![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2779868%2F67fda4eb1ab8e0ff349291f3dd8fbad5%2FScreenshot%20from%202019-09-08%2002-00-04.png?generation=1567900836487871&alt=media\")
Congratulations to the winners and golds!\nLooking forward to read your solutions :)
\n\n@wowfattie @firenero @aispiriants @artemis874 @lanjunyelan @dxchen @huangzihan @garybios @haqishen @bloodaxe @selimsef
\n\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2779868%2Fd114f09c3bc993f5db5a3505a03aac5b%2FScreenshot%20from%202019-09-08%2001-33-34.png?generation=1567900371455618&alt=media\")
well, I have to say I didn't expect this\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2779868%2Fa44799ea775c6f4dd2060768893fe121%2FScreenshot%20from%202019-09-08%2002-07-33.png?generation=1567901290787847&alt=media\")
The competition is hard, my models all have low public lb and high private lb. Iβm not good to trust my models, lol. BTW, learnt a lot from you and Hamid.
", "rawMarkdown": "The competition is hard, my models all have low public lb and high private lb. Iβm not good to trust my models, lol. BTW, learnt a lot from you and Hamid.", "votes": 1 }, { "id": 620795, "postDate": "2019-09-08T01:04:31.187Z", "content": "Sad to hear this, because it seems that simple folking solutions get better rank than some of the hard working script.
", "rawMarkdown": "Sad to hear this, because it seems that simple folking solutions get better rank than some of the hard working script.", "votes": 1 }, { "id": 620886, "postDate": "2019-09-08T03:23:52.440Z", "content": "You know what. Our best selection got only .899, it is our reserve 2nd plan that make us silver >_< In this respect, our team a bit has better luck :)
\n\nEDIT just to clarify, we did select our most promising 2 subs, but this selection process is really difficult!
", "rawMarkdown": "You know what. Our best selection got only .899, it is our reserve 2nd plan that make us silver >___< In this respect, our team a bit has better luck :)\n\n**EDIT** just to clarify, we did select our most promising 2 subs, but this selection process is really difficult!", "votes": 2 }, { "id": 621122, "postDate": "2019-09-08T08:09:05.953Z", "content": "We cannot image the huge gap between public test set and private test. And we donβt following this point: good methods which really works on 2015 data sets will do just as well in 2019 dataset.
", "rawMarkdown": "We cannot image the huge gap between public test set and private test. And we donβt following this point: good methods which really works on 2015 data sets will do just as well in 2019 dataset." }, { "id": 621132, "postDate": "2019-09-08T08:23:47.927Z", "content": "hello from 1046 :D
", "rawMarkdown": "hello from 1046 :D" }, { "id": 622053, "postDate": "2019-09-09T08:13:40.013Z", "content": "That really hurts ! Did you check your other submissions ? Is there any scoring high on Private LB ?
\n\nI checked some of my previous submissions (with lower public LB) scoring 0.922 on private LB. That could have put me in Top-100. But I consider this is what this competition was all about.
", "rawMarkdown": "That really hurts ! Did you check your other submissions ? Is there any scoring high on Private LB ?\n\nI checked some of my previous submissions (with lower public LB) scoring 0.922 on private LB. That could have put me in Top-100. But I consider this is what this competition was all about." }, { "id": 622641, "postDate": "2019-09-09T23:22:27.003Z", "content": "I have senet154 with 0.922, seresnext101 with 0.918, efficient b5 with 0.916. But we haven't choose them to ensemble.
", "rawMarkdown": "I have senet154 with 0.922, seresnext101 with 0.918, efficient b5 with 0.916. But we haven't choose them to ensemble.", "votes": 2 } ] }, { "id": 620745, "postDate": "2019-09-08T00:04:11.687Z", "content": "Congratulations to the winners and golds!\nLooking forward to read your solutions :)
\n\n@wowfattie @firenero @aispiriants @artemis874 @lanjunyelan @dxchen @huangzihan @garybios @haqishen @bloodaxe @selimsef
\n\n
Thank you!\nBut I guess only one of 3rd and 5th place, or neither of them, is going to share his solution.\nI observed that they usually improve their Public LB score at the same time during the last month, and finally shared the same rank shaking on private LB ;)\nBut, of course I have no evidence about that If it turns out that I'm wrong, I'll make an apology to them.
