I understand the need but I really dislike the API submission process and it's lack of any decent error feedback. I do most of my work on local PC's and I am dreading the time in a very short number of days were I need to take my work into a Kaggle kernel. Just in early testing I have already racked up a nice number of submissions failures.
\nThis competition was ideal for Kaggle to try some innovative methods of helping users with submission failures. Also a perfect competition to NOT count submission failures against our daily quota.
\nWhy was it a perfect chance - because a private leader board that has yet to come into existence is impossible to probe. If anyone can probe the future - please message me now - I am interested in betting on the baseball games that will be played in August.
", "messages": [ { "id": "1390718", "postDate": "07/17/2021 03:21:30", "content": "I understand the need but I really dislike the API submission process and it's lack of any decent error feedback. I do most of my work on local PC's and I am dreading the time in a very short number of days were I need to take my work into a Kaggle kernel. Just in early testing I have already racked up a nice number of submissions failures.
\nThis competition was ideal for Kaggle to try some innovative methods of helping users with submission failures. Also a perfect competition to NOT count submission failures against our daily quota.
\nWhy was it a perfect chance - because a private leader board that has yet to come into existence is impossible to probe. If anyone can probe the future - please message me now - I am interested in betting on the baseball games that will be played in August.
", "rawMarkdown": "I understand the need but I really dislike the API submission process and it's lack of any decent error feedback. I do most of my work on local PC's and I am dreading the time in a very short number of days were I need to take my work into a Kaggle kernel. Just in early testing I have already racked up a nice number of submissions failures.\n\nThis competition **was ideal** for Kaggle to try some innovative methods of helping users with submission failures. Also a **perfect** competition to NOT count submission failures against our daily quota.\n\nWhy was it a perfect chance - because a private leader board that has yet to come into existence is impossible to probe. If anyone can probe the future - please message me now - I am interested in betting on the baseball games that will be played in August.", "votes": null }, { "id": "1391170", "postDate": "07/17/2021 11:07:04", "content": "Yeah it sucks, and I think for this competition especially having the lack of error messages is detrimental, as many people will have quite complex pipelines. But obviously it is a challenge to give useful information without potential for probing.
\nMy biggest frustration is the fact that the public test set (that it runs on during commit) is so small. It's something like 5 days? And incidentally these 5 days seem to be quite nice and don't contain a lot of the trip ups that have caused submission errors. I think a bigger public test would potentially alleviate a lot of the issues people are facing.
\nI would advise you to feed in your test pipeline slowly. I found it easier to do a join and submission at a time and build it up. That way it was easier to determine which steps in the pipeline caused the issues.
", "rawMarkdown": "Yeah it sucks, and I think for this competition especially having the lack of error messages is detrimental, as many people will have quite complex pipelines. But obviously it is a challenge to give useful information without potential for probing. \n\nMy biggest frustration is the fact that the public test set (that it runs on during commit) is so small. It's something like 5 days? And incidentally these 5 days seem to be quite nice and don't contain a lot of the trip ups that have caused submission errors. I think a bigger public test would potentially alleviate a lot of the issues people are facing.\n\nI would advise you to feed in your test pipeline slowly. I found it easier to do a join and submission at a time and build it up. That way it was easier to determine which steps in the pipeline caused the issues.", "votes": null }, { "id": "1391619", "postDate": "07/17/2021 19:19:53", "content": "Again - for this competition probing has almost ZERO value since the final private LB data has yet to be written and the games have yet to be played.
\nWhat worries me a lot is what issues will come with the August data and the final submissions :(
\nAgree - build up slow cause failures with no error codes are a real beast to figure out.
", "rawMarkdown": "Again - for this competition probing has almost ZERO value since the final private LB data has yet to be written and the games have yet to be played. \n\nWhat worries me a lot is what issues will come with the August data and the final submissions :(\n\nAgree - build up slow cause failures with no error codes are a real beast to figure out.", "votes": null } ], "comments": [ { "id": 1391170, "author_name": "jacobhowardparker", "author_url": "", "post_date": "07/17/2021 11:07:04", "content": "Yeah it sucks, and I think for this competition especially having the lack of error messages is detrimental, as many people will have quite complex pipelines. But obviously it is a challenge to give useful information without potential for probing.
\nMy biggest frustration is the fact that the public test set (that it runs on during commit) is so small. It's something like 5 days? And incidentally these 5 days seem to be quite nice and don't contain a lot of the trip ups that have caused submission errors. I think a bigger public test would potentially alleviate a lot of the issues people are facing.
\nI would advise you to feed in your test pipeline slowly. I found it easier to do a join and submission at a time and build it up. That way it was easier to determine which steps in the pipeline caused the issues.
", "votes": null, "replies": [ { "id": 1391619, "author_name": "pcjimmmy", "author_url": "", "post_date": "07/17/2021 19:19:53", "content": "Again - for this competition probing has almost ZERO value since the final private LB data has yet to be written and the games have yet to be played.
\nWhat worries me a lot is what issues will come with the August data and the final submissions :(
\nAgree - build up slow cause failures with no error codes are a real beast to figure out.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1390718": "I understand the need but I really dislike the API submission process and it's lack of any decent error feedback. I do most of my work on local PC's and I am dreading the time in a very short number of days were I need to take my work into a Kaggle kernel. Just in early testing I have already racked up a nice number of submissions failures.\n\nThis competition **was ideal** for Kaggle to try some innovative methods of helping users with submission failures. Also a **perfect** competition to NOT count submission failures against our daily quota.\n\nWhy was it a perfect chance - because a private leader board that has yet to come into existence is impossible to probe. If anyone can probe the future - please message me now - I am interested in betting on the baseball games that will be played in August.", "1391170": "Yeah it sucks, and I think for this competition especially having the lack of error messages is detrimental, as many people will have quite complex pipelines. But obviously it is a challenge to give useful information without potential for probing. \n\nMy biggest frustration is the fact that the public test set (that it runs on during commit) is so small. It's something like 5 days? And incidentally these 5 days seem to be quite nice and don't contain a lot of the trip ups that have caused submission errors. I think a bigger public test would potentially alleviate a lot of the issues people are facing.\n\nI would advise you to feed in your test pipeline slowly. I found it easier to do a join and submission at a time and build it up. That way it was easier to determine which steps in the pipeline caused the issues.", "1391619": "Again - for this competition probing has almost ZERO value since the final private LB data has yet to be written and the games have yet to be played. \n\nWhat worries me a lot is what issues will come with the August data and the final submissions :(\n\nAgree - build up slow cause failures with no error codes are a real beast to figure out." }, "source": "meta" }