I am bit confused with the details provided in leaderboard section. It says \"This leaderboard is calculated with approximately 62% of the test data.\", however we are submitting results on only 10 test records.
\n\nAlso, for final submission, are we allowed to upload our locally trained weights and use inference script to generate the submission csv. In that case how the test data will be used to find the scores. These questions might have been already answered, I am new to Kaggle hence any help to understand this will be very much appreciated. ( Pls also share link if the answers are already given somewhere )
", "messages": [ { "id": "844942", "postDate": "05/13/2020 00:22:15", "content": "I am bit confused with the details provided in leaderboard section. It says \"This leaderboard is calculated with approximately 62% of the test data.\", however we are submitting results on only 10 test records.
\n\nAlso, for final submission, are we allowed to upload our locally trained weights and use inference script to generate the submission csv. In that case how the test data will be used to find the scores. These questions might have been already answered, I am new to Kaggle hence any help to understand this will be very much appreciated. ( Pls also share link if the answers are already given somewhere )
", "rawMarkdown": "I am bit confused with the details provided in leaderboard section. It says \"This leaderboard is calculated with approximately 62% of the test data.\", however we are submitting results on only 10 test records. \n\nAlso, for final submission, are we allowed to upload our locally trained weights and use inference script to generate the submission csv. In that case how the test data will be used to find the scores. These questions might have been already answered, I am new to Kaggle hence any help to understand this will be very much appreciated. ( Pls also share link if the answers are already given somewhere )", "votes": null }, { "id": "845058", "postDate": "05/13/2020 03:04:32", "content": "The 10 test records are neither public nor private test dataset. They are provided so you can debug your code. When you submit your code it runs on a new set of test records. Then 62% of those new test records are scored for public LB and 38% are scored for private LB
\n\nRegarding uploaded offline models. Yes you can train offline and upload models for inference only.
", "rawMarkdown": "The 10 test records are neither public nor private test dataset. They are provided so you can debug your code. When you submit your code it runs on a new set of test records. Then 62% of those new test records are scored for public LB and 38% are scored for private LB\n\nRegarding uploaded offline models. Yes you can train offline and upload models for inference only.", "votes": null }, { "id": "845199", "postDate": "05/13/2020 05:08:37", "content": "Hi Chris,\nReally appreciate for responding to the question.
\n\nCould you please let me know how the code finds the test records? I am using 'sample_submission.csv' and reading the test folder. Is there any specific naming standard we need to follow ?
", "rawMarkdown": "Hi Chris,\nReally appreciate for responding to the question.\n\nCould you please let me know how the code finds the test records? I am using 'sample_submission.csv' and reading the test folder. Is there any specific naming standard we need to follow ?", "votes": null } ], "comments": [ { "id": 845058, "author_name": "cdeotte", "author_url": "", "post_date": "05/13/2020 03:04:32", "content": "The 10 test records are neither public nor private test dataset. They are provided so you can debug your code. When you submit your code it runs on a new set of test records. Then 62% of those new test records are scored for public LB and 38% are scored for private LB
\n\nRegarding uploaded offline models. Yes you can train offline and upload models for inference only.
", "votes": null, "replies": [ { "id": 845199, "author_name": "abhisekjana", "author_url": "", "post_date": "05/13/2020 05:08:37", "content": "Hi Chris,\nReally appreciate for responding to the question.
\n\nCould you please let me know how the code finds the test records? I am using 'sample_submission.csv' and reading the test folder. Is there any specific naming standard we need to follow ?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "844942": "I am bit confused with the details provided in leaderboard section. It says \"This leaderboard is calculated with approximately 62% of the test data.\", however we are submitting results on only 10 test records. \n\nAlso, for final submission, are we allowed to upload our locally trained weights and use inference script to generate the submission csv. In that case how the test data will be used to find the scores. These questions might have been already answered, I am new to Kaggle hence any help to understand this will be very much appreciated. ( Pls also share link if the answers are already given somewhere )", "845058": "The 10 test records are neither public nor private test dataset. They are provided so you can debug your code. When you submit your code it runs on a new set of test records. Then 62% of those new test records are scored for public LB and 38% are scored for private LB\n\nRegarding uploaded offline models. Yes you can train offline and upload models for inference only.", "845199": "Hi Chris,\nReally appreciate for responding to the question.\n\nCould you please let me know how the code finds the test records? I am using 'sample_submission.csv' and reading the test folder. Is there any specific naming standard we need to follow ?" }, "source": "meta" }