", "rawMarkdown": "Thank you!\nBut I guess only one of 3rd and 5th place, or neither of them, is going to share his solution.\nI observed that they usually improve their Public LB score at the same time during the last month, and finally shared the same rank shaking on private LB ;)\n**But, of course I have no evidence about that** If it turns out that I'm wrong, I'll make an apology to them.", "votes": 3, "replies": [ { "id": 622117, "postDate": "2019-09-09T09:47:28.927Z", "content": "This is a heavy statement without evidence as you said but since you mentioned this already in another thread and they didn't comment, I guess their silence gave you some extra courage :)
", "rawMarkdown": "This is a **heavy statement** without evidence as you said but since you mentioned this already in another thread and they didn't comment, I guess their silence gave you some extra courage :)", "votes": 2 }, { "id": 622727, "postDate": "2019-09-10T02:30:45.633Z", "content": "@valanm I checked the public LB 4~7 times a day during the competition, the phenomenon I saw is too weird to me. I respect Kaggle community and the rules and I hope other guys respect it as well. I hope Kaggle have some technical ways to identify it.
\n\nAs for that disappeared genius account which up to 0.841 LB in 3 days, being shown in this thread https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/105511#latest-621235
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2Ff3de28a241b927255f85178010841e48%2Finbox_2208353_fbe2cba3d576949e0eff79511c05c775_superman2.png?generation=1568082528574160&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F448347%2F24b838e67386f33e50b56b2a15abcd8c%2Finbox_2208353_55b66fb52f9e7df6290e4e3b0a7b39d1_superman2_excel.jpg?generation=1568082704863680&alt=media\")
We can see there is only around 10~ teams that reach 0.841 public LB in 2019.09.01 referring to the following kernel.\nhttps://www.kaggle.com/tanlikesmath/the-race-to-prevent-blindness-aptos-2019
\n\nOnce more, I hope Kaggle have some technical ways to identify who did it. @sohier
", "rawMarkdown": "@valanm I checked the public LB 4~7 times a day during the competition, the phenomenon I saw is too weird to me. **I respect Kaggle community and the rules and I hope other guys respect it as well.** I hope Kaggle have some technical ways to identify it.\n\nAs for that disappeared genius account which up to 0.841 LB in 3 days, being shown in this thread https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/105511#latest-621235\n\n\n\n\n\n\nWe can see there is only around 10~ teams that reach 0.841 public LB in 2019.09.01 referring to the following kernel.\nhttps://www.kaggle.com/tanlikesmath/the-race-to-prevent-blindness-aptos-2019\n\nOnce more, **I hope Kaggle have some technical ways to identify who did it**. @sohier ", "votes": 3 }, { "id": 622824, "postDate": "2019-09-10T06:15:05.667Z", "content": "@haqishen your hybrid with a lemon @qishenhalemon is pretty good at trolling.
", "rawMarkdown": "@haqishen your hybrid with a lemon @qishenhalemon is pretty good at trolling.", "votes": 2 }, { "id": 622847, "postDate": "2019-09-10T06:38:43.163Z", "content": "@valanm He is pretty good at creating new account as well ;)\nI've read your post, it's amazing that you're doing your job mainly by the phone!
", "rawMarkdown": "@valanm He is pretty good at creating new account as well ;)\nI've read your post, it's amazing that you're doing your job mainly by the phone!", "votes": 1 }, { "id": 622848, "postDate": "2019-09-10T06:40:36.443Z", "content": "> Your behaviors are too aggressive. Kaggle admins will handle such cases in a proper way. It's not necessary to remind them again and again.
\n\n@qishenhalemon i think it's completely necessary... the prove that their system is not working properly is that you created a fake account 3hrs ago just for writing that and troll @haqishen (in a disrespectful way). The community should report this kind of behavior and cheaters (that's why kaggle gave us such tools, like: report kernels, votes etc)
\n\n\n\n> It seems to me that only you want is that Kaggle drops some of the top rankings then you can get the prize?
\n\nKaggle is actually checking that, you can read at the top of the LB:\nThis competition has completed. This leaderboard reflects preliminary final standings. The result will become final after the competition organizers verify the results.
\n\nat the competition https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation they removed 2 top20 teams for cheating.
", "rawMarkdown": "> Your behaviors are too aggressive. Kaggle admins will handle such cases in a proper way. It's not necessary to remind them again and again.\n\n@qishenhalemon i think it's completely necessary... the prove that their system is **not working** properly is that you created a fake account 3hrs ago just for writing that and troll @haqishen (in a disrespectful way). The community should report this kind of behavior and cheaters (that's why kaggle gave us such tools, like: report kernels, votes etc)\n\n\n\n> It seems to me that only you want is that Kaggle drops some of the top rankings then you can get the prize?\n\nKaggle is actually checking that, you can read at the top of the LB:\n*This competition has completed. This leaderboard reflects preliminary final standings. The result will become final after the competition organizers verify the results.*\n\nat the competition https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation they removed 2 top20 teams for cheating.", "votes": 2 }, { "id": 622881, "postDate": "2019-09-10T07:54:38.113Z", "content": "@jesucristo thank you ;) I didn't know that news of SIIM competition.
