Hey guys!
\nI remember I played this game like once a decade ago. I can still vividly feel how difficult it was to turn once you started gaining speed. It DRASTICALLY reduce your ability to turn/change direction. The equations for that are posted somewhere and shouldnt be too hard if someone wants precise info but I was thinking to vectorize the XYZ velocities and maybe log/exp rank them, it is 100% not linear.
I also have a somewhat unrelated Q: I've been working trying to transform the players/ball xyz positions into categorical features based on binning them and feature crossing all three. My idea here is it gives a 3D map of blocks that might be better to represent the areas that are important for scoring/defending by using possibly a transformer model or something. I think the number of 'blocks' are too large though ~344 000. Do you guys have any thoughts on this? Did you transform the coordinates at all?
", "messages": [ { "id": "1988194", "postDate": "10/15/2022 06:08:40", "content": "Hey guys!
\nI remember I played this game like once a decade ago. I can still vividly feel how difficult it was to turn once you started gaining speed. It DRASTICALLY reduce your ability to turn/change direction. The equations for that are posted somewhere and shouldnt be too hard if someone wants precise info but I was thinking to vectorize the XYZ velocities and maybe log/exp rank them, it is 100% not linear.
I also have a somewhat unrelated Q: I've been working trying to transform the players/ball xyz positions into categorical features based on binning them and feature crossing all three. My idea here is it gives a 3D map of blocks that might be better to represent the areas that are important for scoring/defending by using possibly a transformer model or something. I think the number of 'blocks' are too large though ~344 000. Do you guys have any thoughts on this? Did you transform the coordinates at all?
", "rawMarkdown": "Hey guys!\nI remember I played this game like once a decade ago. I can still vividly feel how difficult it was to turn once you started gaining speed. It DRASTICALLY reduce your ability to turn/change direction. The equations for that are posted somewhere and shouldnt be too hard if someone wants precise info but I was thinking to vectorize the XYZ velocities and maybe log/exp rank them, it is 100% not linear.\n\nI also have a somewhat unrelated Q: I've been working trying to transform the players/ball xyz positions into categorical features based on binning them and feature crossing all three. My idea here is it gives a 3D map of blocks that might be better to represent the areas that are important for scoring/defending by using possibly a transformer model or something. I think the number of 'blocks' are too large though ~344 000. Do you guys have any thoughts on this? Did you transform the coordinates at all?", "votes": null }, { "id": "1989370", "postDate": "10/15/2022 21:26:45", "content": "Hi, There are many discussion about feature engineering in this challenge.
\nI'm really interested in the non linearities you wrote about velocities.
\nI haven't played the game but from gameplays I can see that if you are going fast a direction ( espetially when jumping it gets practically impossible to change trajectory and if you try to change direction you just rotate while moving forward with unchanged direction).
About the unrelated question I think that this discussion may give you some useful data and maybe this can also give you a first code to try out transforming positions.
", "rawMarkdown": "Hi, There are many discussion about feature engineering in this challenge.\nI'm really interested in the non linearities you wrote about velocities.\nI haven't played the game but from gameplays I can see that if you are going fast a direction ( espetially when jumping it gets practically impossible to change trajectory and if you try to change direction you just rotate while moving forward with unchanged direction).\n\nAbout the unrelated question I think that [this discussion](https://www.kaggle.com/competitions/tabular-playground-series-oct-2022/discussion/358786) may give you some useful data and maybe this can also give you a first code to try out transforming positions.", "votes": null } ], "comments": [ { "id": 1989370, "author_name": "pietromaldini1", "author_url": "", "post_date": "10/15/2022 21:26:45", "content": "Hi, There are many discussion about feature engineering in this challenge.
\nI'm really interested in the non linearities you wrote about velocities.
\nI haven't played the game but from gameplays I can see that if you are going fast a direction ( espetially when jumping it gets practically impossible to change trajectory and if you try to change direction you just rotate while moving forward with unchanged direction).
About the unrelated question I think that this discussion may give you some useful data and maybe this can also give you a first code to try out transforming positions.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1988194": "Hey guys!\nI remember I played this game like once a decade ago. I can still vividly feel how difficult it was to turn once you started gaining speed. It DRASTICALLY reduce your ability to turn/change direction. The equations for that are posted somewhere and shouldnt be too hard if someone wants precise info but I was thinking to vectorize the XYZ velocities and maybe log/exp rank them, it is 100% not linear.\n\nI also have a somewhat unrelated Q: I've been working trying to transform the players/ball xyz positions into categorical features based on binning them and feature crossing all three. My idea here is it gives a 3D map of blocks that might be better to represent the areas that are important for scoring/defending by using possibly a transformer model or something. I think the number of 'blocks' are too large though ~344 000. Do you guys have any thoughts on this? Did you transform the coordinates at all?", "1989370": "Hi, There are many discussion about feature engineering in this challenge.\nI'm really interested in the non linearities you wrote about velocities.\nI haven't played the game but from gameplays I can see that if you are going fast a direction ( espetially when jumping it gets practically impossible to change trajectory and if you try to change direction you just rotate while moving forward with unchanged direction).\n\nAbout the unrelated question I think that [this discussion](https://www.kaggle.com/competitions/tabular-playground-series-oct-2022/discussion/358786) may give you some useful data and maybe this can also give you a first code to try out transforming positions." }, "source": "meta" }