![\"enter](\"https://storage.googleapis.com/kaggle-forum-message-attachments/354649/9830/trackML_9001.png\")
\nWow, I am 0.9 now. :)My event000001001 stuck at 0.89x, so I need to find a way to break through. Here is my solution:\n\nWow, I am 0.9 now. :)
My event000001001 stuck at 0.89x, so I need to find a way to break through. Here is my solution:\n\nWow, I am 0.9 now. :)
Nice work!
", "rawMarkdown": "Nice work!", "votes": null }, { "id": "354653", "postDate": "07/10/2018 01:57:39", "content": "There are may be some information in cells.cvs
", "rawMarkdown": "There are may be some information in cells.cvs", "votes": null }, { "id": "354677", "postDate": "07/10/2018 03:38:44", "content": "May I ask about your approach if you don't mind? You said you use deep learning, is it right? Regarding your merging steps, I believe it might be similar to this:
\n\n1) clustering long tracks (i.e., >12 hits) with small epsilon. This would be straight tracks with high confidence so it must be intact.\n2) clustering long tracks with higher epsilon without touching the clustered ones in step 1. These should be long curvy tracks.\n3) extend...\n4) do the similar, with shorter tracks...
\n\nIf I am right, may I ask is that your extend/outlier removal step using DL? Thanks so much.
", "rawMarkdown": "May I ask about your approach if you don't mind? You said you use deep learning, is it right? Regarding your merging steps, I believe it might be similar to this:\n\n1) clustering long tracks (i.e., >12 hits) with small epsilon. This would be straight tracks with high confidence so it must be intact.\n2) clustering long tracks with higher epsilon without touching the clustered ones in step 1. These should be long curvy tracks.\n3) extend...\n4) do the similar, with shorter tracks...\n\nIf I am right, may I ask is that your extend/outlier removal step using DL? Thanks so much.", "votes": null }, { "id": "354689", "postDate": "07/10/2018 04:41:42", "content": "The merging steps are similar and I think most of us are similar too. I use deep learning on track finding, and some extension methods are similar as finding stage in my approach. I have trained a model to do outlier removal on good tracks but failed, so I did not apply it yet. And I think every good outlier removal methods are also good for track finding, and it is better to do it on finding stage.
", "rawMarkdown": "The merging steps are similar and I think most of us are similar too. I use deep learning on track finding, and some extension methods are similar as finding stage in my approach. I have trained a model to do outlier removal on good tracks but failed, so I did not apply it yet. And I think every good outlier removal methods are also good for track finding, and it is better to do it on finding stage.", "votes": null }, { "id": "354729", "postDate": "07/10/2018 06:21:05", "content": "\n\n\nThe merging steps are similar and I think most of us are similar too.
\n
@outrunner, I do not think so. You use the worst one. My merging/extension method enables to connect pieces of tracks - full helices, even distorted, are not the problem if there are pieces of them in the soup of track candidates. So, do not call attaching outliers to the tracks of even number of hits as \"Grzegorz's extension\" ;-)
\n\nBut seriously, congratulations and best wishes.
", "rawMarkdown": ">The merging steps are similar and I think most of us are similar too. \n\n@outrunner, I do not think so. You use the worst one. My merging/extension method enables to connect pieces of tracks - full helices, even distorted, are not the problem if there are pieces of them in the soup of track candidates. So, do not call attaching outliers to the tracks of even number of hits as \"Grzegorz's extension\" ;-)\n\nBut seriously, congratulations and best wishes.", "votes": null }, { "id": "354732", "postDate": "07/10/2018 06:32:47", "content": "\n\n\nYou use the worst one.
\n
I like it, as I know @Heng may discover some of my weakness but didn't tell me, haha.