", "rawMarkdown": "@jesucristo thank you ;) I didn't know that news of SIIM competition.", "votes": 1 }, { "id": 622914, "postDate": "2019-09-10T09:00:33.957Z", "content": "@haqishen san, I was 46th at the end of the comp and they removed 2 top-20 teams (I think even one from gold tier) during the post-competition-checkout ... and that's why we reached 44th place hahaha :)
\n\n@qishenhalemon you shouldn't be using that burner account, it's so clear that you created it just for replying/trolling... I don't understand why you hide behind a fake account if you haven't done anything wrong.
", "rawMarkdown": "@haqishen san, I was 46th at the end of the comp and they removed 2 top-20 teams (I think even one from gold tier) during the post-competition-checkout ... and that's why we reached 44th place hahaha :)\n\n@qishenhalemon you shouldn't be using that burner account, it's so clear that you created it just for replying/trolling... I don't understand why you hide behind a fake account if you haven't done anything wrong.", "votes": 1 }, { "id": 623084, "postDate": "2019-09-10T13:21:54.200Z", "content": "@jesucristo san,\nCongratulation! Wow, that's interesting ;) I find that there isn't any official or non-official notification about that. Kaggle is just doing it silently? Is this happens to other competitions?
", "rawMarkdown": "@jesucristo san,\nCongratulation! Wow, that's interesting ;) I find that there isn't any official or non-official notification about that. Kaggle is just doing it silently? Is this happens to other competitions?" }, { "id": 623100, "postDate": "2019-09-10T13:33:40.813Z", "content": "@haqishen I'm \"new\" here so I don't know how often they run/check the solutions... but I think they are starting to do it more often. They don't public a black list or list of cheaters but you can track the LB and see who disappeared, and in my case I realized because my position at the LB changed hahaha \nI had the same experience at generative dogs and jigsaw \nHope Kaggle can find the cheaters here too, and I can go from 990 to 980 π π π \n@drhabib @valanm btw, right now I realized I'm no longer 990 but 989 so they removed 1 person ;)
", "rawMarkdown": "@haqishen I'm \"new\" here so I don't know how often they run/check the solutions... but I think they are starting to do it more often. They don't public a *black list* or *list of cheaters* but you can track the LB and see who disappeared, and in my case I realized because my position at the LB changed hahaha \nI had the same experience at [generative dogs](https://www.kaggle.com/c/generative-dog-images) and [jigsaw](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification) \nHope Kaggle can find the cheaters here too, and I can go from 990 to 980 π π π \n@drhabib @valanm btw, right now I realized I'm no longer 990 but 989 so they removed 1 person ;)", "votes": -1 }, { "id": 623117, "postDate": "2019-09-10T13:43:04.410Z", "content": "@jesucristo Yes, your image attached over there shows that you were 990th 3 days ago π
", "rawMarkdown": "@jesucristo Yes, your image attached over there shows that you were 990th 3 days ago π " }, { "id": 623146, "postDate": "2019-09-10T14:28:51.187Z", "content": "That's most likely someone deleting account themselve
", "rawMarkdown": "That's most likely someone deleting account themselve" }, { "id": 623485, "postDate": "2019-09-11T02:40:31.970Z", "content": "@philippsinger Yes, that make sense, too.
\n\nBut maybe the rank of @jesucristo in SIIM jump from 46th -> 44th is not in this case?\nAs for me I'll definitely not deleting my account that winning a silver medal.
", "rawMarkdown": "@philippsinger Yes, that make sense, too.\n\nBut maybe the rank of @jesucristo in SIIM jump from 46th -> 44th is not in this case?\nAs for me I'll definitely not deleting my account that winning a silver medal." }, { "id": 623493, "postDate": "2019-09-11T02:48:32.143Z", "content": "@jesucristo @philippsinger @valanm I realize that kaggle removed 3 teams in previous top 6 ????\nI suspect 2 of them (previous 3nd, 5th) are cheating but didn't realize that there's a third one.
", "rawMarkdown": "@jesucristo @philippsinger @valanm I realize that kaggle removed 3 teams in previous top 6 ????\nI suspect 2 of them (previous 3nd, 5th) are cheating but didn't realize that there's a third one." }, { "id": 623545, "postDate": "2019-09-11T04:26:06.413Z", "content": "OMG, they got dropped! + rank 2 team also got dropped.
", "rawMarkdown": "OMG, they got dropped! + rank 2 team also got dropped. " }, { "id": 623547, "postDate": "2019-09-11T04:31:05.920Z", "content": "@rishabhiitbhu I'm not surprised to previous 3rd and 5th, but totally have no idea of what happened to previous 2nd team...
