{ "id": 127976, "title": "Worth Seeing Posts and Notebooks", "url": "/competitions/bengaliai-cv19/discussion/127976", "author_name": "Qishen Ha", "post_date": "2020-01-28T07:01:59.072000", "votes": 204, "comment_count": 28, "views": 0, "content": "

Hi,

\n\n

I'm gathering ideas from discussion and notebook, making this summary.

\n\n

If any helpful post is missing please leave a message to me.

\n\n

To be updated...

\n\n
\n\n

Fix Submission Bugs (Most important part come first)

\n\n\n\n

Many people suffered from submission errors when they tried to submit their kernel to the LB. If you are one of them, you will find these posts help a lot.

\n\n

Noise in Dataset

\n\n\n\n

They showed that there are some samples with wrong labels.

\n\n

Stuff said they are dealing with this.

\n\n

\"\"

\n\n

Ideas for Improving Score

\n\n\n\n

Visualization

\n\n", "messages": [ { "id": 730955, "postDate": "2020-01-28T07:01:59.073Z", "content": "

Hi,

\n\n

I'm gathering ideas from discussion and notebook, making this summary.

\n\n

If any helpful post is missing please leave a message to me.

\n\n

To be updated...

\n\n
\n\n

Fix Submission Bugs (Most important part come first)

\n\n\n\n

Many people suffered from submission errors when they tried to submit their kernel to the LB. If you are one of them, you will find these posts help a lot.

\n\n

Noise in Dataset

\n\n\n\n

They showed that there are some samples with wrong labels.

\n\n

Stuff said they are dealing with this.

\n\n

\"\"

\n\n

Ideas for Improving Score

\n\n\n\n

Visualization

\n\n", "rawMarkdown": "Hi,\n\nI'm gathering ideas from discussion and notebook, making this summary.\n\nIf any helpful post is missing please leave a message to me.\n\nTo be updated...\n\n---\n\n## Fix Submission Bugs (Most important part come first)\n\n\n* **Optimization tips, eliminate Kaggle failures**\nby @pestipeti \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126054\n\n* **How to resolve \"Submission Error\" in my case**\nby @inoueu1 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123632\n\n* **Saving Memory few tips**\nby @amit9484 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/124221\n\nMany people suffered from submission errors when they tried to submit their kernel to the LB. If you are one of them, you will find these posts help a lot.\n\n## Noise in Dataset\n\n* **2 graphemes with the same combination of components?**\nby @mnpinto \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123859\n\n* **Wrong Train Data Labels -Organizers plz clarify about Private Set**\nby @phoenix9032 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126833\n\nThey showed that there are some samples with wrong labels.\n\nStuff said they are dealing with this.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2Fc83c95c6cf6e192db81466ba8762a4b9%2F1580192629595.jpg?generation=1580192658713113&alt=media)\n\n\n## Ideas for Improving Score\n\n* **Best single model**\nby @drhabib \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123198\nDespite the title, many people shared ideas and experiment setting in this post. \n\n* **[placeholder] lb0.963+ pytorch starter kit**\nby @hengck23 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123757\nFrog brother's starter kit and bunch of Frog brother's ideas.\n\n* **mixup/cutmix is all you need**\nby @machinelp \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126504\nAn implementation of mixup & cutmix. I find it very useful!\n\n* **why cutout/mixup works**\nby @hengck23 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/127902\nAn explanation of why cutout/mixup works, and gave an idea of making dataset larger.\n\n* **GridMask implementation**\nby @haqishen \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128161\nIt works. I'm sure about it.\n\n* **An Implementation of Augmix Based on albumentations**\nby @haqishen\nhttps://www.kaggle.com/haqishen/augmix-based-on-albumentations\nGoogle's work. I'm doing experiment on this recently.\n\n* **Morphological transformations as image augmentation**\nby @ren4yu \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128198\nSome special augmentation methods\n\n* **Generating More data**\nby @drhabib \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128059\nAnother way to generate data.\n\n* **Image preprocessing (128x128)**\nby @iafoss \nhttps://www.kaggle.com/iafoss/image-preprocessing-128x128\nCrop out black background.\n\n* **iterative stratification**\nby @yiheng \nhttps://www.kaggle.com/yiheng/iterative-stratification\nHow to split training set.\n\n* **Solving grapheme_root is the key to 0.98?**\nby @bibek777 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/124904\nSome ideas were shared in this post.\n\n* **Approach for 0.97 and What's next?**\nby @ildoonet \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/125828\nSome ideas for going to LB 0.97\n\n* **Tricks for Image Classification......**\nby @machinelp \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128914\nMany ideas collected by @machinelp \n\n* **Approach for 0.97 with 64x64x1 input**\nby @shujun717 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128368\nLB 0.9718 reached by 64x64x1 input??? That's fantastic!\n\n* **Efficientnet - Trials and Analysis**\nby @dhakshiin1601 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128911\nEfficientnet and GeM.\n\n* **Flipping properties of Bengali characters?**\nby @roguekk007 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126761\n\n* **Papers worth reading?**\nby @bibek777 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/127719\nPapers that look worth reading.\n\n* **Bengali Data Sets & Publications!**\nby @ipythonx \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/122604\nPapers that in Bengali x CV field.\n\n* **[placeholder] autoML results**\nby @hengck23 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126337\nExperiment result of AutoML.\n\n* **Teacher-Student**\nby @mightyrains \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/125046\nKnowledge distillation.\n\n* **Label Smoothing and TTA**\nby @karan07 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/125221\n\n* **Shake-Shake**\nby @phoenix9032 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/124249\n\n* **Some experiments with CNN tails**\nby @drhabib \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123432\n\n\n\n## Visualization\n\n* **Bengali - Quick EDA**\nby @pestipeti \nhttps://www.kaggle.com/pestipeti/bengali-quick-eda\n\n* **Bengali.AI Handwritten Grapheme - Getting Started**\nby @gpreda \nhttps://www.kaggle.com/gpreda/bengali-ai-handwritten-grapheme-getting-started\n\n* **Visualize your results**\nby @pestipeti \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/127149\n\n\n", "votes": 202 }, { "id": 737985, "postDate": "2020-02-06T02:15:54.220Z", "content": "

