Hi all, as this competition comes to a close I wanted to have a retrospective of what I have learned. For me this helps to forget about the stress of placement and to remind myself of why I came to Kaggle in the first place, to learn! Please do share some of what you learned.
\nMore signal processing : Converting the EEG data to images and spectrograms forced me to focus a bit more on how to process signal in a way that best benefits the model the most. I won't go into too much detail on this before competition end, but stay tuned.
More pytorch/tensorflow : I am definitely not an expert in pytorch yet, and I have not done much with regard to images in tensorflow so this was a great learning experience for me. I also have the help of fantastic teammates to help me throughout the way.
More about Albumentations : There was a ton of support from the Albumentations competition in this competition, particularly interesting masking methods were added recently which was very helpful. This is a lot more useful than the image processing I was doing before!
More on ensemble strategy : A lot of different strategies worked for this competition so ensemble was pretty important. I will save tricks for later, but I am very excited to see what everyone else did here. I have not had many use cases where so many different ideas work well. I saw catboosts, 1d, 2d, 3d, models on spectrograms only, models on eeg only, and so much more!
Good luck everyone! I hope you all feel you learned from this competition! Remember the learning is more important than the placement!
", "messages": [ { "id": "2737790", "postDate": "04/06/2024 00:07:20", "content": "Hi all, as this competition comes to a close I wanted to have a retrospective of what I have learned. For me this helps to forget about the stress of placement and to remind myself of why I came to Kaggle in the first place, to learn! Please do share some of what you learned.
\nMore signal processing : Converting the EEG data to images and spectrograms forced me to focus a bit more on how to process signal in a way that best benefits the model the most. I won't go into too much detail on this before competition end, but stay tuned.
More pytorch/tensorflow : I am definitely not an expert in pytorch yet, and I have not done much with regard to images in tensorflow so this was a great learning experience for me. I also have the help of fantastic teammates to help me throughout the way.
More about Albumentations : There was a ton of support from the Albumentations competition in this competition, particularly interesting masking methods were added recently which was very helpful. This is a lot more useful than the image processing I was doing before!
More on ensemble strategy : A lot of different strategies worked for this competition so ensemble was pretty important. I will save tricks for later, but I am very excited to see what everyone else did here. I have not had many use cases where so many different ideas work well. I saw catboosts, 1d, 2d, 3d, models on spectrograms only, models on eeg only, and so much more!
Good luck everyone! I hope you all feel you learned from this competition! Remember the learning is more important than the placement!
", "rawMarkdown": "Hi all, as this competition comes to a close I wanted to have a retrospective of what I have learned. For me this helps to forget about the stress of placement and to remind myself of why I came to Kaggle in the first place, to learn! Please do share some of what you learned.\n\n1. **More signal processing** : Converting the EEG data to images and spectrograms forced me to focus a bit more on how to process signal in a way that best benefits the model the most. I won't go into too much detail on this before competition end, but stay tuned. \n\n2. **More pytorch/tensorflow** : I am definitely not an expert in pytorch yet, and I have not done much with regard to images in tensorflow so this was a great learning experience for me. I also have the help of fantastic teammates to help me throughout the way.\n\n3. **More about Albumentations** : There was a ton of support from the Albumentations competition in this competition, particularly interesting masking methods were added recently which was very helpful. This is a lot more useful than the image processing I was doing before!\n\n4. **More on ensemble strategy** : A lot of different strategies worked for this competition so ensemble was pretty important. I will save tricks for later, but I am very excited to see what everyone else did here. I have not had many use cases where so many different ideas work well. I saw catboosts, 1d, 2d, 3d, models on spectrograms only, models on eeg only, and so much more! \n\nGood luck everyone! I hope you all feel you learned from this competition! Remember the learning is more important than the placement!", "votes": null }, { "id": "2737845", "postDate": "04/06/2024 01:24:13", "content": "Hello !
\nThanks for this remindš«”!
\nAnother thing is the Learning rate scheduler, is also really important for this competition.
\nThis kind of strategy is extremely useful to make the model converge faster as well as betteršāāļø.
\n Thanks again for Chris had posted about this wonderful discussion.
\nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/488083?rvi=1
Yes! I'm really agreed on that Remember the learning is more important than the placement.
\nAnd then I've also learned a lot of useful skills(especially from chrisš)
Strongly agree!
\nThis competition has greatly improved my deep learning skills.
