Could somebody point to the calculation of the weight for each hit?
", "messages": [ { "id": "344683", "postDate": "06/18/2018 14:10:30", "content": "Could somebody point to the calculation of the weight for each hit?
", "rawMarkdown": "Could somebody point to the calculation of the weight for each hit?", "votes": null }, { "id": "344913", "postDate": "06/18/2018 22:44:55", "content": "If you look at the attached pdf document (taken from this post):
\n\n\"\"\"\nThe weights are incentives to get physically meaningful reconstructions, along two directions: the weight of a point is the product of two independent quantities weight order and weight pt.
\n\nweight order: The points at the beginning of the track, close to the collision, and at the end, are more important than the ones in the middle. The weights reflect this hierarchy and are normalized to sum to 1.
Weight pt: The high energy particles (large transverse momentum pT ) are the most interesting ones. As the bulk of the tracks have low pT , we have to explicitly favor high pT . weight pt is 0.2 if pT < 0.5GeV and 1. for pT > 3GeV, with a linear interpolation in between. Note that the lower the pT , the larger the geometrical curvature; at large pT tracks appear as straight lines.
Particles which generates 3 hits (points) or less are considered spurious, the weights of the associated points are set to zero. There are also random hits, which weight is also set to zero. The assignment of random hits to different tracks do not change the overall score, unless the matching particles lose the majority count.
As the weights disclose important information, they are provided along with the ground truth in the training data, but are kept hidden for the validation data.\n\"\"\"
\n\nNote that the total sum of weights in a given event is equal to 1.
", "rawMarkdown": "If you look at the attached pdf document (taken from this [post][1]):\n\n\"\"\"\nThe weights are incentives to get physically meaningful reconstructions, along two directions: the weight of a point is the product of two independent quantities weight order and weight pt.\n\n - **weight order:** The points at the beginning of the track, close to the collision, and at the end, are more important than the ones in the middle. The weights reflect this hierarchy and are normalized to sum to 1.\n\n - **Weight pt:** The high energy particles (large transverse momentum pT ) are the most interesting ones. As the bulk of the tracks have low pT , we have to explicitly favor high pT . weight pt is 0.2 if pT < 0.5GeV and 1. for pT > 3GeV, with a linear interpolation in between. Note that the lower the pT , the larger the geometrical curvature; at large pT tracks appear as straight lines.\n\n - Particles which generates 3 hits (points) or less are considered spurious, the weights of the associated points are set to zero. There are also random hits, which weight is also set to zero. The assignment of random hits to different tracks do not change the overall score, unless the matching particles lose the majority count.\n\nAs the weights disclose important information, they are provided along with the ground truth in the training data, but are kept hidden for the validation data.\n\"\"\"\n\nNote that the total sum of weights in a given event is equal to 1. \n\n [1]: https://www.kaggle.com/c/trackml-particle-identification/discussion/55708", "votes": null }, { "id": "345211", "postDate": "06/19/2018 12:10:29", "content": "I understand but still I would like to know how the hits weight is calculated. Thank you Osa
", "rawMarkdown": "I understand but still I would like to know how the hits weight is calculated. Thank you Osa", "votes": null }, { "id": "345385", "postDate": "06/19/2018 19:47:04", "content": "Hi, Osa's answer is rather complete (thanks). @agerom, what more would you need ?
", "rawMarkdown": "Hi, Osa's answer is rather complete (thanks). @agerom, what more would you need ?", "votes": null }, { "id": "346230", "postDate": "06/21/2018 10:27:19", "content": "I would like to know the details of how the weight order is calculated. In particular what does hierarchy reflect.
", "rawMarkdown": "I would like to know the details of how the weight order is calculated. In particular what does hierarchy reflect.", "votes": null }, { "id": "347138", "postDate": "06/23/2018 10:47:56", "content": "See https://www.kaggle.com/c/trackml-particle-identification/discussion/59527#latest-347136
", "rawMarkdown": "See https://www.kaggle.com/c/trackml-particle-identification/discussion/59527#latest-347136", "votes": null } ], "comments": [ { "id": 344913, "author_name": "osa111", "author_url": "", "post_date": "06/18/2018 22:44:55", "content": "If you look at the attached pdf document (taken from this post):
\n\n\"\"\"\nThe weights are incentives to get physically meaningful reconstructions, along two directions: the weight of a point is the product of two independent quantities weight order and weight pt.
