\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F5eadf1735b43aac7d39bd51d3ea65756%2Fevent_41185938.gif?generation=1676838679647882&alt=media\")
Inspired by the animations of @pestipeti (1128590, 1799490, 2351046 events) and @rhysie (1128590, 1583, 138523 events), the ability to control the start time, speed and latency of the detectors seems illuminating. The animation can search and load any event in the first 200 batches. One of the bigger events, 41185938 is fun to watch. Animation available here.
\n
Inspired by the animations of @pestipeti (1128590, 1799490, 2351046 events) and @rhysie (1128590, 1583, 138523 events), the ability to control the start time, speed and latency of the detectors seems illuminating. The animation can search and load any event in the first 200 batches. One of the bigger events, 41185938 is fun to watch. Animation available here.
\n
Having looked at a number of events, some observations have come to mind. The first is a combination of number of hits and time frame. Some events have very few hits and should be ignored. All events seem to have a time frame from about 5000 nanoseconds to about 25000 nanoseconds. Light travels at 0.3 m/nsec, but slower in ice, so say even at 0.2 m/nsec (for ease of calculation), it takes about 85 nsecs to get the 17 meters from one vertical sensor to the next. It takes about 5000 nsecs to go about 1000 meters from top to bottom or side to side of the whole structure.
\nWith this in mind, most events suddenly burst at about 9500 nsecs, and by 14500 nsecs are devolving into chaos. This picture from here, shows the expected explosion of light from a colliding neutrino.![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2Ff15abebed670aac0fff7f75e31259356%2Ficecube_hit.jpg?generation=1676955530443610&alt=media\")
The second observation on classification comes down to what is a neutrino and what is not a neutrino. This event #509612375 is clearly not a neutrino. Notice it enters one side with a nice cone of radiation and leaves the other side over about 5000 nanoseconds. This is a charged particle (electron/muon/tau) flying through the detector that may or may not have anything to do with a neutrino. Animation available here.![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F0a0ddfd0fdf99900db06e390ad97c547%2Fevent_509612375.gif?generation=1676956257096545&alt=media\")
As a follow-up to what is not a neutrino, this event #333339210 clearly IS A NEUTRINO. 🙂
\nIt explodes from inside the detector and spreads out in all directions. What is important, is that the spreading light is spreading a little faster in one direction over any other direction. This is the direction the neutrino came from. Animation available here.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F7aae4ef747964f42d82a331d8d64040a%2Fevent_333339210.gif?generation=1676956714333381&alt=media\")
What a nice visualization tool!! 💯😃
", "rawMarkdown": "What a nice visualization tool!! 💯😃", "votes": null }, { "id": "2153738", "postDate": "02/21/2023 15:54:50", "content": "This example leads to a tricky question.
\nIn the shared video introduction it was stated that every event in our train data has a neutrino. This charged particle probably the result of a neutrino collision outside of the detector.
\nSo - is this the promised neutrino or is it an example of noise that happens in Icecube data and we need to find the real neutrino???
\nWe can answer that question by looking at the True vector vs almost any of the math or AI predictions - this is certainly any easy event to compute IF it's the True neutrino.
", "rawMarkdown": "This example leads to a tricky question. \n\nIn the shared video introduction it was stated that every event in our train data has a neutrino. This charged particle probably the result of a neutrino collision outside of the detector. \n\nSo - is this the promised neutrino or is it an example of noise that happens in Icecube data and we need to find the real neutrino??? \n\nWe can answer that question by looking at the True vector vs almost any of the math or AI predictions - this is certainly any easy event to compute IF it's the True neutrino.", "votes": null }, { "id": "2153937", "postDate": "02/21/2023 18:08:52", "content": "I agree. The original idea was to mark all events as IS or ISNOT a neutrino event and use it as an additional training variable. But.. if I understand you, the question is, does the contest \"True\" vector assume all events are neutrino events and mark you based on that assumption?