░░░░░▄▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n▄███▀░◐░░░▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▐░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐▄▄░░░░░░░░░░I need to make one comment :)░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░▄▀▒▒▀▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▒▒▒▒▒▒▒▒▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▀▄░░░░▀▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░▀▄▄▄▄▄█▄▄▄▄▄▄▄▄▄▄▄▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░▄▄▌▌▄▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

", "rawMarkdown": "░░░░░▄▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n▄███▀░◐░░░▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▐░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐▄▄░░░░░░░░░░I need to make one comment :)░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░▄▀▒▒▀▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▒▒▒▒▒▒▒▒▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▀▄░░░░▀▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░▀▄▄▄▄▄█▄▄▄▄▄▄▄▄▄▄▄▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░▄▄▌▌▄▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░", "votes": 8 }, { "id": 731030, "postDate": "2020-01-28T09:19:37.537Z", "content": "

👍

", "rawMarkdown": "👍 ", "votes": 3 }, { "id": 764171, "postDate": "2020-03-05T07:43:38.943Z", "content": "

Cool Notebook!

", "rawMarkdown": "Cool Notebook!", "votes": 1 }, { "id": 757482, "postDate": "2020-02-26T20:13:44.020Z", "content": "

Thanks for the collection, it helped me a lot.

", "rawMarkdown": "Thanks for the collection, it helped me a lot.", "votes": 1 }, { "id": 752217, "postDate": "2020-02-20T19:50:00.157Z", "content": "

nice

", "rawMarkdown": "nice", "votes": 1 }, { "id": 744750, "postDate": "2020-02-13T05:18:16.380Z", "content": "

This is great !

", "rawMarkdown": "This is great !", "votes": 1 }, { "id": 739654, "postDate": "2020-02-08T05:51:34.417Z", "content": "

Thanks a lot of this summarising everything, it's very informative.

", "rawMarkdown": "Thanks a lot of this summarising everything, it's very informative.", "votes": 1 }, { "id": 739633, "postDate": "2020-02-08T04:43:22.207Z", "content": "

Thanks for informative summary, it's very helpful

", "rawMarkdown": "Thanks for informative summary, it's very helpful", "votes": 1 }, { "id": 737057, "postDate": "2020-02-04T21:30:10.947Z", "content": "

Thanks a lot @haqishen , I will refer to this to be able to get inspired and improve my performance

", "rawMarkdown": "Thanks a lot @haqishen , I will refer to this to be able to get inspired and improve my performance", "votes": 1 }, { "id": 734655, "postDate": "2020-02-01T19:38:55.390Z", "content": "

Awesome thread! And the \"Frog brother\" nickname is great. ;)

", "rawMarkdown": "Awesome thread! And the \"Frog brother\" nickname is great. ;)", "votes": 1 }, { "id": 733638, "postDate": "2020-01-31T11:28:22.400Z", "content": "

wow. its fantastiic. one is similar to my idea. but i didnt make it. its very useful. thank you.

