Inspired by one of the solutions on Shopee competition when one of the solutions listed what didn't work (e.g. : \"Some diffusion technique; Attention-based query expantion; MLP trained by embedding based pairwise dataset; Unsupervised model (like CLIP); Arcface with triplet loss\")
\nI'd appreciate to know what won't work on this Comp.
\nThat kind of question is helpful in any other profession. Mostly for beginners, including myself. In fact, I'm below than beginner level. So this topic doubt, isn't a prerogative of Kaggle ML Competitions.
\nTherefore, feel free to share anything till the end of the competition.
\nPreferably, at the end, since your findings will appear during your path.
\nAnd what works will figure in many Winners Solutions.
\nThanks in advance Dear Kagglers.
", "messages": [ { "id": "1685011", "postDate": "02/10/2022 23:53:19", "content": "Inspired by one of the solutions on Shopee competition when one of the solutions listed what didn't work (e.g. : \"Some diffusion technique; Attention-based query expantion; MLP trained by embedding based pairwise dataset; Unsupervised model (like CLIP); Arcface with triplet loss\")
\nI'd appreciate to know what won't work on this Comp.
\nThat kind of question is helpful in any other profession. Mostly for beginners, including myself. In fact, I'm below than beginner level. So this topic doubt, isn't a prerogative of Kaggle ML Competitions.
\nTherefore, feel free to share anything till the end of the competition.
\nPreferably, at the end, since your findings will appear during your path.
\nAnd what works will figure in many Winners Solutions.
\nThanks in advance Dear Kagglers.
", "rawMarkdown": "Inspired by one of the solutions on Shopee competition when one of the solutions listed what didn't work (e.g. : \"Some diffusion technique; Attention-based query expantion; MLP trained by embedding based pairwise dataset; Unsupervised model (like CLIP); Arcface with triplet loss\")\n\nI'd appreciate to know what won't work on this Comp. \n\nThat kind of question is helpful in any other profession. Mostly for beginners, including myself. In fact, I'm below than beginner level. So this topic doubt, isn't a prerogative of Kaggle ML Competitions.\n\nTherefore, feel free to share anything till the end of the competition. \n\nPreferably, at the end, since your findings will appear during your path.\n\nAnd what works will figure in many Winners Solutions.\n\nThanks in advance Dear Kagglers.", "votes": null }, { "id": "1686044", "postDate": "02/11/2022 18:01:41", "content": "\n\nI'd appreciate to know what won't work on this Comp.
\n
There's only one way to find out :)
", "rawMarkdown": "> I'd appreciate to know what won't work on this Comp.\n\nThere's only one way to find out :)", "votes": null }, { "id": "1686097", "postDate": "02/11/2022 18:52:15", "content": "Indeed Sanyam, working on it.
\nMaybe those that joined the comp could share some insights During and mostly After the deadline.
\nWriting about their Dos and Don'ts (work/won't work). In fact, it's more objective than reading \"last competition solutions\" when beginners don't have any clue about where to start. And some codes are outdated.
\nOr even some veterans, that are participating or not could say some inspirational words.
Thanks for commenting here.
", "rawMarkdown": "Indeed Sanyam, working on it.\n\nMaybe those that joined the comp could share some insights During and mostly After the deadline.\nWriting about their Dos and Don'ts (work/won't work). In fact, it's more objective than reading \"last competition solutions\" when beginners don't have any clue about where to start. And some codes are outdated. \nOr even some veterans, that are participating or not could say some inspirational words.\n\nThanks for commenting here.", "votes": null }, { "id": "1687014", "postDate": "02/12/2022 14:34:45", "content": "Thanks for clarifying. I think most winners share things that didn't work and I have found them to be very open in answering questions etc too, we need to ask the right questions 😄
", "rawMarkdown": "Thanks for clarifying. I think most winners share things that didn't work and I have found them to be very open in answering questions etc too, we need to ask the right questions 😄", "votes": null }, { "id": "1687455", "postDate": "02/12/2022 21:55:33", "content": "\"Most winners share things that didn't work\" That's intriguing, revealing and totally unexpected. Unless they write literally what didn't work.
