What's the appropriate value for it when calculating CV? 0.1? Because voxel space is 10?
", "messages": [ { "id": 3108341, "postDate": "2025-01-27T16:32:32.077Z", "content": "For the score() function in the competition metric the distance_multiplier is used to adjust the particle radius (our allowed range of error). According to the competition we must be within 0.5 of the particle radius to count as a true positive so set this value to 0.5 when doing the distance multiplier. Make sure all your predicted points are converted back into the particle coordinates by multiplying your outputs (630x630x184) by 10 because we have a voxel space of 10 but use 0.5 in the score. All coordinates should already be converted to the voxel space before building the dataframe if that makes sense.
", "rawMarkdown": "For the score() function in the competition metric the distance_multiplier is used to adjust the particle radius (our allowed range of error). According to the competition we must be within 0.5 of the particle radius to count as a true positive so set this value to 0.5 when doing the distance multiplier. Make sure all your predicted points are converted back into the particle coordinates by multiplying your outputs (630x630x184) by 10 because we have a voxel space of 10 but use 0.5 in the score. All coordinates should already be converted to the voxel space before building the dataframe if that makes sense.", "votes": 3, "replies": [ { "id": 3108938, "postDate": "2025-01-28T08:46:50.667Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3107929, "postDate": "2025-01-27T06:01:02.313Z", "content": "You mean the scaling from pixels to coordinates? This has a pretty good discussion of that:
\nhttps://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/556057
Thanks, man.
", "rawMarkdown": "Thanks, man.", "votes": 1 } ] }, { "id": 3108490, "postDate": "2025-01-27T17:57:20.113Z", "content": "Does 0.1 seem like the right choice for the distance_multiplier since the voxel space is 10? Curious to know if there's a standard approach for setting this in the competition metric or if it depends on the specific dataset characteristics.
", "rawMarkdown": "Does 0.1 seem like the right choice for the distance_multiplier since the voxel space is 10? Curious to know if there's a standard approach for setting this in the competition metric or if it depends on the specific dataset characteristics." }, { "id": 3107926, "postDate": "2025-01-27T05:52:55.490Z", "content": "What's the appropriate value for it when calculating CV? 0.1? Because voxel space is 10?
", "rawMarkdown": "What's the appropriate value for it when calculating CV? 0.1? Because voxel space is 10?" } ], "comments": [ { "id": 3108341, "author_name": "Connor", "author_url": "", "post_date": "2025-01-27T16:32:32.077000", "content": "For the score() function in the competition metric the distance_multiplier is used to adjust the particle radius (our allowed range of error). According to the competition we must be within 0.5 of the particle radius to count as a true positive so set this value to 0.5 when doing the distance multiplier. Make sure all your predicted points are converted back into the particle coordinates by multiplying your outputs (630x630x184) by 10 because we have a voxel space of 10 but use 0.5 in the score. All coordinates should already be converted to the voxel space before building the dataframe if that makes sense.
", "votes": 3, "replies": [ { "id": 3108938, "author_name": "", "author_url": "", "post_date": "2025-01-28T08:46:50.667000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3107929, "author_name": "David List", "author_url": "", "post_date": "2025-01-27T06:01:02.313000", "content": "You mean the scaling from pixels to coordinates? This has a pretty good discussion of that:
\nhttps://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/556057
Thanks, man.
", "votes": 1, "replies": [] } ] }, { "id": 3108490, "author_name": "", "author_url": "", "post_date": "2025-01-27T17:57:20.113000", "content": "Does 0.1 seem like the right choice for the distance_multiplier since the voxel space is 10? Curious to know if there's a standard approach for setting this in the competition metric or if it depends on the specific dataset characteristics.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3108341": "For the score() function in the competition metric the distance_multiplier is used to adjust the particle radius (our allowed range of error). According to the competition we must be within 0.5 of the particle radius to count as a true positive so set this value to 0.5 when doing the distance multiplier. Make sure all your predicted points are converted back into the particle coordinates by multiplying your outputs (630x630x184) by 10 because we have a voxel space of 10 but use 0.5 in the score. All coordinates should already be converted to the voxel space before building the dataframe if that makes sense.", "3107929": "You mean the scaling from pixels to coordinates? This has a pretty good discussion of that:\n[https://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/556057](https://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/556057)", "3108490": "Does 0.1 seem like the right choice for the distance_multiplier since the voxel space is 10? Curious to know if there's a standard approach for setting this in the competition metric or if it depends on the specific dataset characteristics.", "3107926": "What's the appropriate value for it when calculating CV? 0.1? Because voxel space is 10?" } }