", "rawMarkdown": "wow. its fantastiic. one is similar to my idea. but i didnt make it. its very useful. thank you. ", "votes": 1 }, { "id": 731459, "postDate": "2020-01-28T17:02:33.740Z", "content": "

Ahh Thanks for compiling this =) Helps for new people and for old folks like me =)

", "rawMarkdown": "Ahh Thanks for compiling this =) Helps for new people and for old folks like me =) ", "votes": 1 }, { "id": 731361, "postDate": "2020-01-28T14:44:37.230Z", "content": "

Love all the sharing in the comp, very helpful for beginners!

", "rawMarkdown": "Love all the sharing in the comp, very helpful for beginners!", "votes": 1 }, { "id": 731223, "postDate": "2020-01-28T12:54:19.580Z", "content": "

@haqishen Thanks a lot for gathering all the important pieces in one place. It would be nice if you could add your own suggestions for the participants as well! Of course, I am not talking about 0.9923, but promising directions in an abstract view at least!

", "rawMarkdown": "@haqishen Thanks a lot for gathering all the important pieces in one place. It would be nice if you could add your own suggestions for the participants as well! Of course, I am not talking about 0.9923, but promising directions in an abstract view at least!", "votes": 1, "replies": [ { "id": 731262, "postDate": "2020-01-28T13:14:57.550Z", "content": "

Well... Just one more tip, the most effective idea I've tried within those posts is this one:\nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123198#718926\n4~5 teams broke 0.980 within a week after this post released.

", "rawMarkdown": "Well... Just one more tip, the most effective idea I've tried within those posts is this one:\nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123198#718926\n4~5 teams broke 0.980 within a week after this post released.", "votes": 9 } ] }, { "id": 730964, "postDate": "2020-01-28T07:09:33.780Z", "content": "

Wow, This is extremely helpful. Thanks @haqishen 💚

", "rawMarkdown": "Wow, This is extremely helpful. Thanks @haqishen 💚 ", "votes": 1 }, { "id": 730987, "postDate": "2020-01-28T07:59:16.277Z", "content": "

Thank you for including my Kernel. And congratulations for your top position.

", "rawMarkdown": "Thank you for including my Kernel. And congratulations for your top position.", "votes": 2, "replies": [ { "id": 730991, "postDate": "2020-01-28T08:03:27.517Z", "content": "

Your kernel is wonderful, thanks for sharing 😃

", "rawMarkdown": "Your kernel is wonderful, thanks for sharing 😃", "votes": 2 } ] }, { "id": 764669, "postDate": "2020-03-05T17:59:54.667Z", "rawMarkdown": "", "isDeleted": true }, { "id": 733524, "postDate": "2020-01-31T09:07:52.333Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 759519, "postDate": "2020-02-29T06:29:55.180Z", "content": "

Thank you for sharing @haqishen .

", "rawMarkdown": "Thank you for sharing @haqishen .", "votes": 1 }, { "id": 749846, "postDate": "2020-02-18T23:54:49.123Z", "content": "

Thanks for shairng

", "rawMarkdown": "Thanks for shairng", "votes": 1 }, { "id": 733212, "postDate": "2020-01-30T20:31:47.043Z", "content": "

Thanks for sharing this resource

", "rawMarkdown": "Thanks for sharing this resource", "votes": 1 }, { "id": 732733, "postDate": "2020-01-30T07:06:13.013Z", "content": "

Thanks for the gathering! Very helpful!

", "rawMarkdown": "Thanks for the gathering! Very helpful!", "votes": 1 }, { "id": 732677, "postDate": "2020-01-30T05:01:53.127Z", "content": "

Thank you for the resource!

", "rawMarkdown": "Thank you for the resource!", "votes": 1 }, { "id": 732303, "postDate": "2020-01-29T16:56:04.303Z", "content": "

Thanks for sharing.