", "rawMarkdown": "> You use the worst one.\n\nI like it, as I know @Heng may discover some of my weakness but didn't tell me, haha.", "votes": null }, { "id": "354779", "postDate": "07/10/2018 08:32:40", "content": "@outrunner, Heng and many other Kagglers are optimists. They want to have the same chances to win as you, so they think it makes no sense to help you too much ;). I am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.
\n\nEDIT: emoticon added
", "rawMarkdown": "outrunner, Heng and many other Kagglers are optimists. They want to have the same chances to win as you, so they think it makes no sense to help you too much ;). I am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.\n\nEDIT: emoticon added", "votes": null }, { "id": "354796", "postDate": "07/10/2018 09:12:47", "content": "I think Heng is helping a lot of people, including outrunner.
", "rawMarkdown": "I think Heng is helping a lot of people, including outrunner.", "votes": null }, { "id": "354800", "postDate": "07/10/2018 09:16:16", "content": "\n\n\nI will be glad to see any real Kaggler in TOP10.
\n
You don't think outrunner will be in top 10?
", "rawMarkdown": "> I will be glad to see any real Kaggler in TOP10.\n\nYou don't think outrunner will be in top 10?", "votes": null }, { "id": "354816", "postDate": "07/10/2018 09:43:23", "content": "Thanks outrunner for your comment!
", "rawMarkdown": "Thanks outrunner for your comment!", "votes": null }, { "id": "354829", "postDate": "07/10/2018 10:26:07", "content": "\n\n\nYou don't think outrunner will be in top 10?
\n
@CPMP, Looking at the 3 events above, @outrunner's score is ~0.88, so his/her place at LB will depend on the number of registered crazytrackers (0.89) https://www.kaggle.com/c/trackml-particle-identification/discussion/60735#354465
", "rawMarkdown": ">You don't think outrunner will be in top 10?\n\n@CPMP, Looking at the 3 events above, @outrunner's score is ~0.88, so his/her place at LB will depend on the number of registered crazytrackers (0.89) https://www.kaggle.com/c/trackml-particle-identification/discussion/60735#354465", "votes": null }, { "id": "354831", "postDate": "07/10/2018 10:26:30", "content": "\n\n\nI think Heng is helping a lot of people, including outrunner.
\n
I am just kidding, and you are right.
\n\n\n\n\nTOP10 will be occupied by \"novices\"
\n
I think there are a lot of people doing research on this topic will join this competition.
\n\n\n\n\nso his/her place ...
\n
\"his\", Thanks.
", "rawMarkdown": "> I think Heng is helping a lot of people, including outrunner.\n\nI am just kidding, and you are right.\n\n> TOP10 will be occupied by \"novices\"\n\nI think there are a lot of people doing research on this topic will join this competition.\n\n> so his/her place ...\n\n\"his\", Thanks.", "votes": null }, { "id": "354867", "postDate": "07/10/2018 11:29:47", "content": "\n\n\nI think there are a lot of people doing research on this topic will join this competition.
\n
If so, short explanation:
\n\n\"novice\" - highly educated person who knows how a Kalman filter works
\n\n\"Everything but a Kalman filter\" - subtitle of the competition \"forgotten\" by the organizers; the name of the game we are playing
", "rawMarkdown": "> I think there are a lot of people doing research on this topic will join this competition.\n\nIf so, short explanation: \n\n\"novice\" - highly educated person who knows how a Kalman filter works\n\n\"Everything but a Kalman filter\" - subtitle of the competition \"forgotten\" by the organizers; the name of the game we are playing", "votes": null }, { "id": "354871", "postDate": "07/10/2018 11:43:53", "content": "@Grzegorz, a ML novice who registered Kaggle months ago here, I don't know how a Kalman filter works though. Just a question, may I ask why you stopped making submissions? working on something big?