", "rawMarkdown": "Strongly agree!\n\nThis competition has greatly improved my deep learning skills.", "votes": null }, { "id": "2738406", "postDate": "04/06/2024 10:24:37", "content": "Great content! Your content motivated me to try this competition! š
", "rawMarkdown": "Great content! Your content motivated me to try this competition! š", "votes": null }, { "id": "2738542", "postDate": "04/06/2024 12:02:56", "content": "This competition is so much fun, in a short time I learned many different techniques that work in practice. Thank you to everyone who shared a bit of their knowledge.
", "rawMarkdown": "This competition is so much fun, in a short time I learned many different techniques that work in practice. Thank you to everyone who shared a bit of their knowledge.", "votes": null }, { "id": "2738634", "postDate": "04/06/2024 13:07:37", "content": "Glad to hear it! I hope you had fun!
", "rawMarkdown": "Glad to hear it! I hope you had fun!", "votes": null }, { "id": "2738846", "postDate": "04/06/2024 16:48:14", "content": "This competition was a great excuse for me to learn about signals, filters and spectrograms.
", "rawMarkdown": "This competition was a great excuse for me to learn about signals, filters and spectrograms.", "votes": null }, { "id": "2739364", "postDate": "04/07/2024 01:38:18", "content": "great learning at least for me certainly on signal processing and all types of albumations, creating funky models/architectures modeled around eeg/specs and the human brain. Some dunno how good they worked but was really satisfying to write them. Did write some nice notes on some models at least the reason on that going down the eeg rabbit hole :) will publish at the end of comp if they make any sense š
", "rawMarkdown": "great learning at least for me certainly on signal processing and all types of albumations, creating funky models/architectures modeled around eeg/specs and the human brain. Some dunno how good they worked but was really satisfying to write them. Did write some nice notes on some models at least the reason on that going down the eeg rabbit hole :) will publish at the end of comp if they make any sense š", "votes": null }, { "id": "2740149", "postDate": "04/07/2024 16:13:40", "content": "Thanks to this competition, the communication atmosphere and the competition itself are both exciting. Special thanks to Chris, who has selflessly helped many people like me who donāt know much about this competition. I remember that LB 0.6 was enough to excite everyone when the competition started, but now there have been two LB 0.21.
\nAs a beginner, I really learned a lot of things: processing of EEG raw signals (ResNet), image analysis using spectrogram (EfficientNet), also the processing pipeline: like ensemble of different models, multi-stage, and tuning train schedule (very impressed by Chrisās discussion here) etc.
\nActually, this is my first time participating in a Kaggle competition. I always feel there are too many things to learn. It seems like I do the Knowledge Distillation in my brainš¤£. Cause the limited time, I even had to trade off what I should learn first. I think it might be a nice feeling that I'll remember for a long time. I am also very grateful to my teammates, who gave me a lot of help and support even though I was a beginner.
\n\n\n\"Remember the learning is more important than the placement!\"
\n
As above, this experience means a lot to me, and I also hope to see you all in other competitions in the future. Good luck and let's progress together~
", "rawMarkdown": "Thanks to this competition, the communication atmosphere and the competition itself are both exciting. Special thanks to *Chris*, who has selflessly helped many people like me who donāt know much about this competition. I remember that LB 0.6 was enough to excite everyone when the competition started, but now there have been two LB 0.21.\n\nAs a beginner, I really learned a lot of things: processing of EEG raw signals (ResNet), image analysis using spectrogram (EfficientNet), also the processing pipeline: like ensemble of different models, multi-stage, and tuning train schedule (very impressed by Chrisās discussion [here](https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/488083)) etc.\n\nActually, this is my first time participating in a Kaggle competition. I always feel there are too many things to learn. It seems like I do the Knowledge Distillation in my brainš¤£. Cause the limited time, I even had to trade off what I should learn first. I think it might be a nice feeling that I'll remember for a long time. I am also very grateful to my teammates, who gave me a lot of help and support even though I was a beginner. \n\n>\"Remember the learning is more important than the placement!\"\n\nAs above, this experience means a lot to me, and I also hope to see you all in other competitions in the future. Good luck and let's progress together~", "votes": null }, { "id": "2740904", "postDate": "04/08/2024 02:00:16", "content": "The difficulty in this competition is that how to deal with edge cases for me, and I learn a lot about the preprocessing of spectrograms and signals, though some of them are not successful in this competition. To find the best pipeline, I try many pretrained 2d backbone and utilize many methods to change and fix their own structures. It's a great competition for me to research on the difference between models in timm. What a pity is that I don't have enough time to sink my teeth into it all the time, and I bet it's more challenging than my fulltime work.
