I prefer to write my own code from scratch because of OCD and I don't read public notebooks much. I knew that rle masks are supposed to be transposed while reading them in this competition but I didn't know we have to do that while submitting. My public lb score was bouncing around 0.1 and 0.3 for 2 weeks. When I transposed my predictions, I got 0.7+ public lb score. I lose two weeks of submissions just because I didn't read notebooks. Don't be like me.
", "messages": [ { "id": "1885336", "postDate": "08/05/2022 05:31:01", "content": "I prefer to write my own code from scratch because of OCD and I don't read public notebooks much. I knew that rle masks are supposed to be transposed while reading them in this competition but I didn't know we have to do that while submitting. My public lb score was bouncing around 0.1 and 0.3 for 2 weeks. When I transposed my predictions, I got 0.7+ public lb score. I lose two weeks of submissions just because I didn't read notebooks. Don't be like me.
", "rawMarkdown": "I prefer to write my own code from scratch because of OCD and I don't read public notebooks much. I knew that rle masks are supposed to be transposed while reading them in this competition but I didn't know we have to do that while submitting. My public lb score was bouncing around 0.1 and 0.3 for 2 weeks. When I transposed my predictions, I got 0.7+ public lb score. I lose two weeks of submissions just because I didn't read notebooks. Don't be like me.", "votes": null }, { "id": "1886146", "postDate": "08/05/2022 15:59:56", "content": "I have seen some of the inference NBs using transpose and some not.
\nI tested with one notebook but it did not work for me T-T, the model was previously giving 0.39 but after doing transpose [mask.numpy().T] it is giving 0.21. I guess my model was not that robust.
Thanks for sharing and explaining a mistake. Sorry this happened to you!
", "rawMarkdown": "Thanks for sharing and explaining a mistake. Sorry this happened to you!", "votes": null }, { "id": "1886740", "postDate": "08/06/2022 05:47:58", "content": "I think it depends at what level you are. I am a beginner so I rely a lot of public notebooks. At least it gives me a heads up and learning in the competition. If you are veteran and experienced then its ok to do coding by urself and not to worry about other people’s work.
", "rawMarkdown": "I think it depends at what level you are. I am a beginner so I rely a lot of public notebooks. At least it gives me a heads up and learning in the competition. If you are veteran and experienced then its ok to do coding by urself and not to worry about other people’s work.", "votes": null }, { "id": "1886892", "postDate": "08/06/2022 08:44:39", "content": "I used to watch twitch stream by rob mulla, starting of the competition he used to create his EDA, model by himself, w/o seeing other kernels. I think this helps to create your own personal opinion/view toward the competition without getting biased by other peoples kernels.
\nThis time it happened that he was not aware about transpose that public kernels were applying, but I bet there were many time when he was applying some small method[which he found out by independently working on it] that other people were unaware of which gave him the boost in the LB.
Both of you are right. @cid007 if you finished at least one project, I suggest you to write your own code from scratch. Try implementing ideas in your own way of coding style.
", "rawMarkdown": "Both of you are right. @cid007 if you finished at least one project, I suggest you to write your own code from scratch. Try implementing ideas in your own way of coding style.", "votes": null }, { "id": "1886942", "postDate": "08/06/2022 09:39:01", "content": "\"I am a beginner so I rely a lot of public notebooks.\" : human transfer learning
\n\"to write your own code\" : human learning from scratch
", "rawMarkdown": "\"I am a beginner so I rely a lot of public notebooks.\" : human transfer learning\n\n\"to write your own code\" : human learning from scratch", "votes": null } ], "comments": [ { "id": 1886146, "author_name": "soumya9977", "author_url": "", "post_date": "08/05/2022 15:59:56", "content": "I have seen some of the inference NBs using transpose and some not.
\nI tested with one notebook but it did not work for me T-T, the model was previously giving 0.39 but after doing transpose [mask.numpy().T] it is giving 0.21. I guess my model was not that robust.
Thanks for sharing and explaining a mistake. Sorry this happened to you!
", "votes": null, "replies": [] }, { "id": 1886740, "author_name": "cid007", "author_url": "", "post_date": "08/06/2022 05:47:58", "content": "I think it depends at what level you are. I am a beginner so I rely a lot of public notebooks. At least it gives me a heads up and learning in the competition. If you are veteran and experienced then its ok to do coding by urself and not to worry about other people’s work.
", "votes": null, "replies": [ { "id": 1886892, "author_name": "soumya9977", "author_url": "", "post_date": "08/06/2022 08:44:39", "content": "I used to watch twitch stream by rob mulla, starting of the competition he used to create his EDA, model by himself, w/o seeing other kernels. I think this helps to create your own personal opinion/view toward the competition without getting biased by other peoples kernels.
\nThis time it happened that he was not aware about transpose that public kernels were applying, but I bet there were many time when he was applying some small method[which he found out by independently working on it] that other people were unaware of which gave him the boost in the LB.
Both of you are right. @cid007 if you finished at least one project, I suggest you to write your own code from scratch. Try implementing ideas in your own way of coding style.
", "votes": null, "replies": [] }, { "id": 1886942, "author_name": "hengck23", "author_url": "", "post_date": "08/06/2022 09:39:01", "content": "\"I am a beginner so I rely a lot of public notebooks.\" : human transfer learning
\n\"to write your own code\" : human learning from scratch
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1885336": "I prefer to write my own code from scratch because of OCD and I don't read public notebooks much. I knew that rle masks are supposed to be transposed while reading them in this competition but I didn't know we have to do that while submitting. My public lb score was bouncing around 0.1 and 0.3 for 2 weeks. When I transposed my predictions, I got 0.7+ public lb score. I lose two weeks of submissions just because I didn't read notebooks. Don't be like me.", "1886146": "I have seen some of the inference NBs using transpose and some not.\nI tested with one notebook but it did not work for me `T-T`, the model was previously giving `0.39` but after doing transpose [`mask.numpy().T`] it is giving `0.21`. I guess my model was not that robust.", "1886154": "Thanks for sharing and explaining a mistake. Sorry this happened to you!", "1886740": "I think it depends at what level you are. I am a beginner so I rely a lot of public notebooks. At least it gives me a heads up and learning in the competition. If you are veteran and experienced then its ok to do coding by urself and not to worry about other people’s work.", "1886892": "I used to watch twitch stream by rob mulla, starting of the competition he used to create his EDA, model by himself, w/o seeing other kernels. I think this helps to create your own personal opinion/view toward the competition without getting biased by other peoples kernels. \nThis time it happened that he was not aware about transpose that public kernels were applying, but I bet there were many time when he was applying some small method[which he found out by independently working on it] that other people were unaware of which gave him the boost in the LB.", "1886935": "Both of you are right. @cid007 if you finished at least one project, I suggest you to write your own code from scratch. Try implementing ideas in your own way of coding style.", "1886942": "\"I am a beginner so I rely a lot of public notebooks.\" : human transfer learning\n\n\"to write your own code\" : human learning from scratch" }, "source": "meta" }