Hi,
\n\nThanks a lot to the organizers for putting out this interesting challenge.\nI am not sure if I am missing something here but I have a couple of questions about the data.
\n\nFrom the description, it seems to me that (x,y,z) in hits file and (tx,ty,tz) in the truths file should essentially be the same. Looking at the data, it seems that they are very close but not identical. What is the source of this discrepancy?
The momentum info of the hit (tpx,tpy,tpz) is present in the truth file but not in the hits file. Correct me if I am wrong, but is this info not available for \"unlabelled\" test hits? If yes, this should be helpful in differentiating trajectories.
Thanks in advance.
", "messages": [ { "id": "327790", "postDate": "05/12/2018 13:56:18", "content": "Hi,
\n\nThanks a lot to the organizers for putting out this interesting challenge.\nI am not sure if I am missing something here but I have a couple of questions about the data.
\n\nFrom the description, it seems to me that (x,y,z) in hits file and (tx,ty,tz) in the truths file should essentially be the same. Looking at the data, it seems that they are very close but not identical. What is the source of this discrepancy?
The momentum info of the hit (tpx,tpy,tpz) is present in the truth file but not in the hits file. Correct me if I am wrong, but is this info not available for \"unlabelled\" test hits? If yes, this should be helpful in differentiating trajectories.
Thanks in advance.
", "rawMarkdown": "Hi,\n\nThanks a lot to the organizers for putting out this interesting challenge.\nI am not sure if I am missing something here but I have a couple of questions about the data. \n\n1. From the description, it seems to me that (x,y,z) in hits file and (tx,ty,tz) in the truths file should essentially be the same. Looking at the data, it seems that they are very close but not identical. What is the source of this discrepancy?\n\n2. The momentum info of the hit (tpx,tpy,tpz) is present in the truth file but not in the hits file. Correct me if I am wrong, but is this info not available for \"unlabelled\" test hits? If yes, this should be helpful in differentiating trajectories. \n\nThanks in advance.", "votes": null }, { "id": "327815", "postDate": "05/12/2018 14:59:08", "content": "About the first question. \nFor all recorded hits, the hits file contains the so-called reconstructed positions. These are only an approximation of the real (x, y, z) value of the hit. This is due to the fact that a module is composed of multiple cells (sensors).
\nThere are two possible cases.
The (x, y, z) positions that are given in the training and test files are the reconstructed ones. The positions in the truth files are the exact ones, more precisely, the position that correspond to the intersection of the trajectory with the bisection plane of the module depth-wise.
\n\nMore on that, including figures, in section 3.5 of the summary documentation.
", "rawMarkdown": "About the first question. \nFor all recorded hits, the hits file contains the so-called reconstructed positions. These are only an approximation of the real (x, y, z) value of the hit. This is due to the fact that a module is composed of multiple cells (sensors). \nThere are two possible cases.\n\n - Only one cell of the module is traversed by the particle; then the\n center of the cell is taken as the measurement position. The\n distribution of the discrepancy between the real and measured\n position depends on the situation of the module in the detector. \n - More than one cell of the module is traversed. This happens when a\n particle hits a module tangentially enough. The measured position is\n reconstructed from all involved cells.\n\nThe (x, y, z) positions that are given in the training and test files are the reconstructed ones. The positions in the truth files are the exact ones, more precisely, the position that correspond to the intersection of the trajectory with the bisection plane of the module depth-wise.\n\nMore on that, including figures, in section 3.5 of the [summary documentation][1].\n\n\n [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/321278/9331/trackml-participant-document-particle-v1.0.pdf", "votes": null }, { "id": "327886", "postDate": "05/12/2018 20:51:05", "content": "Thanks a lot for that pointer. Was very helpful. Do you agree with the premise of the second question? I could not find a reason why there should not be at least an estimate of the momenta for the hits. And it seems to me that could be a useful feature for trajectory segmentation.