\nThanks to my teammates.
\nBTW, Chris and Sean, and all of the competitors, you are my heroes.
", "rawMarkdown": "The difficulty in this competition is that how to deal with edge cases for me, and I learn a lot about the preprocessing of spectrograms and signals, though some of them are not successful in this competition. To find the best pipeline, I try many pretrained 2d backbone and utilize many methods to change and fix their own structures. It's a great competition for me to research on the difference between models in timm. What a pity is that I don't have enough time to sink my teeth into it all the time, and I bet it's more challenging than my fulltime work.\n\nThanks to my teammates.\n\nBTW, Chris and Sean, and all of the competitors, you are my heroes.", "votes": null }, { "id": "2741094", "postDate": "04/08/2024 06:16:05", "content": "Good luck everyone! Fun competition! I learning a lot about
\nI'm excited to read the winner discussion posts tomorrow!
", "rawMarkdown": "Good luck everyone! Fun competition! I learning a lot about \n* signal EDA\n* preprocessing with filters (notch, low pass, high pass, etc)\n* preprocess with spectrograms (different libraries and techniques etc)\n* multi-stage training (pretrain, psuedo label, knowledge distillation, fine tune, etc)\n* data augmentation\n* multi modal model architectures\n* train schedules\n* ensemble techniques\n* post process\n\nI'm excited to read the winner discussion posts tomorrow!", "votes": null }, { "id": "2741538", "postDate": "04/08/2024 12:32:34", "content": "I think cv choice I made here is my hardest bet I've ever made in any comp before :)
\nLet's see how this goes!
Yes for sure! Haha fingers crossed it is correct!
", "rawMarkdown": "Yes for sure! Haha fingers crossed it is correct!", "votes": null }, { "id": "2741851", "postDate": "04/08/2024 16:14:22", "content": "I really enjoyed this dataset and challengeā¦ I learned a lot about signal processing, mel spectrograms and about measuring brain waves. Thanks @cdeotte @seshurajup @medali1992 and everyone who shared their knowledge about measuring brain waves - this was so helpful - it took me a few weeks to get to grips with. Also learnt a lot about CNN architectures - although much of it didnt help massively.
\nGood luck everybody on the competition close āļø I can't wait to see the approaches people used - and all the obvious techniques that we missed to try :)
Participating in this competition was a rewarding experience. Being part of a community of like-minded individuals with shared values and interests felt truly energizing. It was refreshing to be surrounded by people who possess strong convictions, unwavering integrity, and an insatiable curiosity for exploring the unknown, even when it may seem daunting or uncertain.
\nAs this was my first real competition, I realized that there is a significant difference between theoretical knowledge and practical experience. I was fortunate to interact with knowledgeable individuals who generously shared their insights and expertise with us.
\nThis competition has taught me that while theoretical understanding is essential, hands-on experience is invaluable. It has reinforced the importance of continuously learning and embracing challenges as opportunities for growth, even when faced with unfamiliar or uncertain situations.
", "rawMarkdown": "Participating in this competition was a rewarding experience. Being part of a community of like-minded individuals with shared values and interests felt truly energizing. It was refreshing to be surrounded by people who possess strong convictions, unwavering integrity, and an insatiable curiosity for exploring the unknown, even when it may seem daunting or uncertain.\n\nAs this was my first real competition, I realized that there is a significant difference between theoretical knowledge and practical experience. I was fortunate to interact with knowledgeable individuals who generously shared their insights and expertise with us.\n\nThis competition has taught me that while theoretical understanding is essential, hands-on experience is invaluable. It has reinforced the importance of continuously learning and embracing challenges as opportunities for growth, even when faced with unfamiliar or uncertain situations.", "votes": null }, { "id": "2741898", "postDate": "04/08/2024 16:32:23", "content": "@darraghdog Thanks, its all because of @cdeotte motivated me to explore more.
", "rawMarkdown": "darraghdog Thanks, its all because of @cdeotte motivated me to explore more.", "votes": null }, { "id": "2742490", "postDate": "04/09/2024 00:05:56", "content": "The most important thing for me is to NEVER overfit on public LBā¦our teamās final result on private LB is very close to the local CV score.