", "rawMarkdown": "@rishabhiitbhu I'm not surprised to previous 3rd and 5th, but totally have no idea of what happened to previous 2nd team..." }, { "id": 623597, "postDate": "2019-09-11T05:54:57.333Z", "content": "@haqishen congratz for in-the-money ;)
", "rawMarkdown": "@haqishen congratz for in-the-money ;)" }, { "id": 623600, "postDate": "2019-09-11T06:01:31.750Z", "content": "@philippsinger Thank you! That's exactly a big surprise to me ;)
", "rawMarkdown": "@philippsinger Thank you! That's exactly a big surprise to me ;)", "votes": 1 }, { "id": 623607, "postDate": "2019-09-11T06:13:45.427Z", "content": "LemonTroll saw it coming
", "rawMarkdown": "LemonTroll saw it coming" }, { "id": 623609, "postDate": "2019-09-11T06:15:31.490Z", "content": "@haqishen and thanks for your kind words regarding my approach
", "rawMarkdown": "@haqishen and thanks for your kind words regarding my approach" }, { "id": 623618, "postDate": "2019-09-11T06:33:38.470Z", "content": "@haqishen congratulations! justice has been done ;) I'm glad kaggle's taking action.
", "rawMarkdown": "@haqishen congratulations! justice has been done ;) I'm glad kaggle's taking action.", "votes": -1 }, { "id": 623633, "postDate": "2019-09-11T06:58:36.053Z", "content": "@jesucristo @valanm \nThanks for replying this thread after all of the mess.\nMaybe we have to do this sort of things again and again as long as we are still competing as a kaggler ;)\nHope to see you guys in future competitions!
", "rawMarkdown": "@jesucristo @valanm \nThanks for replying this thread after all of the mess.\nMaybe we have to do this sort of things again and again as long as we are still competing as a kaggler ;)\nHope to see you guys in future competitions!" }, { "id": 623646, "postDate": "2019-09-11T07:11:08.683Z", "content": "well, we just wanted to help clean that mess...
\n\nI think Kaggle could do more in many regards, and the community should contribute whenever possible. For instance, I came up with this idea to prevent probing on private LB: make all images blank but the same size as the original so people can see their submission is successful but that would be it. And kaggle would then rerun only chosen submissions with original images
\n\nI'll stick around :)
", "rawMarkdown": "well, we just wanted to help clean that mess...\n\nI think Kaggle could do more in many regards, and the community should contribute whenever possible. For instance, I came up with this idea to prevent probing on private LB: make all images blank but the same size as the original so people can see their submission is successful but that would be it. And kaggle would then rerun only chosen submissions with original images\n\nI'll stick around :) ", "votes": 1 }, { "id": 623674, "postDate": "2019-09-11T07:37:43.220Z", "content": "@valanm That's an interesting idea!\nIn this case I think APTOS2019 is successfully prevented private testset probing.\nAll submissions are computed on full testset but the public LB score is only calculated on public testset. So that we can only probing the distribution of public testset (we all know that the distribution of public testset is quite different from training set). But it finally turns out that the distribution of private testset is similar to training set!
\n\nMaybe that's why Kaggle is separating public/private testset in this kind of manner.\nSo they are actually trying to prevent private testset probing and telling us to trust our local CV.
\n\n(Although 1st place is only validating by public LB ;)
", "rawMarkdown": "@valanm That's an interesting idea!\nIn this case I think APTOS2019 is successfully prevented private testset probing.\nAll submissions are computed on full testset but the public LB score is only calculated on public testset. So that we can only probing the distribution of public testset (we all know that the distribution of public testset is quite different from training set). But it finally turns out that the distribution of private testset is similar to training set!\n\nMaybe that's why Kaggle is separating public/private testset in this kind of manner.\nSo they are actually trying to prevent private testset probing and telling us to trust our local CV.\n\n(Although 1st place is only validating by public LB ;)", "votes": 1 }, { "id": 623677, "postDate": "2019-09-11T07:43:27.750Z", "content": "It's possible to probe private test distribution\nVUNO team did it but they did it with fake account otherwise they would not be removed
", "rawMarkdown": "It's possible to probe private test distribution\nVUNO team did it but they did it with fake account otherwise they would not be removed" }, { "id": 623685, "postDate": "2019-09-11T07:55:03.160Z", "content": "REALLY??? I didn't know that. Do you know how they did it?
", "rawMarkdown": "REALLY??? I didn't know that. Do you know how they did it?" } ] }, { "id": 620748, "postDate": "2019-09-08T00:07:04.237Z", "content": "Congrats to all winners and unlucky to those that suffered most with the shakeup
", "rawMarkdown": "Congrats to all winners and unlucky to those that suffered most with the shakeup", "votes": 3 }, { "id": 620818, "postDate": "2019-09-08T01:36:27.293Z", "content": "Hopefully kaggle takes note and tries to make public test similar distribution to private test :/\nOtherwise congrats to the winners!
