notebook, there is a nice ~linear plot. On the other hand, yours is crazy, with some extreme outliers. Clearly, something is not working as expected…I am intrigued; if you find the problem, please share lol.
", "votes": 2, "replies": [ { "id": 2972096, "author_name": "", "author_url": "", "post_date": "2024-08-28T01:41:00.757000", "content": "", "votes": 0, "replies": [] }, { "id": 2977536, "author_name": "Regis Vargas", "author_url": "", "post_date": "2024-09-03T00:34:52.977000", "content": "I believe the issue with the notebook has been partially resolved. Thank you very much for your help.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2971375": "Hi, I wonder if the issue comes from the function load signal before ADC convert during your preprocess. Maybe you should convert the original signal with a load function without any transformation inside ?", "2971219": "Does it make sense? No. I used the normalizations, and the effect was small, and there was definitely no decrease of >0.1. The good news is that the issue is not with test part implementation/test 'surprise'- look at the 'correlation between relative signal reduction and target' plot. For the original notebook, there is a nice ~linear plot. On the other hand, yours is crazy, with some extreme outliers. Clearly, something is not working as expected...I am intrigued; if you find the problem, please share lol.", "2971148": "I’m trying to combine the ideas from two notebooks. I used the `ADC_convert` function from [this notebook](https://www.kaggle.com/code/gordonyip/update-calibrating-and-binning-astronomical-data) within the `read_and_preprocess` function from [this other notebook](https://www.kaggle.com/code/bingyuniu/neurips-ariel-2024-starter-withdifferentparametr). However, my [notebook](https://www.kaggle.com/code/regisvargas/fork-of-neurips-ariel-2024-starter-5be123) ended up with a worse public score. Does this make sense? Theoretically, shouldn’t calibration improve the results? I’d appreciate any help with this. Thanks in advance!" } }