", "rawMarkdown": "I agree. The original idea was to mark all events as IS or ISNOT a neutrino event and use it as an additional training variable. But.. if I understand you, the question is, does the contest \"True\" vector assume all events are neutrino events and mark you based on that assumption?", "votes": null }, { "id": "2153938", "postDate": "02/21/2023 18:10:06", "content": "Thanks, its fun to do 🙂
", "rawMarkdown": "Thanks, its fun to do 🙂", "votes": null }, { "id": "2154369", "postDate": "02/22/2023 01:02:27", "content": "Using one of math shared kernels you can generate a file with the True azimuth and zenith along with the predicted for every event. I also put the score for each event in that file. The math kernel will surely do a good job of finding the angles for this example. If the prediction is good score than neutrino collision outside of the ice cube was the simulated event. If huge error, than this type is noise.
\nAll events should have a neutrino if I understood the discussion in the video correctly. So IS or ISNOT probably not the good feature. I plan to use one of the math models to generate a \"MATH_WORKS' OR 'MATH_NG' feature for each event.
", "rawMarkdown": "Using one of math shared kernels you can generate a file with the True azimuth and zenith along with the predicted for every event. I also put the score for each event in that file. The math kernel will surely do a good job of finding the angles for this example. If the prediction is good score than neutrino collision outside of the ice cube was the simulated event. If huge error, than this type is noise.\n\nAll events should have a neutrino if I understood the discussion in the video correctly. So IS or ISNOT probably not the good feature. I plan to use one of the math models to generate a \"MATH_WORKS' OR 'MATH_NG' feature for each event.", "votes": null }, { "id": "2155712", "postDate": "02/22/2023 20:39:06", "content": "… or maybe 2 math models? one for a passing particle and a second for a hit within the detector that looks more like an explosion? I ran the 3 test events out of interest.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F8b7038996460b93d460a5f80b9d31d11%2Fevent_2092.gif?generation=1677097958130777&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F042efbc483a6a77ca7d6c6868a26f074%2Fevent_7344.gif?generation=1677098009951978&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F74ce9af3d4a7840a8e8517470a5ddf23%2Fevent_9482.gif?generation=1677098053539418&alt=media\")
Brilliant work!
\nCould you add a checkbox to show all or only auxiliary==False pulses?
\nIt would be great.
Thank you.
", "rawMarkdown": "Brilliant work!\n\nCould you add a checkbox to show all or only auxiliary==False pulses?\nIt would be great.\n\nThank you.", "votes": null }, { "id": "2166193", "postDate": "03/02/2023 17:11:28", "content": "Thanks and done. Click for animation
\nAlso, having a second look at the under 150 hit events. There are so many of them and there are some patterns there …
Thank you very much.
\nVery useful tool!
\nProbably still missing a marker (beginning or end) on the direction line of the track. But it is nothing.
P.S. Now the event 2325 looks even more discouraging. :)
", "rawMarkdown": "Thank you very much.\nVery useful tool!\nProbably still missing a marker (beginning or end) on the direction line of the track. But it is nothing.\n\nP.S. Now the event 2325 looks even more discouraging. :)", "votes": null }, { "id": "2166656", "postDate": "03/02/2023 22:49:09", "content": "Switched to auxiliary=false, much cleaner. Wow, hard to figure 2325. There is a muon passing through from top to lower right. That looks like all to me. But if there really is a neutrino hit in there, maybe a small group of about 3 sensors slightly out of line???![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F67794%2F9f74c0fcb517044e9cb0314b262e8e88%2Fevent_2325.gif?generation=1677796378563907&alt=media\")
Having looked at a number of events, some observations have come to mind. The first is a combination of number of hits and time frame. Some events have very few hits and should be ignored. All events seem to have a time frame from about 5000 nanoseconds to about 25000 nanoseconds. Light travels at 0.3 m/nsec, but slower in ice, so say even at 0.2 m/nsec (for ease of calculation), it takes about 85 nsecs to get the 17 meters from one vertical sensor to the next. It takes about 5000 nsecs to go about 1000 meters from top to bottom or side to side of the whole structure.