", "rawMarkdown": "Grzegorz, a ML novice who registered Kaggle months ago here, I don't know how a Kalman filter works though. Just a question, may I ask why you stopped making submissions? working on something big?", "votes": null }, { "id": "354897", "postDate": "07/10/2018 12:36:21", "content": "@Nicole, not only me - train dataset enables validation good enough, so preparing full submission is a waste of time. Up to now, nobody wants to submit anything above 0.82 - let the others think, there is still a barier for (0,0,z). Nobody wants to submit his 0.90 for a moment of fame, because this will activate Kalman filter implementers. Frankly, I have created quite nice merging&extending method, but still do not have a good method for evaluating track candidates. That is the real reason.
", "rawMarkdown": "Nicole, not only me - train dataset enables validation good enough, so preparing full submission is a waste of time. Up to now, nobody wants to submit anything above 0.82 - let the others think, there is still a barier for (0,0,z). Nobody wants to submit his 0.90 for a moment of fame, because this will activate Kalman filter implementers. Frankly, I have created quite nice merging&extending method, but still do not have a good method for evaluating track candidates. That is the real reason.", "votes": null }, { "id": "354915", "postDate": "07/10/2018 13:23:06", "content": "@Nicole, here is a great intro to Kalman filters: http://www.bzarg.com/p/how-a-kalman-filter-works-in-pictures/
", "rawMarkdown": "Nicole, here is a great intro to Kalman filters: http://www.bzarg.com/p/how-a-kalman-filter-works-in-pictures/", "votes": null }, { "id": "355050", "postDate": "07/10/2018 20:17:47", "content": "Hi @Outrunner, which training file did you find event000008841 in ?
\n\nI can only read train_1.zip and train_sample.zip. The other files train_{2,3,4,5}.zip returns a null.
", "rawMarkdown": "Hi @Outrunner, which training file did you find event000008841 in ?\n\nI can only read train_1.zip and train_sample.zip. The other files train_{2,3,4,5}.zip returns a null.", "votes": null }, { "id": "355053", "postDate": "07/10/2018 20:25:43", "content": "That link by @CPMP is a really good one. I found it a few weeks ago and felt my former supervisor (i.e. research supervisor) DR. Bozic will be proud of this author :-)
", "rawMarkdown": "That link by @CPMP is a really good one. I found it a few weeks ago and felt my former supervisor (i.e. research supervisor) DR. Bozic will be proud of this author :-)", "votes": null }, { "id": "355059", "postDate": "07/10/2018 21:03:51", "content": "\n\n\nlet the others think, there is still a barier for (0,0,z)
\n
I think @outrunner has already demonstrated that he is long since \"free from origin\" ;-)
", "rawMarkdown": "> let the others think, there is still a barier for (0,0,z)\n\nI think @outrunner has already demonstrated that he is long since \"free from origin\" ;-)", "votes": null }, { "id": "355065", "postDate": "07/10/2018 21:22:47", "content": "@Edwin, It is not so obvious. A method fitted to one event should be used on another one without any changes. We do not know if it was done and what was the result, so for the world, your method/submission is still the best and still below 0.82.
", "rawMarkdown": "Edwin, It is not so obvious. A method fitted to one event should be used on another one without any changes. We do not know if it was done and what was the result, so for the world, your method/submission is still the best and still below 0.82.", "votes": null }, { "id": "355084", "postDate": "07/10/2018 22:19:04", "content": "@Grzegorz,
\n\n\n\n\nI am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle.
\n
I might fit your description (although I'm afraid it's no guarantee to remain in top 10), but there definitely is another one coming. @demelian's speed of approach is crazy.
", "rawMarkdown": "Grzegorz,\n\n> I am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle.\n\nI might fit your description (although I'm afraid it's no guarantee to remain in top 10), but there definitely is another one coming. @demelian's speed of approach is crazy.", "votes": null }, { "id": "355085", "postDate": "07/10/2018 22:25:03", "content": "(although I'm afraid it's no guarantee to remain in top 10),\n@Edwin, I think 0.81 or above is pretty safe (even for the prize), it's not like you're gonna sit there and do nothing for another month. And you're right, @demelian is making progress like crazy. If I use my human-learning linear regression approach, he'll jump to nr. 1 in no time. ;)
\n\nIncidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?
