With just less than a week of competition left, I wanted to give a reminder to do due diligence in selecting submissions! It is very easy to select a public notebook that gets a good score by fine-tuning a threshold for LB score. However, make sure you think through that the private dataset is to be slightly different than what we have so far. And it is with high resolution. Think through the effects it may have. Being able to beat or being beaten by public models can make a difference of hundreds of places so if placement is your concern be sure to make decisions with all of the information you have and not just LB score!
", "messages": [ { "id": 2627642, "postDate": "2024-01-30T19:38:44.183Z", "content": "With just less than a week of competition left, I wanted to give a reminder to do due diligence in selecting submissions! It is very easy to select a public notebook that gets a good score by fine-tuning a threshold for LB score. However, make sure you think through that the private dataset is to be slightly different than what we have so far. And it is with high resolution. Think through the effects it may have. Being able to beat or being beaten by public models can make a difference of hundreds of places so if placement is your concern be sure to make decisions with all of the information you have and not just LB score!
", "rawMarkdown": "With just less than a week of competition left, I wanted to give a reminder to do due diligence in selecting submissions! It is very easy to select a public notebook that gets a good score by fine-tuning a threshold for LB score. However, make sure you think through that the private dataset is to be slightly different than what we have so far. And it is with high resolution. Think through the effects it may have. Being able to beat or being beaten by public models can make a difference of hundreds of places so if placement is your concern be sure to make decisions with all of the information you have and not just LB score! ", "votes": 7 }, { "id": 2627715, "postDate": "2024-01-30T20:30:17.827Z", "content": "\"Private Test:
\nContinuous 3D part of a whole human kidney imaged with HiP-CT - Originally scanned at 15.77um/voxel binned to 63.08um/voxel (bin x4) before segmentation.\"
\nActually is a bit lower resolution, or I am not understanding it well? But I agree with public notebooks.
you people says \"trust your CV\", it actually means \"make your validation data trustable\"
\nno one would trust validation based on only one object sample, whether it is public test or local cv
", "rawMarkdown": "you people says \"trust your CV\", it actually means \"make your validation data trustable\"\n\nno one would trust validation based on **only one** object sample, whether it is public test or local cv", "votes": 1, "replies": [ { "id": 2628769, "postDate": "2024-01-31T13:42:52.943Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 2627694, "postDate": "2024-01-30T20:12:11.697Z", "content": "@cody11null this is a very important and pertinent point in my opinion.
\nOne needs to know that submissions submitted on the final day may not be the best ones. If one stays away from endowment bias, then he/ she is likely to make an informed decision in this regard!
\"Private Test:
\nContinuous 3D part of a whole human kidney imaged with HiP-CT - Originally scanned at 15.77um/voxel binned to 63.08um/voxel (bin x4) before segmentation.\"
\nActually is a bit lower resolution, or I am not understanding it well? But I agree with public notebooks.
you people says \"trust your CV\", it actually means \"make your validation data trustable\"
\nno one would trust validation based on only one object sample, whether it is public test or local cv
", "votes": 1, "replies": [ { "id": 2628769, "author_name": "", "author_url": "", "post_date": "2024-01-31T13:42:52.943000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 2627694, "author_name": "Ravi Ramakrishnan", "author_url": "", "post_date": "2024-01-30T20:12:11.697000", "content": "@cody11null this is a very important and pertinent point in my opinion.
\nOne needs to know that submissions submitted on the final day may not be the best ones. If one stays away from endowment bias, then he/ she is likely to make an informed decision in this regard!