\nWith this in mind, most events suddenly burst at about 9500 nsecs, and by 14500 nsecs are devolving into chaos. This picture from here, shows the expected explosion of light from a colliding neutrino.
The second observation on classification comes down to what is a neutrino and what is not a neutrino. This event #509612375 is clearly not a neutrino. Notice it enters one side with a nice cone of radiation and leaves the other side over about 5000 nanoseconds. This is a charged particle (electron/muon/tau) flying through the detector that may or may not have anything to do with a neutrino. Animation available here.
This example leads to a tricky question.
\nIn the shared video introduction it was stated that every event in our train data has a neutrino. This charged particle probably the result of a neutrino collision outside of the detector.
\nSo - is this the promised neutrino or is it an example of noise that happens in Icecube data and we need to find the real neutrino???
\nWe can answer that question by looking at the True vector vs almost any of the math or AI predictions - this is certainly any easy event to compute IF it's the True neutrino.
", "votes": null, "replies": [ { "id": 2153937, "author_name": "edguy99", "author_url": "", "post_date": "02/21/2023 18:08:52", "content": "I agree. The original idea was to mark all events as IS or ISNOT a neutrino event and use it as an additional training variable. But.. if I understand you, the question is, does the contest \"True\" vector assume all events are neutrino events and mark you based on that assumption?
", "votes": null, "replies": [ { "id": 2154369, "author_name": "pcjimmmy", "author_url": "", "post_date": "02/22/2023 01:02:27", "content": "Using one of math shared kernels you can generate a file with the True azimuth and zenith along with the predicted for every event. I also put the score for each event in that file. The math kernel will surely do a good job of finding the angles for this example. If the prediction is good score than neutrino collision outside of the ice cube was the simulated event. If huge error, than this type is noise.
\nAll events should have a neutrino if I understood the discussion in the video correctly. So IS or ISNOT probably not the good feature. I plan to use one of the math models to generate a \"MATH_WORKS' OR 'MATH_NG' feature for each event.
", "votes": null, "replies": [ { "id": 2155712, "author_name": "edguy99", "author_url": "", "post_date": "02/22/2023 20:39:06", "content": "… or maybe 2 math models? one for a passing particle and a second for a hit within the detector that looks more like an explosion? I ran the 3 test events out of interest.
\nAs a follow-up to what is not a neutrino, this event #333339210 clearly IS A NEUTRINO. 🙂
\nIt explodes from inside the detector and spreads out in all directions. What is important, is that the spreading light is spreading a little faster in one direction over any other direction. This is the direction the neutrino came from. Animation available here.
\n
What a nice visualization tool!! 💯😃
", "votes": null, "replies": [ { "id": 2153938, "author_name": "edguy99", "author_url": "", "post_date": "02/21/2023 18:10:06", "content": "Thanks, its fun to do 🙂
", "votes": null, "replies": [] } ] }, { "id": 2163102, "author_name": "synset", "author_url": "", "post_date": "02/28/2023 15:30:54", "content": "Brilliant work!
\nCould you add a checkbox to show all or only auxiliary==False pulses?
\nIt would be great.
Thank you.
", "votes": null, "replies": [ { "id": 2166193, "author_name": "edguy99", "author_url": "", "post_date": "03/02/2023 17:11:28", "content": "Thanks and done. Click for animation
\nAlso, having a second look at the under 150 hit events. There are so many of them and there are some patterns there …
Thank you very much.
\nVery useful tool!
\nProbably still missing a marker (beginning or end) on the direction line of the track. But it is nothing.
P.S. Now the event 2325 looks even more discouraging. :)
", "votes": null, "replies": [ { "id": 2166656, "author_name": "edguy99", "author_url": "", "post_date": "03/02/2023 22:49:09", "content": "Switched to auxiliary=false, much cleaner. Wow, hard to figure 2325. There is a muon passing through from top to lower right. That looks like all to me. But if there really is a neutrino hit in there, maybe a small group of about 3 sensors slightly out of line???