", "rawMarkdown": "Hopefully kaggle takes note and tries to make public test similar distribution to private test :/\nOtherwise congrats to the winners!", "votes": 2 }, { "id": 620747, "postDate": "2019-09-08T00:06:33.137Z", "content": "Thank youuuu!!
", "rawMarkdown": "Thank youuuu!! ", "votes": 2, "replies": [ { "id": 620754, "postDate": "2019-09-08T00:13:19.723Z", "content": "@drhabib Congratulations!
", "rawMarkdown": "@drhabib Congratulations! ", "votes": 1 }, { "id": 620756, "postDate": "2019-09-08T00:14:20.193Z", "content": "thank you! you to with silver!
", "rawMarkdown": "thank you! you to with silver! " }, { "id": 620761, "postDate": "2019-09-08T00:18:20.113Z", "content": "Congratulations and thank you for your contributions to this competition!
", "rawMarkdown": "Congratulations and thank you for your contributions to this competition!", "votes": 1 }, { "id": 620762, "postDate": "2019-09-08T00:18:28.757Z", "content": "Gegeee
", "rawMarkdown": "Gegeee", "votes": 1 }, { "id": 620829, "postDate": "2019-09-08T01:58:06.600Z", "content": "Congrats @drhabib your kernel and discussion helped us a lot!
", "rawMarkdown": "Congrats @drhabib your kernel and discussion helped us a lot!", "votes": 1 }, { "id": 620831, "postDate": "2019-09-08T02:01:05.647Z", "content": "you are welcome! Congrats on silver! =)
", "rawMarkdown": "you are welcome! Congrats on silver! =) ", "votes": 1 }, { "id": 620841, "postDate": "2019-09-08T02:14:49.903Z", "content": "Congratulations @drhabib!!!\nAnd thank you for your insightful posts throughout the competition. Core of my model is based on your posts.
", "rawMarkdown": "Congratulations @drhabib!!!\nAnd thank you for your insightful posts throughout the competition. Core of my model is based on your posts.", "votes": 1 }, { "id": 620843, "postDate": "2019-09-08T02:16:52.483Z", "content": "This is awesome! Congrats on your first silver medal! =)
", "rawMarkdown": "This is awesome! Congrats on your first silver medal! =) ", "votes": 1 }, { "id": 620852, "postDate": "2019-09-08T02:23:34.393Z", "content": "Thanks!!
", "rawMarkdown": "Thanks!!", "votes": 1 }, { "id": 620862, "postDate": "2019-09-08T02:32:18.190Z", "content": "@drhabib congratulations :) Learned a lot from you. Hope you will Share your solution and strategies.
", "rawMarkdown": "@drhabib congratulations :) Learned a lot from you. Hope you will Share your solution and strategies. ", "votes": 1 }, { "id": 620864, "postDate": "2019-09-08T02:36:18.100Z", "content": "Congrats on silver medal =) I am gonna share tomorrow if I will survive drinking =)
", "rawMarkdown": "Congrats on silver medal =) I am gonna share tomorrow if I will survive drinking =) ", "votes": 3 } ] }, { "id": 623621, "postDate": "2019-09-11T06:37:44.253Z", "content": "I created a new discussion about the new LB, I'd like to know why that people was removed...\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108368
", "rawMarkdown": "I created a new discussion about the new LB, I'd like to know why that people was removed...\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108368", "votes": -1 }, { "id": 620828, "postDate": "2019-09-08T01:57:06.483Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 620750, "author_name": "Nanashi", "author_url": "", "post_date": "2019-09-08T00:08:17.950000", "content": "well, I have to say I didn't expect this\n
The competition is hard, my models all have low public lb and high private lb. Iβm not good to trust my models, lol. BTW, learnt a lot from you and Hamid.
", "votes": 1, "replies": [] }, { "id": 620795, "author_name": "SchenbergZ", "author_url": "", "post_date": "2019-09-08T01:04:31.187000", "content": "Sad to hear this, because it seems that simple folking solutions get better rank than some of the hard working script.
", "votes": 1, "replies": [] }, { "id": 620886, "author_name": "Neuron Engineer", "author_url": "", "post_date": "2019-09-08T03:23:52.440000", "content": "You know what. Our best selection got only .899, it is our reserve 2nd plan that make us silver >_< In this respect, our team a bit has better luck :)
\n\nEDIT just to clarify, we did select our most promising 2 subs, but this selection process is really difficult!
", "votes": 2, "replies": [] }, { "id": 621122, "author_name": "seefun", "author_url": "", "post_date": "2019-09-08T08:09:05.953000", "content": "We cannot image the huge gap between public test set and private test. And we donβt following this point: good methods which really works on 2015 data sets will do just as well in 2019 dataset.
