The eval page says \"the horizontal distance (in meters) is computed between the predicted lat/lng and the ground truth lat/lng\". How is horizontal distance computed?
", "messages": [ { "id": "1304768", "postDate": "05/12/2021 21:19:20", "content": "The eval page says \"the horizontal distance (in meters) is computed between the predicted lat/lng and the ground truth lat/lng\". How is horizontal distance computed?
", "rawMarkdown": "The eval page says \"the horizontal distance (in meters) is computed between the predicted lat/lng and the ground truth lat/lng\". How is horizontal distance computed?", "votes": null }, { "id": "1304775", "postDate": "05/12/2021 21:29:29", "content": "I'd have to double check, but I believe it's the great circle distance. All plausible methods give very similar results for gaps this small anyway.
", "rawMarkdown": "I'd have to double check, but I believe it's the [great circle distance](https://en.wikipedia.org/wiki/Great-circle_distance). All plausible methods give very similar results for gaps this small anyway.", "votes": null }, { "id": "1304777", "postDate": "05/12/2021 21:36:59", "content": "Cool. It might be helpful to know which formula is used for great circle and what constants are used in the formula. Like you say, the difference might not affect the score.
", "rawMarkdown": "Cool. It might be helpful to know which formula is used for great circle and what constants are used in the formula. Like you say, the difference might not affect the score.", "votes": null }, { "id": "1304786", "postDate": "05/12/2021 21:57:35", "content": "@jpmiller The official metric uses the haversine method of computing the great circle distance, with RadiusEarthInMeters = 6367000
I can dig into that if we get to a point where it can be shown to matter, but I don't think we're there yet.
", "rawMarkdown": "I can dig into that if we get to a point where it can be shown to matter, but I don't think we're there yet.", "votes": null }, { "id": "1306042", "postDate": "05/13/2021 15:34:29", "content": "It does not appear to matter, either for the competition or in absolute terms. I compared two points where the lat and long differ at the 5th decimal point, which is typical for the baseline values.
\n(37.42828, -122.07254), (37.42827, -122.07253)
Haversine as described above shows a distance of 1.41902m. Vincenty, which is considered to be more accurate, shows 1.41957m. So, sub-mm differences only in this range.
Thanks for looking into it!
", "rawMarkdown": "It does not appear to matter, either for the competition or in absolute terms. I compared two points where the lat and long differ at the 5th decimal point, which is typical for the baseline values.\n`(37.42828, -122.07254), (37.42827, -122.07253)`\n\nHaversine as described above shows a distance of `1.41902m`. Vincenty, which is considered to be more accurate, shows `1.41957m`. So, sub-mm differences only in this range.\n\nThanks for looking into it!", "votes": null }, { "id": "1306562", "postDate": "05/13/2021 21:59:01", "content": "@jpmiller thanks for reporting this:)
", "rawMarkdown": "jpmiller thanks for reporting this:)", "votes": null } ], "comments": [ { "id": 1304775, "author_name": "sohier", "author_url": "", "post_date": "05/12/2021 21:29:29", "content": "I'd have to double check, but I believe it's the great circle distance. All plausible methods give very similar results for gaps this small anyway.
", "votes": null, "replies": [ { "id": 1304777, "author_name": "jpmiller", "author_url": "", "post_date": "05/12/2021 21:36:59", "content": "Cool. It might be helpful to know which formula is used for great circle and what constants are used in the formula. Like you say, the difference might not affect the score.
", "votes": null, "replies": [] }, { "id": 1304786, "author_name": "wcukierski", "author_url": "", "post_date": "05/12/2021 21:57:35", "content": "@jpmiller The official metric uses the haversine method of computing the great circle distance, with RadiusEarthInMeters = 6367000
I can dig into that if we get to a point where it can be shown to matter, but I don't think we're there yet.
", "votes": null, "replies": [] }, { "id": 1306042, "author_name": "jpmiller", "author_url": "", "post_date": "05/13/2021 15:34:29", "content": "It does not appear to matter, either for the competition or in absolute terms. I compared two points where the lat and long differ at the 5th decimal point, which is typical for the baseline values.
\n(37.42828, -122.07254), (37.42827, -122.07253)
Haversine as described above shows a distance of 1.41902m. Vincenty, which is considered to be more accurate, shows 1.41957m. So, sub-mm differences only in this range.
Thanks for looking into it!
", "votes": null, "replies": [] }, { "id": 1306562, "author_name": "emaerthin", "author_url": "", "post_date": "05/13/2021 21:59:01", "content": "@jpmiller thanks for reporting this:)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1304768": "The eval page says \"the horizontal distance (in meters) is computed between the predicted lat/lng and the ground truth lat/lng\". How is horizontal distance computed?", "1304775": "I'd have to double check, but I believe it's the [great circle distance](https://en.wikipedia.org/wiki/Great-circle_distance). All plausible methods give very similar results for gaps this small anyway.", "1304777": "Cool. It might be helpful to know which formula is used for great circle and what constants are used in the formula. Like you say, the difference might not affect the score.", "1304786": "jpmiller The official metric uses the haversine method of computing the great circle distance, with `RadiusEarthInMeters = 6367000`", "1304788": "I can dig into that if we get to a point where it can be shown to matter, but I don't think we're there yet.", "1306042": "It does not appear to matter, either for the competition or in absolute terms. I compared two points where the lat and long differ at the 5th decimal point, which is typical for the baseline values.\n`(37.42828, -122.07254), (37.42827, -122.07253)`\n\nHaversine as described above shows a distance of `1.41902m`. Vincenty, which is considered to be more accurate, shows `1.41957m`. So, sub-mm differences only in this range.\n\nThanks for looking into it!", "1306562": "jpmiller thanks for reporting this:)" }, "source": "meta" }