", "rawMarkdown": "Thanks for sharing.", "votes": 1 }, { "id": 732161, "postDate": "2020-01-29T14:43:06.243Z", "content": "

Thanks, it will help me a lot!!! 😁

", "rawMarkdown": "Thanks, it will help me a lot!!! 😁 ", "votes": 1 }, { "id": 732298, "postDate": "2020-01-29T16:44:30.617Z", "content": "

Very helpful! Thanks for sharing!!😄 👍

", "rawMarkdown": "Very helpful! Thanks for sharing!!😄 👍 ", "votes": 2 } ], "comments": [ { "id": 737985, "author_name": "ESH", "author_url": "", "post_date": "2020-02-06T02:15:54.220000", "content": "

░░░░░▄▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n▄███▀░◐░░░▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▐░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐▄▄░░░░░░░░░░I need to make one comment :)░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░▄▀▒▒▀▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▒▒▒▒▒▒▒▒▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▀▄░░░░▀▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░▀▄▄▄▄▄█▄▄▄▄▄▄▄▄▄▄▄▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░▄▄▌▌▄▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

", "votes": 8, "replies": [] }, { "id": 731030, "author_name": "MachineLP", "author_url": "", "post_date": "2020-01-28T09:19:37.537000", "content": "

👍

", "votes": 3, "replies": [] }, { "id": 764171, "author_name": "Sumit Mishra", "author_url": "", "post_date": "2020-03-05T07:43:38.943000", "content": "

Cool Notebook!

", "votes": 1, "replies": [] }, { "id": 757482, "author_name": "gnyana teja", "author_url": "", "post_date": "2020-02-26T20:13:44.020000", "content": "

Thanks for the collection, it helped me a lot.

", "votes": 1, "replies": [] }, { "id": 752217, "author_name": "", "author_url": "", "post_date": "2020-02-20T19:50:00.157000", "content": "

nice

", "votes": 1, "replies": [] }, { "id": 744750, "author_name": "Md Zarif Ul Alam", "author_url": "", "post_date": "2020-02-13T05:18:16.380000", "content": "

This is great !

", "votes": 1, "replies": [] }, { "id": 739654, "author_name": "Naman Jaswani", "author_url": "", "post_date": "2020-02-08T05:51:34.417000", "content": "

Thanks a lot of this summarising everything, it's very informative.

", "votes": 1, "replies": [] }, { "id": 739633, "author_name": "ccchang", "author_url": "", "post_date": "2020-02-08T04:43:22.207000", "content": "

Thanks for informative summary, it's very helpful

", "votes": 1, "replies": [] }, { "id": 737057, "author_name": "Rohit Agarwal", "author_url": "", "post_date": "2020-02-04T21:30:10.947000", "content": "

Thanks a lot @haqishen , I will refer to this to be able to get inspired and improve my performance

", "votes": 1, "replies": [] }, { "id": 734655, "author_name": "Yassine Alouini", "author_url": "", "post_date": "2020-02-01T19:38:55.390000", "content": "

Awesome thread! And the \"Frog brother\" nickname is great. ;)

", "votes": 1, "replies": [] }, { "id": 733638, "author_name": "HELLOWORLD", "author_url": "", "post_date": "2020-01-31T11:28:22.400000", "content": "

wow. its fantastiic. one is similar to my idea. but i didnt make it. its very useful. thank you.

", "votes": 1, "replies": [] }, { "id": 731459, "author_name": "DrHB", "author_url": "", "post_date": "2020-01-28T17:02:33.740000", "content": "

Ahh Thanks for compiling this =) Helps for new people and for old folks like me =)

", "votes": 1, "replies": [] }, { "id": 731361, "author_name": "GreatGameDota", "author_url": "", "post_date": "2020-01-28T14:44:37.230000", "content": "

Love all the sharing in the comp, very helpful for beginners!

", "votes": 1, "replies": [] }, { "id": 731223, "author_name": "Mohammad Azam Khan", "author_url": "", "post_date": "2020-01-28T12:54:19.580000", "content": "

@haqishen Thanks a lot for gathering all the important pieces in one place. It would be nice if you could add your own suggestions for the participants as well! Of course, I am not talking about 0.9923, but promising directions in an abstract view at least!

", "votes": 1, "replies": [ { "id": 731262, "author_name": "Qishen Ha", "author_url": "", "post_date": "2020-01-28T13:14:57.550000", "content": "

Well... Just one more tip, the most effective idea I've tried within those posts is this one:\nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123198#718926\n4~5 teams broke 0.980 within a week after this post released.

", "votes": 9, "replies": [] } ] }, { "id": 730964, "author_name": "Innat", "author_url": "", "post_date": "2020-01-28T07:09:33.780000", "content": "

Wow, This is extremely helpful. Thanks @haqishen 💚

", "votes": 1, "replies": [] }, { "id": 730987, "author_name": "Gabriel Preda", "author_url": "", "post_date": "2020-01-28T07:59:16.277000", "content": "

Thank you for including my Kernel. And congratulations for your top position.