", "votes": 0, "replies": [] }, { "id": 621132, "author_name": "Kupchanski", "author_url": "", "post_date": "2019-09-08T08:23:47.927000", "content": "hello from 1046 :D
", "votes": 0, "replies": [] }, { "id": 622053, "author_name": "Filemon", "author_url": "", "post_date": "2019-09-09T08:13:40.013000", "content": "That really hurts ! Did you check your other submissions ? Is there any scoring high on Private LB ?
\n\nI checked some of my previous submissions (with lower public LB) scoring 0.922 on private LB. That could have put me in Top-100. But I consider this is what this competition was all about.
", "votes": 0, "replies": [] }, { "id": 622641, "author_name": "jionie", "author_url": "", "post_date": "2019-09-09T23:22:27.003000", "content": "I have senet154 with 0.922, seresnext101 with 0.918, efficient b5 with 0.916. But we haven't choose them to ensemble.
", "votes": 2, "replies": [] } ] }, { "id": 621873, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-09T04:03:17.120000", "content": "Thank you!\nBut I guess only one of 3rd and 5th place, or neither of them, is going to share his solution.\nI observed that they usually improve their Public LB score at the same time during the last month, and finally shared the same rank shaking on private LB ;)\nBut, of course I have no evidence about that If it turns out that I'm wrong, I'll make an apology to them.
", "votes": 3, "replies": [ { "id": 622117, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-09T09:47:28.927000", "content": "This is a heavy statement without evidence as you said but since you mentioned this already in another thread and they didn't comment, I guess their silence gave you some extra courage :)
", "votes": 2, "replies": [] }, { "id": 622727, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-10T02:30:45.633000", "content": "@valanm I checked the public LB 4~7 times a day during the competition, the phenomenon I saw is too weird to me. I respect Kaggle community and the rules and I hope other guys respect it as well. I hope Kaggle have some technical ways to identify it.
\n\nAs for that disappeared genius account which up to 0.841 LB in 3 days, being shown in this thread https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/105511#latest-621235
\n\n
We can see there is only around 10~ teams that reach 0.841 public LB in 2019.09.01 referring to the following kernel.\nhttps://www.kaggle.com/tanlikesmath/the-race-to-prevent-blindness-aptos-2019
\n\nOnce more, I hope Kaggle have some technical ways to identify who did it. @sohier
", "votes": 3, "replies": [] }, { "id": 622824, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-10T06:15:05.667000", "content": "@haqishen your hybrid with a lemon @qishenhalemon is pretty good at trolling.
", "votes": 2, "replies": [] }, { "id": 622847, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-10T06:38:43.163000", "content": "@valanm He is pretty good at creating new account as well ;)\nI've read your post, it's amazing that you're doing your job mainly by the phone!
", "votes": 1, "replies": [] }, { "id": 622848, "author_name": "Nanashi", "author_url": "", "post_date": "2019-09-10T06:40:36.443000", "content": "> Your behaviors are too aggressive. Kaggle admins will handle such cases in a proper way. It's not necessary to remind them again and again.
\n\n@qishenhalemon i think it's completely necessary... the prove that their system is not working properly is that you created a fake account 3hrs ago just for writing that and troll @haqishen (in a disrespectful way). The community should report this kind of behavior and cheaters (that's why kaggle gave us such tools, like: report kernels, votes etc)
\n\n\n\n> It seems to me that only you want is that Kaggle drops some of the top rankings then you can get the prize?
\n\nKaggle is actually checking that, you can read at the top of the LB:\nThis competition has completed. This leaderboard reflects preliminary final standings. The result will become final after the competition organizers verify the results.
\n\nat the competition https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation they removed 2 top20 teams for cheating.
", "votes": 2, "replies": [] }, { "id": 622881, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-10T07:54:38.113000", "content": "@jesucristo thank you ;) I didn't know that news of SIIM competition.
", "votes": 1, "replies": [] }, { "id": 622914, "author_name": "Nanashi", "author_url": "", "post_date": "2019-09-10T09:00:33.957000", "content": "@haqishen san, I was 46th at the end of the comp and they removed 2 top-20 teams (I think even one from gold tier) during the post-competition-checkout ... and that's why we reached 44th place hahaha :)
\n\n@qishenhalemon you shouldn't be using that burner account, it's so clear that you created it just for replying/trolling... I don't understand why you hide behind a fake account if you haven't done anything wrong.
", "votes": 1, "replies": [] }, { "id": 623084, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-10T13:21:54.200000", "content": "@jesucristo san,\nCongratulation! Wow, that's interesting ;) I find that there isn't any official or non-official notification about that. Kaggle is just doing it silently? Is this happens to other competitions?