", "rawMarkdown": "Thanks a lot for that pointer. Was very helpful. Do you agree with the premise of the second question? I could not find a reason why there should not be at least an estimate of the momenta for the hits. And it seems to me that could be a useful feature for trajectory segmentation.", "votes": null }, { "id": "327909", "postDate": "05/12/2018 21:41:09", "content": "", "rawMarkdown": "", "votes": null }, { "id": "328236", "postDate": "05/13/2018 18:52:04", "content": "About the momentum information. The basic answer is that the hits file corresponds to what is actually measured in the detector, and that the momentum is not measured. If it were, the problem would be much easier...
", "rawMarkdown": "About the momentum information. The basic answer is that the hits file corresponds to what is actually measured in the detector, and that the momentum is not measured. If it were, the problem would be much easier...", "votes": null }, { "id": "328421", "postDate": "05/14/2018 09:36:16", "content": "In real world, position sensors don't give the value of momentum. The estimated position of hits is known thanks to cells position but it is not possible to know from which direction came the particle.
", "rawMarkdown": "In real world, position sensors don't give the value of momentum. The estimated position of hits is known thanks to cells position but it is not possible to know from which direction came the particle.", "votes": null }, { "id": "328782", "postDate": "05/15/2018 04:20:17", "content": "For detectors like the ones in this competition, you need tracks to estimate the momentum. When particles are at highly relativistic energies, the energy loss due to ionization is only very weakly dependent on the momentum, and the detectors are too thin for much showering to happen. You won't get much beyond the position information.
", "rawMarkdown": "For detectors like the ones in this competition, you need tracks to estimate the momentum. When particles are at highly relativistic energies, the energy loss due to ionization is only very weakly dependent on the momentum, and the detectors are too thin for much showering to happen. You won't get much beyond the position information.", "votes": null }, { "id": "329119", "postDate": "05/15/2018 18:57:38", "content": "Regarding the second question: current detectors can only measure position. The curvature of a particle trajectory is a measure of its momentum and multiple position measurements are required to be able to estimate a curve and thus the momentum.
", "rawMarkdown": "Regarding the second question: current detectors can only measure position. The curvature of a particle trajectory is a measure of its momentum and multiple position measurements are required to be able to estimate a curve and thus the momentum.", "votes": null }, { "id": "329219", "postDate": "05/16/2018 03:06:44", "content": "Thanks for the responses folks.
", "rawMarkdown": "Thanks for the responses folks.", "votes": null } ], "comments": [ { "id": 327815, "author_name": "cecilegermain", "author_url": "", "post_date": "05/12/2018 14:59:08", "content": "About the first question. \nFor all recorded hits, the hits file contains the so-called reconstructed positions. These are only an approximation of the real (x, y, z) value of the hit. This is due to the fact that a module is composed of multiple cells (sensors).
\nThere are two possible cases.
The (x, y, z) positions that are given in the training and test files are the reconstructed ones. The positions in the truth files are the exact ones, more precisely, the position that correspond to the intersection of the trajectory with the bisection plane of the module depth-wise.
\n\nMore on that, including figures, in section 3.5 of the summary documentation.
", "votes": null, "replies": [ { "id": 327886, "author_name": "hodoor", "author_url": "", "post_date": "05/12/2018 20:51:05", "content": "Thanks a lot for that pointer. Was very helpful. Do you agree with the premise of the second question? I could not find a reason why there should not be at least an estimate of the momenta for the hits. And it seems to me that could be a useful feature for trajectory segmentation.
", "votes": null, "replies": [] }, { "id": 328421, "author_name": "darkdoval", "author_url": "", "post_date": "05/14/2018 09:36:16", "content": "In real world, position sensors don't give the value of momentum. The estimated position of hits is known thanks to cells position but it is not possible to know from which direction came the particle.