", "rawMarkdown": "(although I'm afraid it's no guarantee to remain in top 10),\n\n@Edwin, I think 0.81 or above is pretty safe (even for the prize), it's not like you're gonna sit there and do nothing for another month. And you're right, @demelian is making progress like crazy. If I use my human-learning linear regression approach, he'll jump to nr. 1 in no time. ;)\n\nIncidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?", "votes": null }, { "id": "355086", "postDate": "07/10/2018 22:37:38", "content": "@Edwin, All novices are warmly welcomed and invited to the game \"Everything but a Kalman filter\".
", "rawMarkdown": "Edwin, All novices are warmly welcomed and invited to the game \"Everything but a Kalman filter\".", "votes": null }, { "id": "355087", "postDate": "07/10/2018 22:57:22", "content": "\n\n\nIncidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?
\n
Judging by the forum posts I'm using a very different approach than @outrunner. That's about as much as I'm inclined to disclose at this point, I think.
", "rawMarkdown": "> Incidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?\n\nJudging by the forum posts I'm using a very different approach than @outrunner. That's about as much as I'm inclined to disclose at this point, I think.", "votes": null }, { "id": "355115", "postDate": "07/11/2018 01:24:44", "content": "4481? is in train_2.zip, you can download it.
", "rawMarkdown": "4481? is in train_2.zip, you can download it.", "votes": null }, { "id": "355121", "postDate": "07/11/2018 01:53:47", "content": "Thanks I can download it but it is empty as reported here https://www.kaggle.com/c/trackml-particle-identification/discussion/56787#352750
\n\nUPDATE
\n\nI have downloaded train_2.zip and observed the following:-
\n\nFor anyone else having the same issue, the reason kernels do not recognize the files in train_{2,3,4,5}.zip is not because they are empty. I believe there is some kind of invisible characters in the file names. Once I renamed them to the same names by retyping, they are readable. I had the same issue reading them locally until I did the renaming.
", "rawMarkdown": "Thanks I can download it but it is empty as reported here https://www.kaggle.com/c/trackml-particle-identification/discussion/56787#352750\n\n**UPDATE**\n\nI have downloaded train_2.zip and observed the following:-\n\nFor anyone else having the same issue, the reason kernels do not recognize the files in train_{2,3,4,5}.zip is not because they are empty. I believe there is some kind of invisible characters in the file names. Once I renamed them to the same names by retyping, they are readable. I had the same issue reading them locally until I did the renaming.", "votes": null }, { "id": "370293", "postDate": "08/14/2018 15:20:54", "content": "\n\n\nI am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.
\n
Grzegorz,
\n\nI was puzzled when you made that comment, which is why I revisit it now that we know the end result. You're right, few newcomers at the top or near it, but also 5 Kaggle masters or grandmasters in top 10. I guess these can count as \"real Kagglers\" ;)
", "rawMarkdown": "> I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.\n\nGrzegorz,\n\nI was puzzled when you made that comment, which is why I revisit it now that we know the end result. You're right, few newcomers at the top or near it, but also 5 Kaggle masters or grandmasters in top 10. I guess these can count as \"real Kagglers\" ;)", "votes": null } ], "comments": [ { "id": 354652, "author_name": "hengck23", "author_url": "", "post_date": "07/10/2018 01:57:12", "content": "Nice work!