", "rawMarkdown": "The most important thing for me is to NEVER overfit on public LBā¦our teamās final result on private LB is very close to the local CV score.", "votes": null }, { "id": "2742498", "postDate": "04/09/2024 00:16:09", "content": "Correct, but this is only the case if you ignore the low votes side of the training dataset
", "rawMarkdown": "Correct, but this is only the case if you ignore the low votes side of the training dataset", "votes": null } ], "comments": [ { "id": 2737845, "author_name": "lmhongkhnh", "author_url": "", "post_date": "04/06/2024 01:24:13", "content": "Hello !
\nThanks for this remindš«”!
\nAnother thing is the Learning rate scheduler, is also really important for this competition.
\nThis kind of strategy is extremely useful to make the model converge faster as well as betteršāāļø.
\n Thanks again for Chris had posted about this wonderful discussion.
\nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/488083?rvi=1
Yes! I'm really agreed on that Remember the learning is more important than the placement.
\nAnd then I've also learned a lot of useful skills(especially from chrisš)
Strongly agree!
\nThis competition has greatly improved my deep learning skills.
", "votes": null, "replies": [] }, { "id": 2738406, "author_name": "aaronfriasr", "author_url": "", "post_date": "04/06/2024 10:24:37", "content": "Great content! Your content motivated me to try this competition! š
", "votes": null, "replies": [ { "id": 2738634, "author_name": "cody11null", "author_url": "", "post_date": "04/06/2024 13:07:37", "content": "Glad to hear it! I hope you had fun!
", "votes": null, "replies": [] } ] }, { "id": 2738542, "author_name": "robsonsan", "author_url": "", "post_date": "04/06/2024 12:02:56", "content": "This competition is so much fun, in a short time I learned many different techniques that work in practice. Thank you to everyone who shared a bit of their knowledge.
", "votes": null, "replies": [] }, { "id": 2738846, "author_name": "gunesevitan", "author_url": "", "post_date": "04/06/2024 16:48:14", "content": "This competition was a great excuse for me to learn about signals, filters and spectrograms.
", "votes": null, "replies": [] }, { "id": 2739364, "author_name": "gauravbrills", "author_url": "", "post_date": "04/07/2024 01:38:18", "content": "great learning at least for me certainly on signal processing and all types of albumations, creating funky models/architectures modeled around eeg/specs and the human brain. Some dunno how good they worked but was really satisfying to write them. Did write some nice notes on some models at least the reason on that going down the eeg rabbit hole :) will publish at the end of comp if they make any sense š
", "votes": null, "replies": [] }, { "id": 2740149, "author_name": "fangzitao", "author_url": "", "post_date": "04/07/2024 16:13:40", "content": "Thanks to this competition, the communication atmosphere and the competition itself are both exciting. Special thanks to Chris, who has selflessly helped many people like me who donāt know much about this competition. I remember that LB 0.6 was enough to excite everyone when the competition started, but now there have been two LB 0.21.
\nAs a beginner, I really learned a lot of things: processing of EEG raw signals (ResNet), image analysis using spectrogram (EfficientNet), also the processing pipeline: like ensemble of different models, multi-stage, and tuning train schedule (very impressed by Chrisās discussion here) etc.
\nActually, this is my first time participating in a Kaggle competition. I always feel there are too many things to learn. It seems like I do the Knowledge Distillation in my brainš¤£. Cause the limited time, I even had to trade off what I should learn first. I think it might be a nice feeling that I'll remember for a long time. I am also very grateful to my teammates, who gave me a lot of help and support even though I was a beginner.
\n\n\n\"Remember the learning is more important than the placement!\"
\n
As above, this experience means a lot to me, and I also hope to see you all in other competitions in the future. Good luck and let's progress together~
", "votes": null, "replies": [] }, { "id": 2740904, "author_name": "sweetyheehee", "author_url": "", "post_date": "04/08/2024 02:00:16", "content": "The difficulty in this competition is that how to deal with edge cases for me, and I learn a lot about the preprocessing of spectrograms and signals, though some of them are not successful in this competition. To find the best pipeline, I try many pretrained 2d backbone and utilize many methods to change and fix their own structures. It's a great competition for me to research on the difference between models in timm. What a pity is that I don't have enough time to sink my teeth into it all the time, and I bet it's more challenging than my fulltime work.
\nThanks to my teammates.
\nBTW, Chris and Sean, and all of the competitors, you are my heroes.