", "votes": 0, "replies": [] }, { "id": 623100, "author_name": "Nanashi", "author_url": "", "post_date": "2019-09-10T13:33:40.813000", "content": "@haqishen I'm \"new\" here so I don't know how often they run/check the solutions... but I think they are starting to do it more often. They don't public a black list or list of cheaters but you can track the LB and see who disappeared, and in my case I realized because my position at the LB changed hahaha \nI had the same experience at generative dogs and jigsaw \nHope Kaggle can find the cheaters here too, and I can go from 990 to 980 π π π \n@drhabib @valanm btw, right now I realized I'm no longer 990 but 989 so they removed 1 person ;)
", "votes": -1, "replies": [] }, { "id": 623117, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-10T13:43:04.410000", "content": "@jesucristo Yes, your image attached over there shows that you were 990th 3 days ago π
", "votes": 0, "replies": [] }, { "id": 623146, "author_name": "Psi", "author_url": "", "post_date": "2019-09-10T14:28:51.187000", "content": "That's most likely someone deleting account themselve
", "votes": 0, "replies": [] }, { "id": 623485, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T02:40:31.970000", "content": "@philippsinger Yes, that make sense, too.
\n\nBut maybe the rank of @jesucristo in SIIM jump from 46th -> 44th is not in this case?\nAs for me I'll definitely not deleting my account that winning a silver medal.
", "votes": 0, "replies": [] }, { "id": 623493, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T02:48:32.143000", "content": "@jesucristo @philippsinger @valanm I realize that kaggle removed 3 teams in previous top 6 ????\nI suspect 2 of them (previous 3nd, 5th) are cheating but didn't realize that there's a third one.
", "votes": 0, "replies": [] }, { "id": 623545, "author_name": "Rishabh Agrahari", "author_url": "", "post_date": "2019-09-11T04:26:06.413000", "content": "OMG, they got dropped! + rank 2 team also got dropped.
", "votes": 0, "replies": [] }, { "id": 623547, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T04:31:05.920000", "content": "@rishabhiitbhu I'm not surprised to previous 3rd and 5th, but totally have no idea of what happened to previous 2nd team...
", "votes": 0, "replies": [] }, { "id": 623597, "author_name": "Psi", "author_url": "", "post_date": "2019-09-11T05:54:57.333000", "content": "@haqishen congratz for in-the-money ;)
", "votes": 0, "replies": [] }, { "id": 623600, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T06:01:31.750000", "content": "@philippsinger Thank you! That's exactly a big surprise to me ;)
", "votes": 1, "replies": [] }, { "id": 623607, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-11T06:13:45.427000", "content": "LemonTroll saw it coming
", "votes": 0, "replies": [] }, { "id": 623609, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-11T06:15:31.490000", "content": "@haqishen and thanks for your kind words regarding my approach
", "votes": 0, "replies": [] }, { "id": 623618, "author_name": "Nanashi", "author_url": "", "post_date": "2019-09-11T06:33:38.470000", "content": "@haqishen congratulations! justice has been done ;) I'm glad kaggle's taking action.
", "votes": -1, "replies": [] }, { "id": 623633, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T06:58:36.053000", "content": "@jesucristo @valanm \nThanks for replying this thread after all of the mess.\nMaybe we have to do this sort of things again and again as long as we are still competing as a kaggler ;)\nHope to see you guys in future competitions!
", "votes": 0, "replies": [] }, { "id": 623646, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-11T07:11:08.683000", "content": "well, we just wanted to help clean that mess...
\n\nI think Kaggle could do more in many regards, and the community should contribute whenever possible. For instance, I came up with this idea to prevent probing on private LB: make all images blank but the same size as the original so people can see their submission is successful but that would be it. And kaggle would then rerun only chosen submissions with original images
\n\nI'll stick around :)
", "votes": 1, "replies": [] }, { "id": 623674, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T07:37:43.220000", "content": "@valanm That's an interesting idea!\nIn this case I think APTOS2019 is successfully prevented private testset probing.\nAll submissions are computed on full testset but the public LB score is only calculated on public testset. So that we can only probing the distribution of public testset (we all know that the distribution of public testset is quite different from training set). But it finally turns out that the distribution of private testset is similar to training set!
\n\nMaybe that's why Kaggle is separating public/private testset in this kind of manner.\nSo they are actually trying to prevent private testset probing and telling us to trust our local CV.
\n\n(Although 1st place is only validating by public LB ;)
", "votes": 1, "replies": [] }, { "id": 623677, "author_name": "Miroslav Valan", "author_url": "", "post_date": "2019-09-11T07:43:27.750000", "content": "It's possible to probe private test distribution\nVUNO team did it but they did it with fake account otherwise they would not be removed
", "votes": 0, "replies": [] }, { "id": 623685, "author_name": "Qishen Ha", "author_url": "", "post_date": "2019-09-11T07:55:03.160000", "content": "REALLY??? I didn't know that. Do you know how they did it?