", "votes": null, "replies": [] }, { "id": 354653, "author_name": "hengck23", "author_url": "", "post_date": "07/10/2018 01:57:39", "content": "There are may be some information in cells.cvs
", "votes": null, "replies": [] }, { "id": 354677, "author_name": "khahuras", "author_url": "", "post_date": "07/10/2018 03:38:44", "content": "May I ask about your approach if you don't mind? You said you use deep learning, is it right? Regarding your merging steps, I believe it might be similar to this:
\n\n1) clustering long tracks (i.e., >12 hits) with small epsilon. This would be straight tracks with high confidence so it must be intact.\n2) clustering long tracks with higher epsilon without touching the clustered ones in step 1. These should be long curvy tracks.\n3) extend...\n4) do the similar, with shorter tracks...
\n\nIf I am right, may I ask is that your extend/outlier removal step using DL? Thanks so much.
", "votes": null, "replies": [ { "id": 354689, "author_name": "outrunner", "author_url": "", "post_date": "07/10/2018 04:41:42", "content": "The merging steps are similar and I think most of us are similar too. I use deep learning on track finding, and some extension methods are similar as finding stage in my approach. I have trained a model to do outlier removal on good tracks but failed, so I did not apply it yet. And I think every good outlier removal methods are also good for track finding, and it is better to do it on finding stage.
", "votes": null, "replies": [] }, { "id": 354729, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 06:21:05", "content": "\n\n\nThe merging steps are similar and I think most of us are similar too.
\n
@outrunner, I do not think so. You use the worst one. My merging/extension method enables to connect pieces of tracks - full helices, even distorted, are not the problem if there are pieces of them in the soup of track candidates. So, do not call attaching outliers to the tracks of even number of hits as \"Grzegorz's extension\" ;-)
\n\nBut seriously, congratulations and best wishes.
", "votes": null, "replies": [] }, { "id": 354732, "author_name": "outrunner", "author_url": "", "post_date": "07/10/2018 06:32:47", "content": "\n\n\nYou use the worst one.
\n
I like it, as I know @Heng may discover some of my weakness but didn't tell me, haha.
", "votes": null, "replies": [] }, { "id": 354779, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 08:32:40", "content": "@outrunner, Heng and many other Kagglers are optimists. They want to have the same chances to win as you, so they think it makes no sense to help you too much ;). I am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.
\n\nEDIT: emoticon added
", "votes": null, "replies": [] }, { "id": 354796, "author_name": "cpmpml", "author_url": "", "post_date": "07/10/2018 09:12:47", "content": "I think Heng is helping a lot of people, including outrunner.
", "votes": null, "replies": [] }, { "id": 354800, "author_name": "cpmpml", "author_url": "", "post_date": "07/10/2018 09:16:16", "content": "\n\n\nI will be glad to see any real Kaggler in TOP10.
\n
You don't think outrunner will be in top 10?
", "votes": null, "replies": [] }, { "id": 354816, "author_name": "khahuras", "author_url": "", "post_date": "07/10/2018 09:43:23", "content": "Thanks outrunner for your comment!
", "votes": null, "replies": [] }, { "id": 354829, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 10:26:07", "content": "\n\n\nYou don't think outrunner will be in top 10?
\n
@CPMP, Looking at the 3 events above, @outrunner's score is ~0.88, so his/her place at LB will depend on the number of registered crazytrackers (0.89) https://www.kaggle.com/c/trackml-particle-identification/discussion/60735#354465
", "votes": null, "replies": [] }, { "id": 354831, "author_name": "outrunner", "author_url": "", "post_date": "07/10/2018 10:26:30", "content": "\n\n\nI think Heng is helping a lot of people, including outrunner.
\n
I am just kidding, and you are right.
\n\n\n\n\nTOP10 will be occupied by \"novices\"
\n
I think there are a lot of people doing research on this topic will join this competition.
\n\n\n\n\nso his/her place ...
\n
\"his\", Thanks.
", "votes": null, "replies": [] }, { "id": 354867, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 11:29:47", "content": "\n\n\nI think there are a lot of people doing research on this topic will join this competition.
