![\"\"](\"https://i.ibb.co/bbRMrZ8/the-first-rule.png\")
Hi to all!
\nIt has been more than 5 months since the introduction of EfficientNetV2. But still this family of models is practically not used in competitions. In my few attempts in previous competitions, these models showed worse results than the classic EfficientNet. For the first time in this competition, the results are similar, but I can't say that EfficientNetV2 gives at least some advantage over EfficientNet.
\nWhat are your thoughts on this?
\nGood luck to all!
Hi to all!
\nIt has been more than 5 months since the introduction of EfficientNetV2. But still this family of models is practically not used in competitions. In my few attempts in previous competitions, these models showed worse results than the classic EfficientNet. For the first time in this competition, the results are similar, but I can't say that EfficientNetV2 gives at least some advantage over EfficientNet.
\nWhat are your thoughts on this?
\nGood luck to all!
Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
looks like you forget to relogin
", "rawMarkdown": "looks like you forget to relogin ", "votes": 10 }, { "id": 1519801, "postDate": "2021-09-21T23:34:53.073Z", "content": "lol 😆
\nIs it allowed to have more than one account?
", "rawMarkdown": "lol 😆\n\nIs it allowed to have more than one account?" }, { "id": 1519810, "postDate": "2021-09-21T23:59:11.780Z", "content": "\n\nlooks like you forget to relogin
\n
lolol. I was thinking the same thing but didn't wanna stir anything 😹
\n\n\nIs it allowed to have more than one account?
\n
Literally the first rule of Kaggle Club
\n
Guys today we have a lot of opportunities to train models on TPU and do not need to use kaggle :)
", "rawMarkdown": "Guys today we have a lot of opportunities to train models on TPU and do not need to use kaggle :)" }, { "id": 1520145, "postDate": "2021-09-22T06:51:38.397Z", "content": "Still, Kaggle TPUs beat Colab TPUs since they have more memory.
\nAnd Colab alternatives are generally more expensive.
One can violate the rules and create N different accounts to use more Kaggle TPUs, forcing other participants to be badly queued.
\nIt'd be nice to investigate this one. TS said about \"public score\", but the account has zero submissions to this competition. Otherwise, we'd see \"1234th in this Competition\" near the author's nickname.
To me, the capacity provided by the colab to train one or two models a day is enough. The kaggle team finds people who use multiple accounts. I may be wrong, but one of the grandmasters was caught.
", "rawMarkdown": "To me, the capacity provided by the colab to train one or two models a day is enough. The kaggle team finds people who use multiple accounts. I may be wrong, but one of the grandmasters was caught." }, { "id": 1520441, "postDate": "2021-09-22T10:29:08.510Z", "content": "It's not enough for us.
\nOur pipeline uses multiple steps (signal preprocessing, normalization, CQT, model training and validation, predictions post-processing).
\nEach of these steps requires multiple parameters that are difficult to tune via some heuristics.
\nThe best way in terms of quality is to adjust the parameters by running the entire pipeline, at least on the reduced dataset.
Right now, we use 2 Kaggle accounts (personal accounts, of course) and 2 Colab accounts (Pro and Pro+), and we still have a lot of parameters to tune and hypotheses to verify.
", "rawMarkdown": "It's not enough for us.\n\nOur pipeline uses multiple steps (signal preprocessing, normalization, CQT, model training and validation, predictions post-processing).\nEach of these steps requires multiple parameters that are difficult to tune via some heuristics. \nThe best way in terms of quality is to adjust the parameters by running the entire pipeline, at least on the reduced dataset.\n\nRight now, we use 2 Kaggle accounts (personal accounts, of course) and 2 Colab accounts (Pro and Pro+), and we still have a lot of parameters to tune and hypotheses to verify.", "votes": 1 }, { "id": 1520960, "postDate": "2021-09-22T18:14:56.683Z", "content": "@hannerandler Oh no, it looks like you accidentally click delete! no worries i got you: ![\"\"](\"https://i.imgur.com/YmtI2eg.png\")
What a coincidence! 🙂
", "rawMarkdown": "What a coincidence! 🙂" } ] } ], "comments": [ { "id": 1559733, "author_name": "ChristopherZerafa", "author_url": "", "post_date": "2021-10-27T07:10:10.023000", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
looks like you forget to relogin
", "votes": 10, "replies": [] }, { "id": 1519801, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2021-09-21T23:34:53.073000", "content": "lol 😆
\nIs it allowed to have more than one account?
", "votes": 0, "replies": [] }, { "id": 1519810, "author_name": "عثمان", "author_url": "", "post_date": "2021-09-21T23:59:11.780000", "content": "\n\nlooks like you forget to relogin
\n
lolol. I was thinking the same thing but didn't wanna stir anything 😹
\n\n\nIs it allowed to have more than one account?
\n
Literally the first rule of Kaggle Club
\n
Guys today we have a lot of opportunities to train models on TPU and do not need to use kaggle :)
", "votes": 0, "replies": [] }, { "id": 1520145, "author_name": "Ruslan Vdovychenko", "author_url": "", "post_date": "2021-09-22T06:51:38.397000", "content": "Still, Kaggle TPUs beat Colab TPUs since they have more memory.
\nAnd Colab alternatives are generally more expensive.
One can violate the rules and create N different accounts to use more Kaggle TPUs, forcing other participants to be badly queued.
\nIt'd be nice to investigate this one. TS said about \"public score\", but the account has zero submissions to this competition. Otherwise, we'd see \"1234th in this Competition\" near the author's nickname.
To me, the capacity provided by the colab to train one or two models a day is enough. The kaggle team finds people who use multiple accounts. I may be wrong, but one of the grandmasters was caught.
", "votes": 0, "replies": [] }, { "id": 1520441, "author_name": "Ruslan Vdovychenko", "author_url": "", "post_date": "2021-09-22T10:29:08.510000", "content": "It's not enough for us.
\nOur pipeline uses multiple steps (signal preprocessing, normalization, CQT, model training and validation, predictions post-processing).
\nEach of these steps requires multiple parameters that are difficult to tune via some heuristics.
\nThe best way in terms of quality is to adjust the parameters by running the entire pipeline, at least on the reduced dataset.
Right now, we use 2 Kaggle accounts (personal accounts, of course) and 2 Colab accounts (Pro and Pro+), and we still have a lot of parameters to tune and hypotheses to verify.
", "votes": 1, "replies": [] }, { "id": 1520960, "author_name": "Gleb", "author_url": "", "post_date": "2021-09-22T18:14:56.683000", "content": "@hannerandler Oh no, it looks like you accidentally click delete! no worries i got you:
What a coincidence! 🙂
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1519644": "Hi to all! \nIt has been more than 5 months since the introduction of EfficientNetV2. But still this family of models is practically not used in competitions. In my few attempts in previous competitions, these models showed worse results than the classic EfficientNet. For the first time in this competition, the results are similar, but I can't say that EfficientNetV2 gives at least some advantage over EfficientNet. \nWhat are your thoughts on this?\nGood luck to all!", "1559733": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris", "1519760": "" } }