", "votes": null, "replies": [] }, { "id": 2741094, "author_name": "cdeotte", "author_url": "", "post_date": "04/08/2024 06:16:05", "content": "Good luck everyone! Fun competition! I learning a lot about
\nI'm excited to read the winner discussion posts tomorrow!
", "votes": null, "replies": [] }, { "id": 2741538, "author_name": "mohammad2012191", "author_url": "", "post_date": "04/08/2024 12:32:34", "content": "I think cv choice I made here is my hardest bet I've ever made in any comp before :)
\nLet's see how this goes!
Yes for sure! Haha fingers crossed it is correct!
", "votes": null, "replies": [] } ] }, { "id": 2741851, "author_name": "darraghdog", "author_url": "", "post_date": "04/08/2024 16:14:22", "content": "I really enjoyed this dataset and challengeā¦ I learned a lot about signal processing, mel spectrograms and about measuring brain waves. Thanks @cdeotte @seshurajup @medali1992 and everyone who shared their knowledge about measuring brain waves - this was so helpful - it took me a few weeks to get to grips with. Also learnt a lot about CNN architectures - although much of it didnt help massively.
\nGood luck everybody on the competition close āļø I can't wait to see the approaches people used - and all the obvious techniques that we missed to try :)
@darraghdog Thanks, its all because of @cdeotte motivated me to explore more.
", "votes": null, "replies": [] } ] }, { "id": 2741893, "author_name": "nartaa", "author_url": "", "post_date": "04/08/2024 16:29:43", "content": "Participating in this competition was a rewarding experience. Being part of a community of like-minded individuals with shared values and interests felt truly energizing. It was refreshing to be surrounded by people who possess strong convictions, unwavering integrity, and an insatiable curiosity for exploring the unknown, even when it may seem daunting or uncertain.
\nAs this was my first real competition, I realized that there is a significant difference between theoretical knowledge and practical experience. I was fortunate to interact with knowledgeable individuals who generously shared their insights and expertise with us.
\nThis competition has taught me that while theoretical understanding is essential, hands-on experience is invaluable. It has reinforced the importance of continuously learning and embracing challenges as opportunities for growth, even when faced with unfamiliar or uncertain situations.
", "votes": null, "replies": [] }, { "id": 2742490, "author_name": "phoenix1203", "author_url": "", "post_date": "04/09/2024 00:05:56", "content": "The most important thing for me is to NEVER overfit on public LBā¦our teamās final result on private LB is very close to the local CV score.
", "votes": null, "replies": [ { "id": 2742498, "author_name": "cody11null", "author_url": "", "post_date": "04/09/2024 00:16:09", "content": "Correct, but this is only the case if you ignore the low votes side of the training dataset
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2737790": "Hi all, as this competition comes to a close I wanted to have a retrospective of what I have learned. For me this helps to forget about the stress of placement and to remind myself of why I came to Kaggle in the first place, to learn! Please do share some of what you learned.\n\n1. **More signal processing** : Converting the EEG data to images and spectrograms forced me to focus a bit more on how to process signal in a way that best benefits the model the most. I won't go into too much detail on this before competition end, but stay tuned. \n\n2. **More pytorch/tensorflow** : I am definitely not an expert in pytorch yet, and I have not done much with regard to images in tensorflow so this was a great learning experience for me. I also have the help of fantastic teammates to help me throughout the way.\n\n3. **More about Albumentations** : There was a ton of support from the Albumentations competition in this competition, particularly interesting masking methods were added recently which was very helpful. This is a lot more useful than the image processing I was doing before!\n\n4. **More on ensemble strategy** : A lot of different strategies worked for this competition so ensemble was pretty important. I will save tricks for later, but I am very excited to see what everyone else did here. I have not had many use cases where so many different ideas work well. I saw catboosts, 1d, 2d, 3d, models on spectrograms only, models on eeg only, and so much more! \n\nGood luck everyone! I hope you all feel you learned from this competition! Remember the learning is more important than the placement!", "2737845": "Hello !\nThanks for this remindš«”!\nAnother thing is the Learning rate scheduler, is also really important for this competition.\nThis kind of strategy is extremely useful to make the model converge faster as well as betteršāāļø.\n Thanks again for Chris had posted about this wonderful discussion. \nhttps://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/488083?rvi=1", "2737953": "Yes! I'm really agreed on that Remember the learning is more important than the placement.\nAnd then I've also learned a lot of useful skills(especially from chrisš)", "2738333": "Strongly agree!