\n
If so, short explanation:
\n\n\"novice\" - highly educated person who knows how a Kalman filter works
\n\n\"Everything but a Kalman filter\" - subtitle of the competition \"forgotten\" by the organizers; the name of the game we are playing
", "votes": null, "replies": [] }, { "id": 354871, "author_name": "nicolefinnie", "author_url": "", "post_date": "07/10/2018 11:43:53", "content": "@Grzegorz, a ML novice who registered Kaggle months ago here, I don't know how a Kalman filter works though. Just a question, may I ask why you stopped making submissions? working on something big?
", "votes": null, "replies": [] }, { "id": 354897, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 12:36:21", "content": "@Nicole, not only me - train dataset enables validation good enough, so preparing full submission is a waste of time. Up to now, nobody wants to submit anything above 0.82 - let the others think, there is still a barier for (0,0,z). Nobody wants to submit his 0.90 for a moment of fame, because this will activate Kalman filter implementers. Frankly, I have created quite nice merging&extending method, but still do not have a good method for evaluating track candidates. That is the real reason.
", "votes": null, "replies": [] }, { "id": 354915, "author_name": "cpmpml", "author_url": "", "post_date": "07/10/2018 13:23:06", "content": "@Nicole, here is a great intro to Kalman filters: http://www.bzarg.com/p/how-a-kalman-filter-works-in-pictures/
", "votes": null, "replies": [] }, { "id": 355053, "author_name": "sheriytm", "author_url": "", "post_date": "07/10/2018 20:25:43", "content": "That link by @CPMP is a really good one. I found it a few weeks ago and felt my former supervisor (i.e. research supervisor) DR. Bozic will be proud of this author :-)
", "votes": null, "replies": [] }, { "id": 355059, "author_name": "edwinst", "author_url": "", "post_date": "07/10/2018 21:03:51", "content": "\n\n\nlet the others think, there is still a barier for (0,0,z)
\n
I think @outrunner has already demonstrated that he is long since \"free from origin\" ;-)
", "votes": null, "replies": [] }, { "id": 355065, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 21:22:47", "content": "@Edwin, It is not so obvious. A method fitted to one event should be used on another one without any changes. We do not know if it was done and what was the result, so for the world, your method/submission is still the best and still below 0.82.
", "votes": null, "replies": [] }, { "id": 355084, "author_name": "edwinst", "author_url": "", "post_date": "07/10/2018 22:19:04", "content": "@Grzegorz,
\n\n\n\n\nI am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle.
\n
I might fit your description (although I'm afraid it's no guarantee to remain in top 10), but there definitely is another one coming. @demelian's speed of approach is crazy.
", "votes": null, "replies": [] }, { "id": 355085, "author_name": "nicolefinnie", "author_url": "", "post_date": "07/10/2018 22:25:03", "content": "(although I'm afraid it's no guarantee to remain in top 10),\n@Edwin, I think 0.81 or above is pretty safe (even for the prize), it's not like you're gonna sit there and do nothing for another month. And you're right, @demelian is making progress like crazy. If I use my human-learning linear regression approach, he'll jump to nr. 1 in no time. ;)
\n\nIncidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?
", "votes": null, "replies": [] }, { "id": 355086, "author_name": "sionek", "author_url": "", "post_date": "07/10/2018 22:37:38", "content": "@Edwin, All novices are warmly welcomed and invited to the game \"Everything but a Kalman filter\".
", "votes": null, "replies": [] }, { "id": 355087, "author_name": "edwinst", "author_url": "", "post_date": "07/10/2018 22:57:22", "content": "\n\n\nIncidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?
\n
Judging by the forum posts I'm using a very different approach than @outrunner. That's about as much as I'm inclined to disclose at this point, I think.
", "votes": null, "replies": [] }, { "id": 370293, "author_name": "cpmpml", "author_url": "", "post_date": "08/14/2018 15:20:54", "content": "\n\n\nI am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.