", "votes": 2, "replies": [ { "id": 730991, "author_name": "Qishen Ha", "author_url": "", "post_date": "2020-01-28T08:03:27.517000", "content": "

Your kernel is wonderful, thanks for sharing 😃

", "votes": 2, "replies": [] } ] }, { "id": 764669, "author_name": "", "author_url": "", "post_date": "2020-03-05T17:59:54.667000", "content": "", "votes": 0, "replies": [] }, { "id": 733524, "author_name": "", "author_url": "", "post_date": "2020-01-31T09:07:52.333000", "content": "", "votes": 1, "replies": [] }, { "id": 759519, "author_name": "Joel Hanson", "author_url": "", "post_date": "2020-02-29T06:29:55.180000", "content": "

Thank you for sharing @haqishen .

", "votes": 1, "replies": [] }, { "id": 749846, "author_name": "Kurian Benoy", "author_url": "", "post_date": "2020-02-18T23:54:49.123000", "content": "

Thanks for shairng

", "votes": 1, "replies": [] }, { "id": 733212, "author_name": "Manish Nayak", "author_url": "", "post_date": "2020-01-30T20:31:47.043000", "content": "

Thanks for sharing this resource

", "votes": 1, "replies": [] }, { "id": 732733, "author_name": "Helen", "author_url": "", "post_date": "2020-01-30T07:06:13.013000", "content": "

Thanks for the gathering! Very helpful!

", "votes": 1, "replies": [] }, { "id": 732677, "author_name": "John Lang", "author_url": "", "post_date": "2020-01-30T05:01:53.127000", "content": "

Thank you for the resource!

", "votes": 1, "replies": [] }, { "id": 732303, "author_name": "karan", "author_url": "", "post_date": "2020-01-29T16:56:04.303000", "content": "

Thanks for sharing.

", "votes": 1, "replies": [] }, { "id": 732161, "author_name": "Victor Menuzzo", "author_url": "", "post_date": "2020-01-29T14:43:06.243000", "content": "

Thanks, it will help me a lot!!! 😁

", "votes": 1, "replies": [] }, { "id": 732298, "author_name": "Miyabon", "author_url": "", "post_date": "2020-01-29T16:44:30.617000", "content": "

