This competition is so strange that the high score kernel comes up frequently and it is hard to keep the ranking . When i wake up, my rank is greatly reduced.
", "messages": [ { "id": "400225", "postDate": "10/07/2018 23:41:57", "content": "This competition is so strange that the high score kernel comes up frequently and it is hard to keep the ranking . When i wake up, my rank is greatly reduced.
", "rawMarkdown": "This competition is so strange that the high score kernel comes up frequently and it is hard to keep the ranking . When i wake up, my rank is greatly reduced.", "votes": null }, { "id": "400330", "postDate": "10/08/2018 05:54:55", "content": "Relax
", "rawMarkdown": "Relax", "votes": null }, { "id": "400385", "postDate": "10/08/2018 08:19:22", "content": "I think since there is still one month to go, it seems ok. However, there was a recent discussion about high scoring kernels where most people were favoring sharing stuff below the medal range. I also agree with this for the sake of competition spirit. Instead of giving people one click ahead silver medal submissions, it's more appropriate to give them insights and let them figure out new features by themselves. Like this.
", "rawMarkdown": "I think since there is still one month to go, it seems ok. However, there was [a recent discussion][1] about high scoring kernels where most people were favoring sharing stuff below the medal range. I also agree with this for the sake of competition spirit. Instead of giving people one click ahead silver medal submissions, it's more appropriate to give them insights and let them figure out new features by themselves. [Like this][2].\n\n\n [1]: http://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/62162\n [2]: http://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/67767", "votes": null }, { "id": "400476", "postDate": "10/08/2018 11:41:41", "content": "Thanks for your reply. I'll take care of getting insights or features. and I'll try hard for the rest of the month. \n:)
", "rawMarkdown": "Thanks for your reply. I'll take care of getting insights or features. and I'll try hard for the rest of the month. \n:)", "votes": null }, { "id": "400509", "postDate": "10/08/2018 12:43:01", "content": ": D
", "rawMarkdown": ": D", "votes": null }, { "id": "400855", "postDate": "10/09/2018 03:13:17", "content": "Effective ensembling is your best friend in this case. When someone makes available a high-scoring public kernel, think of it as a gift. You can slightly modify the code to generate OOF predictions and then blend it with your own excellent solution to outdo the one-clickers!
", "rawMarkdown": "Effective ensembling is your best friend in this case. When someone makes available a high-scoring public kernel, think of it as a gift. You can slightly modify the code to generate OOF predictions and then blend it with your own excellent solution to outdo the one-clickers!", "votes": null }, { "id": "400858", "postDate": "10/09/2018 03:19:18", "content": "Also, there's value beyond the ranking for a single contest. You can learn so many great techniques from these superior kernels that will benefit you outside of Kaggle!
", "rawMarkdown": "Also, there's value beyond the ranking for a single contest. You can learn so many great techniques from these superior kernels that will benefit you outside of Kaggle!", "votes": null }, { "id": "401075", "postDate": "10/09/2018 11:46:33", "content": "Thanks for the reply. There are many things to learn while doing this contest. I'll try harder.
", "rawMarkdown": "Thanks for the reply. There are many things to learn while doing this contest. I'll try harder.", "votes": null }, { "id": "401096", "postDate": "10/09/2018 12:35:45", "content": "Or you can sit back wait for Olivier to produce a script AND THEN wait for Ashish to produce the ensemble! LOL ;)
", "rawMarkdown": "Or you can sit back wait for Olivier to produce a script AND THEN wait for Ashish to produce the ensemble! LOL ;)", "votes": null }, { "id": "401110", "postDate": "10/09/2018 12:57:18", "content": "First time i ever laughed out loud reading kaggle discussions =D
", "rawMarkdown": "First time i ever laughed out loud reading kaggle discussions =D", "votes": null }, { "id": "401190", "postDate": "10/09/2018 15:20:40", "content": "Don't forget that there are only two ways to get gold medals on kernels; a good EDA at the start and a leader board killer during the competition.
\n\nThis is unfortunate but is real life I am afraid.
\n\nI am a cynic and think Kaggle may have used kernels to artificially boost the overall mean scores as a sort of vanity metric for their clients. They could do something about LB killers but they have chosen not to especially in the last week. Hopefully this won't be a problem this time but I am not holding my breath.
\n\nI am coming to the realization that for efficiency, submit a placeholder when the competition starts and then join the comp for real in the last two weeks.
\n\nI should mention that this isn't intended as a criticism of either Olivier or Ashish as I know they don't release LB shakers in the last week.
", "rawMarkdown": "Don't forget that there are only two ways to get gold medals on kernels; a good EDA at the start and a leader board killer during the competition.\n\nThis is unfortunate but is real life I am afraid.\n\nI am a cynic and think Kaggle may have used kernels to artificially boost the overall mean scores as a sort of vanity metric for their clients. They could do something about LB killers but they have chosen not to especially in the last week. Hopefully this won't be a problem this time but I am not holding my breath.\n\nI am coming to the realization that for efficiency, submit a placeholder when the competition starts and then join the comp for real in the last two weeks.\n\nI should mention that this isn't intended as a criticism of either Olivier or Ashish as I know they don't release LB shakers in the last week.", "votes": null }, { "id": "401206", "postDate": "10/09/2018 16:00:53", "content": "Agree.
", "rawMarkdown": "Agree.", "votes": null }, { "id": "401356", "postDate": "10/09/2018 23:19:39", "content": "Thanks. I'll keep in mind.