\nI could never write about this since I don't have any knowledge. Besides, I barely understand competitions and everything about them.
\nCommunication skills help us to evolve. You should know what to ask, what you intend to learn. Never forget timing: when to ask and mostly who are you going to ask. You know what I mean.
\nIt may sound easy, however I had already some moments on Kaggle that I had to explain what I intended to say.
\nMaybe exact sciences can cause less confusion than users communication.
A beginner has no way to discriminate between the more specific techniques and theories one would find in a winning solutions listing of \"What didn't work\". The beginner still has to grasp the fundamentals (i.e. kNN or Naive Bayes or any other model, AUC or RMSE or any other means of evaluation, and so on), so variants or cutting-edge developments may not be as productive and rewarding as experimenting (and failing) with more basic implementations.
\nPerhaps the question to ask as a beginner is which solutions will make you beat the benchmarks, not which solutions will make you a winner. And twisting it: which solutions will perform worse than the benchmarks?
\nBut yeah, the problem in this competition doesn't seem to be a regression problem, right? Also, this is a recommendations problem, so both some knowledge of that whole field and probably some domain knowledge of retail, (fast) fashion, consumer behavior and so on will be helpful in understanding the results of your work.
", "rawMarkdown": "A beginner has no way to discriminate between the more specific techniques and theories one would find in a winning solutions listing of *\"What didn't work\"*. The beginner still has to grasp the fundamentals (i.e. *kNN* or *Naive Bayes* or any other model, *AUC* or *RMSE* or any other means of evaluation, and so on), so variants or cutting-edge developments may not be as productive and rewarding as experimenting (and failing) with more basic implementations.\n\nPerhaps the question to ask as a beginner is which solutions will make you beat the benchmarks, not which solutions will make you a winner. And twisting it: *which solutions will perform worse than the benchmarks?*\n\nBut yeah, the problem in this competition doesn't seem to be a regression problem, right? Also, this is a recommendations problem, so both some knowledge of that whole field and probably some domain knowledge of retail, (fast) fashion, consumer behavior and so on will be helpful in understanding the results of your work.", "votes": null }, { "id": "1692951", "postDate": "02/16/2022 11:11:39", "content": "Thanks for sharing your valuable insight of the competition.
", "rawMarkdown": "Thanks for sharing your valuable insight of the competition.", "votes": null } ], "comments": [ { "id": 1686044, "author_name": "init27", "author_url": "", "post_date": "02/11/2022 18:01:41", "content": "\n\nI'd appreciate to know what won't work on this Comp.
\n
There's only one way to find out :)
", "votes": null, "replies": [ { "id": 1686097, "author_name": "mpwolke", "author_url": "", "post_date": "02/11/2022 18:52:15", "content": "Indeed Sanyam, working on it.
\nMaybe those that joined the comp could share some insights During and mostly After the deadline.
\nWriting about their Dos and Don'ts (work/won't work). In fact, it's more objective than reading \"last competition solutions\" when beginners don't have any clue about where to start. And some codes are outdated.
\nOr even some veterans, that are participating or not could say some inspirational words.
Thanks for commenting here.
", "votes": null, "replies": [] }, { "id": 1687014, "author_name": "init27", "author_url": "", "post_date": "02/12/2022 14:34:45", "content": "Thanks for clarifying. I think most winners share things that didn't work and I have found them to be very open in answering questions etc too, we need to ask the right questions 😄
", "votes": null, "replies": [] }, { "id": 1687455, "author_name": "mpwolke", "author_url": "", "post_date": "02/12/2022 21:55:33", "content": "\"Most winners share things that didn't work\" That's intriguing, revealing and totally unexpected. Unless they write literally what didn't work.
\nI could never write about this since I don't have any knowledge. Besides, I barely understand competitions and everything about them.
\nCommunication skills help us to evolve. You should know what to ask, what you intend to learn. Never forget timing: when to ask and mostly who are you going to ask. You know what I mean.