\n\nThis competition has greatly improved my deep learning skills.", "2738406": "Great content! Your content motivated me to try this competition! š", "2738542": "This competition is so much fun, in a short time I learned many different techniques that work in practice. Thank you to everyone who shared a bit of their knowledge.", "2738634": "Glad to hear it! I hope you had fun!", "2738846": "This competition was a great excuse for me to learn about signals, filters and spectrograms.", "2739364": "great learning at least for me certainly on signal processing and all types of albumations, creating funky models/architectures modeled around eeg/specs and the human brain. Some dunno how good they worked but was really satisfying to write them. Did write some nice notes on some models at least the reason on that going down the eeg rabbit hole :) will publish at the end of comp if they make any sense š", "2740149": "Thanks to this competition, the communication atmosphere and the competition itself are both exciting. Special thanks to *Chris*, who has selflessly helped many people like me who donāt know much about this competition. I remember that LB 0.6 was enough to excite everyone when the competition started, but now there have been two LB 0.21.\n\nAs a beginner, I really learned a lot of things: processing of EEG raw signals (ResNet), image analysis using spectrogram (EfficientNet), also the processing pipeline: like ensemble of different models, multi-stage, and tuning train schedule (very impressed by Chrisās discussion [here](https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/discussion/488083)) etc.\n\nActually, this is my first time participating in a Kaggle competition. I always feel there are too many things to learn. It seems like I do the Knowledge Distillation in my brainš¤£. Cause the limited time, I even had to trade off what I should learn first. I think it might be a nice feeling that I'll remember for a long time. I am also very grateful to my teammates, who gave me a lot of help and support even though I was a beginner. \n\n>\"Remember the learning is more important than the placement!\"\n\nAs above, this experience means a lot to me, and I also hope to see you all in other competitions in the future. Good luck and let's progress together~", "2740904": "The difficulty in this competition is that how to deal with edge cases for me, and I learn a lot about the preprocessing of spectrograms and signals, though some of them are not successful in this competition. To find the best pipeline, I try many pretrained 2d backbone and utilize many methods to change and fix their own structures. It's a great competition for me to research on the difference between models in timm. What a pity is that I don't have enough time to sink my teeth into it all the time, and I bet it's more challenging than my fulltime work.\n\nThanks to my teammates.\n\nBTW, Chris and Sean, and all of the competitors, you are my heroes.", "2741094": "Good luck everyone! Fun competition! I learning a lot about \n* signal EDA\n* preprocessing with filters (notch, low pass, high pass, etc)\n* preprocess with spectrograms (different libraries and techniques etc)\n* multi-stage training (pretrain, psuedo label, knowledge distillation, fine tune, etc)\n* data augmentation\n* multi modal model architectures\n* train schedules\n* ensemble techniques\n* post process\n\nI'm excited to read the winner discussion posts tomorrow!", "2741538": "I think cv choice I made here is my hardest bet I've ever made in any comp before :) \nLet's see how this goes!", "2741564": "Yes for sure! Haha fingers crossed it is correct!", "2741851": "I really enjoyed this dataset and challenge... I learned a lot about signal processing, mel spectrograms and about measuring brain waves. Thanks @cdeotte @seshurajup @medali1992 and everyone who shared their knowledge about measuring brain waves - this was so helpful - it took me a few weeks to get to grips with. Also learnt a lot about CNN architectures - although much of it didnt help massively. \nGood luck everybody on the competition close āļø I can't wait to see the approaches people used - and all the obvious techniques that we missed to try :)", "2741893": "Participating in this competition was a rewarding experience. Being part of a community of like-minded individuals with shared values and interests felt truly energizing. It was refreshing to be surrounded by people who possess strong convictions, unwavering integrity, and an insatiable curiosity for exploring the unknown, even when it may seem daunting or uncertain.\n\nAs this was my first real competition, I realized that there is a significant difference between theoretical knowledge and practical experience. I was fortunate to interact with knowledgeable individuals who generously shared their insights and expertise with us.\n\nThis competition has taught me that while theoretical understanding is essential, hands-on experience is invaluable. It has reinforced the importance of continuously learning and embracing challenges as opportunities for growth, even when faced with unfamiliar or uncertain situations.", "2741898": "darraghdog Thanks, its all because of @cdeotte motivated me to explore more.", "2742490": "The most important thing for me is to NEVER overfit on public LBā¦our teamās final result on private LB is very close to the local CV score.", "2742498": "Correct, but this is only the case if you ignore the low votes side of the training dataset" }, "source": "meta" }