", "votes": null, "replies": [] }, { "id": 328782, "author_name": "muonneutrino", "author_url": "", "post_date": "05/15/2018 04:20:17", "content": "For detectors like the ones in this competition, you need tracks to estimate the momentum. When particles are at highly relativistic energies, the energy loss due to ionization is only very weakly dependent on the momentum, and the detectors are too thin for much showering to happen. You won't get much beyond the position information.
", "votes": null, "replies": [] }, { "id": 329119, "author_name": "msmk00", "author_url": "", "post_date": "05/15/2018 18:57:38", "content": "Regarding the second question: current detectors can only measure position. The curvature of a particle trajectory is a measure of its momentum and multiple position measurements are required to be able to estimate a curve and thus the momentum.
", "votes": null, "replies": [] } ] }, { "id": 327909, "author_name": "aeevrubin", "author_url": "", "post_date": "05/12/2018 21:41:09", "content": "", "votes": null, "replies": [] }, { "id": 328236, "author_name": "cecilegermain", "author_url": "", "post_date": "05/13/2018 18:52:04", "content": "About the momentum information. The basic answer is that the hits file corresponds to what is actually measured in the detector, and that the momentum is not measured. If it were, the problem would be much easier...
", "votes": null, "replies": [] }, { "id": 329219, "author_name": "hodoor", "author_url": "", "post_date": "05/16/2018 03:06:44", "content": "Thanks for the responses folks.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "327790": "Hi,\n\nThanks a lot to the organizers for putting out this interesting challenge.\nI am not sure if I am missing something here but I have a couple of questions about the data. \n\n1. From the description, it seems to me that (x,y,z) in hits file and (tx,ty,tz) in the truths file should essentially be the same. Looking at the data, it seems that they are very close but not identical. What is the source of this discrepancy?\n\n2. The momentum info of the hit (tpx,tpy,tpz) is present in the truth file but not in the hits file. Correct me if I am wrong, but is this info not available for \"unlabelled\" test hits? If yes, this should be helpful in differentiating trajectories. \n\nThanks in advance.", "327815": "About the first question. \nFor all recorded hits, the hits file contains the so-called reconstructed positions. These are only an approximation of the real (x, y, z) value of the hit. This is due to the fact that a module is composed of multiple cells (sensors). \nThere are two possible cases.\n\n - Only one cell of the module is traversed by the particle; then the\n center of the cell is taken as the measurement position. The\n distribution of the discrepancy between the real and measured\n position depends on the situation of the module in the detector. \n - More than one cell of the module is traversed. This happens when a\n particle hits a module tangentially enough. The measured position is\n reconstructed from all involved cells.\n\nThe (x, y, z) positions that are given in the training and test files are the reconstructed ones. The positions in the truth files are the exact ones, more precisely, the position that correspond to the intersection of the trajectory with the bisection plane of the module depth-wise.\n\nMore on that, including figures, in section 3.5 of the [summary documentation][1].\n\n\n [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/321278/9331/trackml-participant-document-particle-v1.0.pdf", "327886": "Thanks a lot for that pointer. Was very helpful. Do you agree with the premise of the second question? I could not find a reason why there should not be at least an estimate of the momenta for the hits. And it seems to me that could be a useful feature for trajectory segmentation.", "327909": "", "328236": "About the momentum information. The basic answer is that the hits file corresponds to what is actually measured in the detector, and that the momentum is not measured. If it were, the problem would be much easier...", "328421": "In real world, position sensors don't give the value of momentum. The estimated position of hits is known thanks to cells position but it is not possible to know from which direction came the particle.", "328782": "For detectors like the ones in this competition, you need tracks to estimate the momentum. When particles are at highly relativistic energies, the energy loss due to ionization is only very weakly dependent on the momentum, and the detectors are too thin for much showering to happen. You won't get much beyond the position information.", "329119": "Regarding the second question: current detectors can only measure position. The curvature of a particle trajectory is a measure of its momentum and multiple position measurements are required to be able to estimate a curve and thus the momentum.", "329219": "Thanks for the responses folks." }, "source": "meta" }