", "votes": 0, "replies": [] } ] }, { "id": 620748, "author_name": "Tom Aindow", "author_url": "", "post_date": "2019-09-08T00:07:04.237000", "content": "Congrats to all winners and unlucky to those that suffered most with the shakeup
", "votes": 3, "replies": [] }, { "id": 620818, "author_name": "sh", "author_url": "", "post_date": "2019-09-08T01:36:27.293000", "content": "Hopefully kaggle takes note and tries to make public test similar distribution to private test :/\nOtherwise congrats to the winners!
", "votes": 2, "replies": [] }, { "id": 620747, "author_name": "DrHB", "author_url": "", "post_date": "2019-09-08T00:06:33.137000", "content": "Thank youuuu!!
", "votes": 2, "replies": [ { "id": 620754, "author_name": "Chanhu", "author_url": "", "post_date": "2019-09-08T00:13:19.723000", "content": "@drhabib Congratulations!
", "votes": 1, "replies": [] }, { "id": 620756, "author_name": "DrHB", "author_url": "", "post_date": "2019-09-08T00:14:20.193000", "content": "thank you! you to with silver!
", "votes": 0, "replies": [] }, { "id": 620761, "author_name": "ilovescience", "author_url": "", "post_date": "2019-09-08T00:18:20.113000", "content": "Congratulations and thank you for your contributions to this competition!
", "votes": 1, "replies": [] }, { "id": 620762, "author_name": "Psi", "author_url": "", "post_date": "2019-09-08T00:18:28.757000", "content": "Gegeee
", "votes": 1, "replies": [] }, { "id": 620829, "author_name": "Udbhav Bamba", "author_url": "", "post_date": "2019-09-08T01:58:06.600000", "content": "Congrats @drhabib your kernel and discussion helped us a lot!
", "votes": 1, "replies": [] }, { "id": 620831, "author_name": "DrHB", "author_url": "", "post_date": "2019-09-08T02:01:05.647000", "content": "you are welcome! Congrats on silver! =)
", "votes": 1, "replies": [] }, { "id": 620841, "author_name": "Ravi Kumar Vadapalli", "author_url": "", "post_date": "2019-09-08T02:14:49.903000", "content": "Congratulations @drhabib!!!\nAnd thank you for your insightful posts throughout the competition. Core of my model is based on your posts.
", "votes": 1, "replies": [] }, { "id": 620843, "author_name": "DrHB", "author_url": "", "post_date": "2019-09-08T02:16:52.483000", "content": "This is awesome! Congrats on your first silver medal! =)
", "votes": 1, "replies": [] }, { "id": 620852, "author_name": "Udbhav Bamba", "author_url": "", "post_date": "2019-09-08T02:23:34.393000", "content": "Thanks!!
", "votes": 1, "replies": [] }, { "id": 620862, "author_name": "Tahsin Mostafiz", "author_url": "", "post_date": "2019-09-08T02:32:18.190000", "content": "@drhabib congratulations :) Learned a lot from you. Hope you will Share your solution and strategies.
", "votes": 1, "replies": [] }, { "id": 620864, "author_name": "DrHB", "author_url": "", "post_date": "2019-09-08T02:36:18.100000", "content": "Congrats on silver medal =) I am gonna share tomorrow if I will survive drinking =)
", "votes": 3, "replies": [] } ] }, { "id": 623621, "author_name": "Nanashi", "author_url": "", "post_date": "2019-09-11T06:37:44.253000", "content": "I created a new discussion about the new LB, I'd like to know why that people was removed...\nhttps://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108368
", "votes": -1, "replies": [] }, { "id": 620828, "author_name": "", "author_url": "", "post_date": "2019-09-08T01:57:06.483000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "620750": "well, I have to say I didn't expect this\n\n", "620745": "Congratulations to the winners and golds!\nLooking forward to read your solutions :)\n\n@wowfattie @firenero @aispiriants @artemis874 @lanjunyelan @dxchen @huangzihan @garybios @haqishen @bloodaxe @selimsef\n\n---\n\n# LB Shake\n\n### PrivateLB\n\n\n![https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2779868%2F1bdf6014e57461b2c9725b458df6e7ee%2FScreenshot%20from%202019-09-10%2008-31-57.png?generation=1568097506504363&alt=media](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2779868%2F1bdf6014e57461b2c9725b458df6e7ee%2FScreenshot%20from%202019-09-10%2008-31-57.png?generation=1568097506504363&alt=media\"){kind=link}