\n\nweight order: The points at the beginning of the track, close to the collision, and at the end, are more important than the ones in the middle. The weights reflect this hierarchy and are normalized to sum to 1.
Weight pt: The high energy particles (large transverse momentum pT ) are the most interesting ones. As the bulk of the tracks have low pT , we have to explicitly favor high pT . weight pt is 0.2 if pT < 0.5GeV and 1. for pT > 3GeV, with a linear interpolation in between. Note that the lower the pT , the larger the geometrical curvature; at large pT tracks appear as straight lines.
Particles which generates 3 hits (points) or less are considered spurious, the weights of the associated points are set to zero. There are also random hits, which weight is also set to zero. The assignment of random hits to different tracks do not change the overall score, unless the matching particles lose the majority count.
As the weights disclose important information, they are provided along with the ground truth in the training data, but are kept hidden for the validation data.\n\"\"\"
\n\nNote that the total sum of weights in a given event is equal to 1.
", "votes": null, "replies": [ { "id": 345211, "author_name": "artemiosgeromitsos", "author_url": "", "post_date": "06/19/2018 12:10:29", "content": "I understand but still I would like to know how the hits weight is calculated. Thank you Osa
", "votes": null, "replies": [] }, { "id": 345385, "author_name": "droussea", "author_url": "", "post_date": "06/19/2018 19:47:04", "content": "Hi, Osa's answer is rather complete (thanks). @agerom, what more would you need ?
", "votes": null, "replies": [] }, { "id": 346230, "author_name": "artemiosgeromitsos", "author_url": "", "post_date": "06/21/2018 10:27:19", "content": "I would like to know the details of how the weight order is calculated. In particular what does hierarchy reflect.
", "votes": null, "replies": [] } ] }, { "id": 347138, "author_name": "cpmpml", "author_url": "", "post_date": "06/23/2018 10:47:56", "content": "See https://www.kaggle.com/c/trackml-particle-identification/discussion/59527#latest-347136
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "344683": "Could somebody point to the calculation of the weight for each hit?", "344913": "If you look at the attached pdf document (taken from this [post][1]):\n\n\"\"\"\nThe weights are incentives to get physically meaningful reconstructions, along two directions: the weight of a point is the product of two independent quantities weight order and weight pt.\n\n - **weight order:** The points at the beginning of the track, close to the collision, and at the end, are more important than the ones in the middle. The weights reflect this hierarchy and are normalized to sum to 1.\n\n - **Weight pt:** The high energy particles (large transverse momentum pT ) are the most interesting ones. As the bulk of the tracks have low pT , we have to explicitly favor high pT . weight pt is 0.2 if pT < 0.5GeV and 1. for pT > 3GeV, with a linear interpolation in between. Note that the lower the pT , the larger the geometrical curvature; at large pT tracks appear as straight lines.\n\n - Particles which generates 3 hits (points) or less are considered spurious, the weights of the associated points are set to zero. There are also random hits, which weight is also set to zero. The assignment of random hits to different tracks do not change the overall score, unless the matching particles lose the majority count.\n\nAs the weights disclose important information, they are provided along with the ground truth in the training data, but are kept hidden for the validation data.\n\"\"\"\n\nNote that the total sum of weights in a given event is equal to 1. \n\n [1]: https://www.kaggle.com/c/trackml-particle-identification/discussion/55708", "345211": "I understand but still I would like to know how the hits weight is calculated. Thank you Osa", "345385": "Hi, Osa's answer is rather complete (thanks). @agerom, what more would you need ?", "346230": "I would like to know the details of how the weight order is calculated. In particular what does hierarchy reflect.", "347138": "See https://www.kaggle.com/c/trackml-particle-identification/discussion/59527#latest-347136" }, "source": "meta" }