\nIt may sound easy, however I had already some moments on Kaggle that I had to explain what I intended to say.
\nMaybe exact sciences can cause less confusion than users communication.
A beginner has no way to discriminate between the more specific techniques and theories one would find in a winning solutions listing of \"What didn't work\". The beginner still has to grasp the fundamentals (i.e. kNN or Naive Bayes or any other model, AUC or RMSE or any other means of evaluation, and so on), so variants or cutting-edge developments may not be as productive and rewarding as experimenting (and failing) with more basic implementations.
\nPerhaps the question to ask as a beginner is which solutions will make you beat the benchmarks, not which solutions will make you a winner. And twisting it: which solutions will perform worse than the benchmarks?
\nBut yeah, the problem in this competition doesn't seem to be a regression problem, right? Also, this is a recommendations problem, so both some knowledge of that whole field and probably some domain knowledge of retail, (fast) fashion, consumer behavior and so on will be helpful in understanding the results of your work.
", "votes": null, "replies": [ { "id": 1692951, "author_name": "mpwolke", "author_url": "", "post_date": "02/16/2022 11:11:39", "content": "Thanks for sharing your valuable insight of the competition.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1685011": "Inspired by one of the solutions on Shopee competition when one of the solutions listed what didn't work (e.g. : \"Some diffusion technique; Attention-based query expantion; MLP trained by embedding based pairwise dataset; Unsupervised model (like CLIP); Arcface with triplet loss\")\n\nI'd appreciate to know what won't work on this Comp. \n\nThat kind of question is helpful in any other profession. Mostly for beginners, including myself. In fact, I'm below than beginner level. So this topic doubt, isn't a prerogative of Kaggle ML Competitions.\n\nTherefore, feel free to share anything till the end of the competition. \n\nPreferably, at the end, since your findings will appear during your path.\n\nAnd what works will figure in many Winners Solutions.\n\nThanks in advance Dear Kagglers.", "1686044": "> I'd appreciate to know what won't work on this Comp.\n\nThere's only one way to find out :)", "1686097": "Indeed Sanyam, working on it.\n\nMaybe those that joined the comp could share some insights During and mostly After the deadline.\nWriting about their Dos and Don'ts (work/won't work). In fact, it's more objective than reading \"last competition solutions\" when beginners don't have any clue about where to start. And some codes are outdated. \nOr even some veterans, that are participating or not could say some inspirational words.\n\nThanks for commenting here.", "1687014": "Thanks for clarifying. I think most winners share things that didn't work and I have found them to be very open in answering questions etc too, we need to ask the right questions 😄", "1687455": "\"Most winners share things that didn't work\" That's intriguing, revealing and totally unexpected. Unless they write literally what didn't work.\n\nI could never write about this since I don't have any knowledge. Besides, I barely understand competitions and everything about them. \n\nCommunication skills help us to evolve. You should know what to ask, what you intend to learn. Never forget timing: when to ask and mostly who are you going to ask. You know what I mean.\n\n It may sound easy, however I had already some moments on Kaggle that I had to explain what I intended to say. \nMaybe exact sciences can cause less confusion than users communication.", "1692342": "A beginner has no way to discriminate between the more specific techniques and theories one would find in a winning solutions listing of *\"What didn't work\"*. The beginner still has to grasp the fundamentals (i.e. *kNN* or *Naive Bayes* or any other model, *AUC* or *RMSE* or any other means of evaluation, and so on), so variants or cutting-edge developments may not be as productive and rewarding as experimenting (and failing) with more basic implementations.\n\nPerhaps the question to ask as a beginner is which solutions will make you beat the benchmarks, not which solutions will make you a winner. And twisting it: *which solutions will perform worse than the benchmarks?*\n\nBut yeah, the problem in this competition doesn't seem to be a regression problem, right? Also, this is a recommendations problem, so both some knowledge of that whole field and probably some domain knowledge of retail, (fast) fashion, consumer behavior and so on will be helpful in understanding the results of your work.", "1692951": "Thanks for sharing your valuable insight of the competition." }, "source": "meta" }