\n
Grzegorz,
\n\nI was puzzled when you made that comment, which is why I revisit it now that we know the end result. You're right, few newcomers at the top or near it, but also 5 Kaggle masters or grandmasters in top 10. I guess these can count as \"real Kagglers\" ;)
", "votes": null, "replies": [] } ] }, { "id": 355050, "author_name": "sheriytm", "author_url": "", "post_date": "07/10/2018 20:17:47", "content": "Hi @Outrunner, which training file did you find event000008841 in ?
\n\nI can only read train_1.zip and train_sample.zip. The other files train_{2,3,4,5}.zip returns a null.
", "votes": null, "replies": [ { "id": 355115, "author_name": "outrunner", "author_url": "", "post_date": "07/11/2018 01:24:44", "content": "4481? is in train_2.zip, you can download it.
", "votes": null, "replies": [] }, { "id": 355121, "author_name": "sheriytm", "author_url": "", "post_date": "07/11/2018 01:53:47", "content": "Thanks I can download it but it is empty as reported here https://www.kaggle.com/c/trackml-particle-identification/discussion/56787#352750
\n\nUPDATE
\n\nI have downloaded train_2.zip and observed the following:-
\n\nFor anyone else having the same issue, the reason kernels do not recognize the files in train_{2,3,4,5}.zip is not because they are empty. I believe there is some kind of invisible characters in the file names. Once I renamed them to the same names by retyping, they are readable. I had the same issue reading them locally until I did the renaming.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "354649": "My event000001001 stuck at 0.89x, so I need to find a way to break through. Here is my solution:\n![enter image description here][1]\nWow, I am 0.9 now. :)\n\n\n [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/354649/9830/trackML_9001.png", "354652": "Nice work!", "354653": "There are may be some information in cells.cvs", "354677": "May I ask about your approach if you don't mind? You said you use deep learning, is it right? Regarding your merging steps, I believe it might be similar to this:\n\n1) clustering long tracks (i.e., >12 hits) with small epsilon. This would be straight tracks with high confidence so it must be intact.\n2) clustering long tracks with higher epsilon without touching the clustered ones in step 1. These should be long curvy tracks.\n3) extend...\n4) do the similar, with shorter tracks...\n\nIf I am right, may I ask is that your extend/outlier removal step using DL? Thanks so much.", "354689": "The merging steps are similar and I think most of us are similar too. I use deep learning on track finding, and some extension methods are similar as finding stage in my approach. I have trained a model to do outlier removal on good tracks but failed, so I did not apply it yet. And I think every good outlier removal methods are also good for track finding, and it is better to do it on finding stage.", "354729": ">The merging steps are similar and I think most of us are similar too. \n\n@outrunner, I do not think so. You use the worst one. My merging/extension method enables to connect pieces of tracks - full helices, even distorted, are not the problem if there are pieces of them in the soup of track candidates. So, do not call attaching outliers to the tracks of even number of hits as \"Grzegorz's extension\" ;-)\n\nBut seriously, congratulations and best wishes.", "354732": "> You use the worst one.\n\nI like it, as I know @Heng may discover some of my weakness but didn't tell me, haha.", "354779": "outrunner, Heng and many other Kagglers are optimists. They want to have the same chances to win as you, so they think it makes no sense to help you too much ;). I am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.\n\nEDIT: emoticon added", "354796": "I think Heng is helping a lot of people, including outrunner.", "354800": "> I will be glad to see any real Kaggler in TOP10.\n\nYou don't think outrunner will be in top 10?", "354816": "Thanks outrunner for your comment!", "354829": ">You don't think outrunner will be in top 10?\n\n@CPMP, Looking at the 3 events above, @outrunner's score is ~0.88, so his/her place at LB will depend on the number of registered crazytrackers (0.89) https://www.kaggle.com/c/trackml-particle-identification/discussion/60735#354465", "354831": "> I think Heng is helping a lot of people, including outrunner.\n\nI am just kidding, and you are right.