Nice to meet you.
\nI'm a beginner.I have created simple models for several competitions.
\nI don't know how to proceed with this kind of competition, so I would like to ask you something.
\nNot limited to this competition, but how do you achieve results? I feel that there are many people who are able to produce analytical results even without specialized knowledge. Since I can't really participate in the discussions, I would like to ask you all, how do you achieve results? I would like to know how to proceed and how to think.
\nI tried to refer to similar competitions in the past, but I couldn't find anything that looked like it.
Similar competitions in last year: Ribonanza, American Sign Language Fingerspelling Recognition, Isolated Sign Language Recognition, and iceCube. Each competition is unique, but all of those competitions deal with 1D data (i.e., things you can put into LSTM/GRU/Transformers) and massive data sets. Not coincidentally, I have a silver and two golds in 3 of these competitions, and this past knowledge is why I'm doing well in this competition, too.
", "rawMarkdown": "Similar competitions in last year: Ribonanza, American Sign Language Fingerspelling Recognition, Isolated Sign Language Recognition, and iceCube. Each competition is unique, but all of those competitions deal with 1D data (i.e., things you can put into LSTM/GRU/Transformers) and massive data sets. Not coincidentally, I have a silver and two golds in 3 of these competitions, and this past knowledge is why I'm doing well in this competition, too.", "votes": 11, "replies": [ { "id": 2889482, "postDate": "2024-06-25T14:22:20.187Z", "content": "@shlomoron, do you think that this competition will help with model architecture?
\nSome days ago, I was looking at some past competitions, but they were not very useful. I will check the competitions you mentioned.
Try and find out :)
\nOf course, no competition is the same. Ribonanza is a bit tricky since it had also edge data, ASLFR was encoder-decoder kind of competiton and iceCube data was the kind of data that encourages GNNs solutions.
\nBut the similarities exist and experience and ideas from those past competitions helps.
thank you for your kindness.
", "rawMarkdown": "thank you for your kindness." }, { "id": 2891891, "postDate": "2024-06-27T01:39:22.557Z", "content": "Your suggestion is very helpful! I have tried model posted in Isolated Sign Language Recognition, which is dealing with 1d data, but my local cv cannot achieve good results. How much domain knowledge helps to improve model architecture?
", "rawMarkdown": "Your suggestion is very helpful! I have tried model posted in Isolated Sign Language Recognition, which is dealing with 1d data, but my local cv cannot achieve good results. How much domain knowledge helps to improve model architecture?", "replies": [ { "id": 2892722, "postDate": "2024-06-27T11:39:36.407Z", "content": "If 'domain knowledge' = atmospheric physics, None. If 'domain knowledge' = deep learning tricks, architectures, finetuning and best practices- a lot. Don't expect to copy paste, plug in and get good result out of the box…
", "rawMarkdown": "If 'domain knowledge' = atmospheric physics, None. If 'domain knowledge' = deep learning tricks, architectures, finetuning and best practices- a lot. Don't expect to copy paste, plug in and get good result out of the box...", "votes": 1 }, { "id": 2893673, "postDate": "2024-06-28T01:53:05.670Z", "content": "Thanks for sharing! I just start from copy and paste and learned a lot from your codes in previous competitions🤣. By modifing model architectures I can see improvements step by step.
Nice to meet you.
\nI'm a beginner.I have created simple models for several competitions.
\nI don't know how to proceed with this kind of competition, so I would like to ask you something.
\nNot limited to this competition, but how do you achieve results? I feel that there are many people who are able to produce analytical results even without specialized knowledge. Since I can't really participate in the discussions, I would like to ask you all, how do you achieve results? I would like to know how to proceed and how to think.
\nI tried to refer to similar competitions in the past, but I couldn't find anything that looked like it.
First, try to create a model by yourself without copying someone else's notebook. This is the best way to learn.
\nAfter you get stuck, I usually go to the code section and pick the best code that is clear to read and understand. In this competition, I think there are two (maybe more) awesome notebooks: Giba's and Amadeo's.
\nThese two are really good for learning and trying to pursue a medal.
After these notebooks are completely understood, go to the internet and the Discussion section to find new ideas.
\nTo understand both notebooks and if you have an unanswered question, ask ChatGPT.