", "rawMarkdown": "Thanks. I'll keep in mind.", "votes": null } ], "comments": [ { "id": 400330, "author_name": "marcuslin", "author_url": "", "post_date": "10/08/2018 05:54:55", "content": "Relax
", "votes": null, "replies": [ { "id": 400509, "author_name": "chocozzz", "author_url": "", "post_date": "10/08/2018 12:43:01", "content": ": D
", "votes": null, "replies": [] } ] }, { "id": 400385, "author_name": "fatihozturk", "author_url": "", "post_date": "10/08/2018 08:19:22", "content": "I think since there is still one month to go, it seems ok. However, there was a recent discussion about high scoring kernels where most people were favoring sharing stuff below the medal range. I also agree with this for the sake of competition spirit. Instead of giving people one click ahead silver medal submissions, it's more appropriate to give them insights and let them figure out new features by themselves. Like this.
", "votes": null, "replies": [ { "id": 400476, "author_name": "chocozzz", "author_url": "", "post_date": "10/08/2018 11:41:41", "content": "Thanks for your reply. I'll take care of getting insights or features. and I'll try hard for the rest of the month. \n:)
", "votes": null, "replies": [] }, { "id": 401190, "author_name": "scirpus", "author_url": "", "post_date": "10/09/2018 15:20:40", "content": "Don't forget that there are only two ways to get gold medals on kernels; a good EDA at the start and a leader board killer during the competition.
\n\nThis is unfortunate but is real life I am afraid.
\n\nI am a cynic and think Kaggle may have used kernels to artificially boost the overall mean scores as a sort of vanity metric for their clients. They could do something about LB killers but they have chosen not to especially in the last week. Hopefully this won't be a problem this time but I am not holding my breath.
\n\nI am coming to the realization that for efficiency, submit a placeholder when the competition starts and then join the comp for real in the last two weeks.
\n\nI should mention that this isn't intended as a criticism of either Olivier or Ashish as I know they don't release LB shakers in the last week.
", "votes": null, "replies": [] }, { "id": 401206, "author_name": "chinta", "author_url": "", "post_date": "10/09/2018 16:00:53", "content": "Agree.
", "votes": null, "replies": [] }, { "id": 401356, "author_name": "chocozzz", "author_url": "", "post_date": "10/09/2018 23:19:39", "content": "Thanks. I'll keep in mind.
", "votes": null, "replies": [] } ] }, { "id": 400855, "author_name": "jpmiller", "author_url": "", "post_date": "10/09/2018 03:13:17", "content": "Effective ensembling is your best friend in this case. When someone makes available a high-scoring public kernel, think of it as a gift. You can slightly modify the code to generate OOF predictions and then blend it with your own excellent solution to outdo the one-clickers!
", "votes": null, "replies": [ { "id": 400858, "author_name": "jpmiller", "author_url": "", "post_date": "10/09/2018 03:19:18", "content": "Also, there's value beyond the ranking for a single contest. You can learn so many great techniques from these superior kernels that will benefit you outside of Kaggle!
", "votes": null, "replies": [] }, { "id": 401075, "author_name": "chocozzz", "author_url": "", "post_date": "10/09/2018 11:46:33", "content": "Thanks for the reply. There are many things to learn while doing this contest. I'll try harder.
", "votes": null, "replies": [] }, { "id": 401096, "author_name": "scirpus", "author_url": "", "post_date": "10/09/2018 12:35:45", "content": "Or you can sit back wait for Olivier to produce a script AND THEN wait for Ashish to produce the ensemble! LOL ;)
", "votes": null, "replies": [] }, { "id": 401110, "author_name": "tpthegreat", "author_url": "", "post_date": "10/09/2018 12:57:18", "content": "First time i ever laughed out loud reading kaggle discussions =D
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "400225": "This competition is so strange that the high score kernel comes up frequently and it is hard to keep the ranking . When i wake up, my rank is greatly reduced.", "400330": "Relax", "400385": "I think since there is still one month to go, it seems ok. However, there was [a recent discussion][1] about high scoring kernels where most people were favoring sharing stuff below the medal range. I also agree with this for the sake of competition spirit. Instead of giving people one click ahead silver medal submissions, it's more appropriate to give them insights and let them figure out new features by themselves. [Like this][2].\n\n\n [1]: http://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/62162\n [2]: http://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/67767", "400476": "Thanks for your reply. I'll take care of getting insights or features. and I'll try hard for the rest of the month. \n:)", "400509": ": D", "400855": "Effective ensembling is your best friend in this case. When someone makes available a high-scoring public kernel, think of it as a gift. You can slightly modify the code to generate OOF predictions and then blend it with your own excellent solution to outdo the one-clickers!", "400858": "Also, there's value beyond the ranking for a single contest. You can learn so many great techniques from these superior kernels that will benefit you outside of Kaggle!", "401075": "Thanks for the reply. There are many things to learn while doing this contest. I'll try harder.", "401096": "Or you can sit back wait for Olivier to produce a script AND THEN wait for Ashish to produce the ensemble! LOL ;)", "401110": "First time i ever laughed out loud reading kaggle discussions =D", "401190": "Don't forget that there are only two ways to get gold medals on kernels; a good EDA at the start and a leader board killer during the competition.\n\nThis is unfortunate but is real life I am afraid.\n\nI am a cynic and think Kaggle may have used kernels to artificially boost the overall mean scores as a sort of vanity metric for their clients. They could do something about LB killers but they have chosen not to especially in the last week. Hopefully this won't be a problem this time but I am not holding my breath.\n\nI am coming to the realization that for efficiency, submit a placeholder when the competition starts and then join the comp for real in the last two weeks.\n\nI should mention that this isn't intended as a criticism of either Olivier or Ashish as I know they don't release LB shakers in the last week.", "401206": "Agree.", "401356": "Thanks. I'll keep in mind." }, "source": "meta" }