\n\n> TOP10 will be occupied by \"novices\"\n\nI think there are a lot of people doing research on this topic will join this competition.\n\n> so his/her place ...\n\n\"his\", Thanks.", "354867": "> I think there are a lot of people doing research on this topic will join this competition.\n\nIf so, short explanation: \n\n\"novice\" - highly educated person who knows how a Kalman filter works\n\n\"Everything but a Kalman filter\" - subtitle of the competition \"forgotten\" by the organizers; the name of the game we are playing", "354871": "Grzegorz, a ML novice who registered Kaggle months ago here, I don't know how a Kalman filter works though. Just a question, may I ask why you stopped making submissions? working on something big?", "354897": "Nicole, not only me - train dataset enables validation good enough, so preparing full submission is a waste of time. Up to now, nobody wants to submit anything above 0.82 - let the others think, there is still a barier for (0,0,z). Nobody wants to submit his 0.90 for a moment of fame, because this will activate Kalman filter implementers. Frankly, I have created quite nice merging&extending method, but still do not have a good method for evaluating track candidates. That is the real reason.", "354915": "Nicole, here is a great intro to Kalman filters: http://www.bzarg.com/p/how-a-kalman-filter-works-in-pictures/", "355050": "Hi @Outrunner, which training file did you find event000008841 in ?\n\nI can only read train_1.zip and train_sample.zip. The other files train_{2,3,4,5}.zip returns a null.", "355053": "That link by @CPMP is a really good one. I found it a few weeks ago and felt my former supervisor (i.e. research supervisor) DR. Bozic will be proud of this author :-)", "355059": "> let the others think, there is still a barier for (0,0,z)\n\nI think @outrunner has already demonstrated that he is long since \"free from origin\" ;-)", "355065": "Edwin, It is not so obvious. A method fitted to one event should be used on another one without any changes. We do not know if it was done and what was the result, so for the world, your method/submission is still the best and still below 0.82.", "355084": "Grzegorz,\n\n> I am skeptical, I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle.\n\nI might fit your description (although I'm afraid it's no guarantee to remain in top 10), but there definitely is another one coming. @demelian's speed of approach is crazy.", "355085": "(although I'm afraid it's no guarantee to remain in top 10),\n\n@Edwin, I think 0.81 or above is pretty safe (even for the prize), it's not like you're gonna sit there and do nothing for another month. And you're right, @demelian is making progress like crazy. If I use my human-learning linear regression approach, he'll jump to nr. 1 in no time. ;)\n\nIncidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?", "355086": "Edwin, All novices are warmly welcomed and invited to the game \"Everything but a Kalman filter\".", "355087": "> Incidentally, are you using a supervised learning (DL) approach as well, just like @outrunner?\n\nJudging by the forum posts I'm using a very different approach than @outrunner. That's about as much as I'm inclined to disclose at this point, I think.", "355115": "4481? is in train_2.zip, you can download it.", "355121": "Thanks I can download it but it is empty as reported here https://www.kaggle.com/c/trackml-particle-identification/discussion/56787#352750\n\n**UPDATE**\n\nI have downloaded train_2.zip and observed the following:-\n\nFor anyone else having the same issue, the reason kernels do not recognize the files in train_{2,3,4,5}.zip is not because they are empty. I believe there is some kind of invisible characters in the file names. Once I renamed them to the same names by retyping, they are readable. I had the same issue reading them locally until I did the renaming.", "370293": "> I am affraid, TOP10 will be occupied by \"novices\" registered one or two months ago, for whom this competition is the first and the only one at Kaggle. I will be glad to see any real Kaggler in TOP10.\n\nGrzegorz,\n\nI was puzzled when you made that comment, which is why I revisit it now that we know the end result. You're right, few newcomers at the top or near it, but also 5 Kaggle masters or grandmasters in top 10. I guess these can count as \"real Kagglers\" ;)" }, "source": "meta" }