Very helpful! Thanks for sharing!!😄 👍

", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "730955": "Hi,\n\nI'm gathering ideas from discussion and notebook, making this summary.\n\nIf any helpful post is missing please leave a message to me.\n\nTo be updated...\n\n---\n\n## Fix Submission Bugs (Most important part come first)\n\n\n* **Optimization tips, eliminate Kaggle failures**\nby @pestipeti \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126054\n\n* **How to resolve \"Submission Error\" in my case**\nby @inoueu1 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123632\n\n* **Saving Memory few tips**\nby @amit9484 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/124221\n\nMany people suffered from submission errors when they tried to submit their kernel to the LB. If you are one of them, you will find these posts help a lot.\n\n## Noise in Dataset\n\n* **2 graphemes with the same combination of components?**\nby @mnpinto \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123859\n\n* **Wrong Train Data Labels -Organizers plz clarify about Private Set**\nby @phoenix9032 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126833\n\nThey showed that there are some samples with wrong labels.\n\nStuff said they are dealing with this.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2Fc83c95c6cf6e192db81466ba8762a4b9%2F1580192629595.jpg?generation=1580192658713113&alt=media)\n\n\n## Ideas for Improving Score\n\n* **Best single model**\nby @drhabib \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123198\nDespite the title, many people shared ideas and experiment setting in this post. \n\n* **[placeholder] lb0.963+ pytorch starter kit**\nby @hengck23 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123757\nFrog brother's starter kit and bunch of Frog brother's ideas.\n\n* **mixup/cutmix is all you need**\nby @machinelp \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126504\nAn implementation of mixup & cutmix. I find it very useful!\n\n* **why cutout/mixup works**\nby @hengck23 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/127902\nAn explanation of why cutout/mixup works, and gave an idea of making dataset larger.\n\n* **GridMask implementation**\nby @haqishen \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128161\nIt works. I'm sure about it.\n\n* **An Implementation of Augmix Based on albumentations**\nby @haqishen\nhttps://www.kaggle.com/haqishen/augmix-based-on-albumentations\nGoogle's work. I'm doing experiment on this recently.\n\n* **Morphological transformations as image augmentation**\nby @ren4yu \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128198\nSome special augmentation methods\n\n* **Generating More data**\nby @drhabib \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128059\nAnother way to generate data.\n\n* **Image preprocessing (128x128)**\nby @iafoss \nhttps://www.kaggle.com/iafoss/image-preprocessing-128x128\nCrop out black background.\n\n* **iterative stratification**\nby @yiheng \nhttps://www.kaggle.com/yiheng/iterative-stratification\nHow to split training set.\n\n* **Solving grapheme_root is the key to 0.98?**\nby @bibek777 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/124904\nSome ideas were shared in this post.\n\n* **Approach for 0.97 and What's next?**\nby @ildoonet \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/125828\nSome ideas for going to LB 0.97\n\n* **Tricks for Image Classification......**\nby @machinelp \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128914\nMany ideas collected by @machinelp \n\n* **Approach for 0.97 with 64x64x1 input**\nby @shujun717 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128368\nLB 0.9718 reached by 64x64x1 input??? That's fantastic!\n\n* **Efficientnet - Trials and Analysis**\nby @dhakshiin1601 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/128911\nEfficientnet and GeM.\n\n* **Flipping properties of Bengali characters?**\nby @roguekk007 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126761\n\n* **Papers worth reading?**\nby @bibek777 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/127719\nPapers that look worth reading.\n\n* **Bengali Data Sets & Publications!**\nby @ipythonx \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/122604\nPapers that in Bengali x CV field.\n\n* **[placeholder] autoML results**\nby @hengck23 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/126337\nExperiment result of AutoML.\n\n* **Teacher-Student**\nby @mightyrains \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/125046\nKnowledge distillation.\n\n* **Label Smoothing and TTA**\nby @karan07 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/125221\n\n* **Shake-Shake**\nby @phoenix9032 \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/124249\n\n* **Some experiments with CNN tails**\nby @drhabib \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/123432\n\n\n\n## Visualization\n\n* **Bengali - Quick EDA**\nby @pestipeti \nhttps://www.kaggle.com/pestipeti/bengali-quick-eda\n\n* **Bengali.AI Handwritten Grapheme - Getting Started**\nby @gpreda \nhttps://www.kaggle.com/gpreda/bengali-ai-handwritten-grapheme-getting-started\n\n* **Visualize your results**\nby @pestipeti \nhttps://www.kaggle.com/c/bengaliai-cv19/discussion/127149\n\n\n", "737985": "░░░░░▄▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n▄███▀░◐░░░▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▐░░░░░▐░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░░▐▄▄░░░░░░░░░░I need to make one comment :)░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▌░░░░▄▀▒▒▀▀▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▒▒▒▒▒▒▒▒▀▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░▐░░░░▐▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░▀▄░░░░▀▄▒▒▒▒▒▒▒▒▒▒▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░▀▄▄▄▄▄█▄▄▄▄▄▄▄▄▄▄▄▀▄░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░░░▌▌░▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░\n░░░░░░░░░▄▄▌▌▄▌▌░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░", "731030": "👍 ", "764171": "Cool Notebook!", "757482": "Thanks for the collection, it helped me a lot.", "752217": "nice", "744750": "This is great !", "739654": "Thanks a lot of this summarising everything, it's very informative.", "739633": "Thanks for informative summary, it's very helpful", "737057": "Thanks a lot @haqishen , I will refer to this to be able to get inspired and improve my performance", "734655": "Awesome thread! And the \"Frog brother\" nickname is great. ;)", "733638": "wow. its fantastiic. one is similar to my idea. but i didnt make it. its very useful. thank you. ", "731459": "Ahh Thanks for compiling this =) Helps for new people and for old folks like me =) ", "731361": "Love all the sharing in the comp, very helpful for beginners!", "731223": "@haqishen Thanks a lot for gathering all the important pieces in one place. It would be nice if you could add your own suggestions for the participants as well! Of course, I am not talking about 0.9923, but promising directions in an abstract view at least!", "730964": "Wow, This is extremely helpful. Thanks @haqishen 💚 ", "730987": "Thank you for including my Kernel. And congratulations for your top position.", "764669": "", "733524": "", "759519": "Thank you for sharing @haqishen .", "749846": "Thanks for shairng", "733212": "Thanks for sharing this resource", "732733": "Thanks for the gathering! Very helpful!", "732677": "Thank you for the resource!", "732303": "Thanks for sharing.", "732161": "Thanks, it will help me a lot!!! 😁 ", "732298": "Very helpful! Thanks for sharing!!😄 👍 " } }