", "rawMarkdown": "First, try to create a model by yourself without copying someone else's notebook. This is the best way to learn.\n\nAfter you get stuck, I usually go to the code section and pick the best code that is clear to read and understand. In this competition, I think there are two (maybe more) awesome notebooks: [Giba's](https://www.kaggle.com/code/titericz/giba-baseline-xgboost) and [Amadeo's](https://www.kaggle.com/code/abiolatti/keras-baseline-seq2seq).\nThese two are really good for learning and trying to pursue a medal.\n\nAfter these notebooks are completely understood, go to the internet and the Discussion section to find new ideas.\n\nTo understand both notebooks and if you have an unanswered question, ask ChatGPT.\n\n", "replies": [ { "id": 2890171, "postDate": "2024-06-26T01:28:23.857Z", "content": "Thank you very much
", "rawMarkdown": "Thank you very much" } ] } ], "comments": [ { "id": 2889472, "author_name": "greySnow", "author_url": "", "post_date": "2024-06-25T14:18:06.850000", "content": "Similar competitions in last year: Ribonanza, American Sign Language Fingerspelling Recognition, Isolated Sign Language Recognition, and iceCube. Each competition is unique, but all of those competitions deal with 1D data (i.e., things you can put into LSTM/GRU/Transformers) and massive data sets. Not coincidentally, I have a silver and two golds in 3 of these competitions, and this past knowledge is why I'm doing well in this competition, too.
", "votes": 11, "replies": [ { "id": 2889482, "author_name": "Fernando Melo", "author_url": "", "post_date": "2024-06-25T14:22:20.187000", "content": "@shlomoron, do you think that this competition will help with model architecture?
\nSome days ago, I was looking at some past competitions, but they were not very useful. I will check the competitions you mentioned.
Try and find out :)
\nOf course, no competition is the same. Ribonanza is a bit tricky since it had also edge data, ASLFR was encoder-decoder kind of competiton and iceCube data was the kind of data that encourages GNNs solutions.
\nBut the similarities exist and experience and ideas from those past competitions helps.
thank you for your kindness.
", "votes": 0, "replies": [] }, { "id": 2891891, "author_name": "Sijun Xu", "author_url": "", "post_date": "2024-06-27T01:39:22.557000", "content": "Your suggestion is very helpful! I have tried model posted in Isolated Sign Language Recognition, which is dealing with 1d data, but my local cv cannot achieve good results. How much domain knowledge helps to improve model architecture?
", "votes": 0, "replies": [ { "id": 2892722, "author_name": "greySnow", "author_url": "", "post_date": "2024-06-27T11:39:36.407000", "content": "If 'domain knowledge' = atmospheric physics, None. If 'domain knowledge' = deep learning tricks, architectures, finetuning and best practices- a lot. Don't expect to copy paste, plug in and get good result out of the box…
", "votes": 1, "replies": [] }, { "id": 2893673, "author_name": "Sijun Xu", "author_url": "", "post_date": "2024-06-28T01:53:05.670000", "content": "Thanks for sharing! I just start from copy and paste and learned a lot from your codes in previous competitions🤣. By modifing model architectures I can see improvements step by step.
First, try to create a model by yourself without copying someone else's notebook. This is the best way to learn.
\nAfter you get stuck, I usually go to the code section and pick the best code that is clear to read and understand. In this competition, I think there are two (maybe more) awesome notebooks: Giba's and Amadeo's.
\nThese two are really good for learning and trying to pursue a medal.
After these notebooks are completely understood, go to the internet and the Discussion section to find new ideas.
\nTo understand both notebooks and if you have an unanswered question, ask ChatGPT.
", "votes": 0, "replies": [ { "id": 2890171, "author_name": "syuz00", "author_url": "", "post_date": "2024-06-26T01:28:23.857000", "content": "Thank you very much
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2889472": "Similar competitions in last year: Ribonanza, American Sign Language Fingerspelling Recognition, Isolated Sign Language Recognition, and iceCube. Each competition is unique, but all of those competitions deal with 1D data (i.e., things you can put into LSTM/GRU/Transformers) and massive data sets. Not coincidentally, I have a silver and two golds in 3 of these competitions, and this past knowledge is why I'm doing well in this competition, too.", "2889459": "Nice to meet you.\nI'm a beginner.I have created simple models for several competitions.\nI don't know how to proceed with this kind of competition, so I would like to ask you something.\nNot limited to this competition, but how do you achieve results? I feel that there are many people who are able to produce analytical results even without specialized knowledge. Since I can't really participate in the discussions, I would like to ask you all, how do you achieve results? I would like to know how to proceed and how to think.\nI tried to refer to similar competitions in the past, but I couldn't find anything that looked like it.", "2889476": "First, try to create a model by yourself without copying someone else's notebook. This is the best way to learn.\n\nAfter you get stuck, I usually go to the code section and pick the best code that is clear to read and understand. In this competition, I think there are two (maybe more) awesome notebooks: [Giba's](https://www.kaggle.com/code/titericz/giba-baseline-xgboost) and [Amadeo's](https://www.kaggle.com/code/abiolatti/keras-baseline-seq2seq).\nThese two are really good for learning and trying to pursue a medal.\n\nAfter these notebooks are completely understood, go to the internet and the Discussion section to find new ideas.\n\nTo understand both notebooks and if you have an unanswered question, ask ChatGPT.\n\n" } }