A similar competition wqas recently held by Dacon.ai from Korea. Links to top solution code has been shared in this competition forum, and some Kaggle public notebooks were created from that.
\nNow this is all gone for a non valid reason (copyright). Linking to external content cannot be a copyright violation. Also, Dacon removed the code from github as well as slides form where they were publicly available.
\nSome of these solutions seem very relevant to this competition. They are most probably used by current top teams who could download the code before it has been removed. Issue is that newcomers cannot do the same.
\nI guess people who downloaded the code cannot republish it because of copyright issues. But the summary of the solutions that were posted in this forum should be put back to level this unfairness a bit IMHO.
\nEdit: I shared a summary of Dacon top solution as a first step to fix this unfairness: https://www.kaggle.com/c/bms-molecular-translation/discussion/229984
\nEdit2: @yasufuminakama has found clones of the erased code and kindly shared in a comment below. I reproduce the links here:
\nhttps://github.com/HamdiTarek/LG_SMILES_1st
\nhttps://github.com/testmailtt/LG_SMILES_2nd
\nhttps://github.com/snapbuy/LG_SMILES_3rd
\nWe are still missing the slides but this is good enough to close the issue IMHO.
\nEdit2: @hasann also cloned one of the solutions and shared it in a comment below. Reproducing it here for convenience: https://github.com/hasan-nn/LG_SMILES_2nd
\nEdit3: I shared the code for these competitions in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions
", "messages": [ { "id": 1259515, "postDate": "2021-04-01T13:22:08.003Z", "content": "\nA similar competition wqas recently held by Dacon.ai from Korea. Links to top solution code has been shared in this competition forum, and some Kaggle public notebooks were created from that.
\nNow this is all gone for a non valid reason (copyright). Linking to external content cannot be a copyright violation. Also, Dacon removed the code from github as well as slides form where they were publicly available.
\nSome of these solutions seem very relevant to this competition. They are most probably used by current top teams who could download the code before it has been removed. Issue is that newcomers cannot do the same.
\nI guess people who downloaded the code cannot republish it because of copyright issues. But the summary of the solutions that were posted in this forum should be put back to level this unfairness a bit IMHO.
\nEdit: I shared a summary of Dacon top solution as a first step to fix this unfairness: https://www.kaggle.com/c/bms-molecular-translation/discussion/229984
\nEdit2: @yasufuminakama has found clones of the erased code and kindly shared in a comment below. I reproduce the links here:
\nhttps://github.com/HamdiTarek/LG_SMILES_1st
\nhttps://github.com/testmailtt/LG_SMILES_2nd
\nhttps://github.com/snapbuy/LG_SMILES_3rd
\nWe are still missing the slides but this is good enough to close the issue IMHO.
\nEdit2: @hasann also cloned one of the solutions and shared it in a comment below. Reproducing it here for convenience: https://github.com/hasan-nn/LG_SMILES_2nd
\nEdit3: I shared the code for these competitions in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions
", "rawMarkdown": "@addisonhoward \n\nA similar competition wqas recently held by Dacon.ai from Korea. [ Links to top solution code](https://www.kaggle.com/c/bms-molecular-translation/discussion/223381) has been shared in this competition forum, and some Kaggle public notebooks were created from that.\n\nNow this is all gone for a non valid reason (copyright). Linking to external content cannot be a copyright violation. Also, Dacon removed the code from github as well as slides form where they were publicly available.\n\nSome of these solutions seem very relevant to this competition. They are most probably used by current top teams who could download the code before it has been removed. Issue is that newcomers cannot do the same.\n\nI guess people who downloaded the code cannot republish it because of copyright issues. But the summary of the solutions that were posted in this forum should be put back to level this unfairness a bit IMHO.\n\nEdit: I shared a summary of Dacon top solution as a first step to fix this unfairness: https://www.kaggle.com/c/bms-molecular-translation/discussion/229984\n\nEdit2: @yasufuminakama has found clones of the erased code and kindly shared in a comment below. I reproduce the links here:\n\nhttps://github.com/HamdiTarek/LG_SMILES_1st\n\nhttps://github.com/testmailtt/LG_SMILES_2nd\n\nhttps://github.com/snapbuy/LG_SMILES_3rd\n\nWe are still missing the slides but this is good enough to close the issue IMHO.\n\nEdit2: @hasann also cloned one of the solutions and shared it in a comment below. Reproducing it here for convenience: https://github.com/hasan-nn/LG_SMILES_2nd\n\nEdit3: I shared the code for these competitions in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions", "votes": 93 }, { "id": 1259780, "postDate": "2021-04-01T16:51:51.700Z", "content": "Dacon removed the codes from github, but it seems the codes are still public from other users github.
\nI guess they forked them before the these got deleted.
\nThat's why it's nonsense to remove open sourced solution.
", "rawMarkdown": "I guess they forked them before the these got deleted. \n\nThat's why it's nonsense to remove open sourced solution. ", "votes": 7 }, { "id": 1259979, "postDate": "2021-04-01T18:46:04.120Z", "content": "@yasufuminakama Thanks for sharing! I will update my post.
", "rawMarkdown": "@yasufuminakama Thanks for sharing! I will update my post.", "votes": 2 } ] }, { "id": 1259730, "postDate": "2021-04-01T16:03:31.413Z", "content": "It is funny to get downvoted on this one. I take it as a way to count who is reusing this top solution approach.
\nSorry guys, I dislike unfairness so much that I act against my own interest by sharing what I know.
", "rawMarkdown": "It is funny to get downvoted on this one. I take it as a way to count who is reusing this top solution approach.\n\nSorry guys, I dislike unfairness so much that I act against my own interest by sharing what I know.", "votes": 19 }, { "id": 1262441, "postDate": "2021-04-04T09:11:09.253Z", "content": "If I remember correctly, the 1st solution only used 120k images from all the training images. They calculated some weights to all the images and chose the 120k based on that; for calculating (summing up?) weights for molecules included things like:
\nI also don't remember the exact specifics, most probably the code can tell more.
", "rawMarkdown": "If I remember correctly, the 1st solution only used 120k images from all the training images. They calculated some weights to all the images and chose the 120k based on that; for calculating (summing up?) weights for molecules included things like:\n\n- does it contain rare atoms\n- how complex they are (i.e. are there atoms that are close to each other)\n- how many atoms they contain\n- and there could be more, but I don't remember\n\nI also don't remember the exact specifics, most probably the code can tell more.", "votes": 5 }, { "id": 1265880, "postDate": "2021-04-07T09:21:29.597Z", "content": "I just found out that this Youtuber had took a recording of my post before it got deleted. People interested in reading the summary of the solutions can check it out.
", "rawMarkdown": "I just found out that [this Youtuber](https://youtu.be/dmg4iFCVW5s?t=576) had took a recording of my post before it got deleted. People interested in reading the summary of the solutions can check it out.", "votes": 6 }, { "id": 1262296, "postDate": "2021-04-04T04:42:42.980Z", "content": "@cpmpml Thank you for bringing this topic up! I just followed up with all the discussions in last 3 days that has been done by the community to minimize the imbalance in access to (already) shared ideas. (I also do not like unfairness) I was somewhat restricted to act on this issue on my own behalf. Glad to see the work done by the community.
", "rawMarkdown": "@cpmpml Thank you for bringing this topic up! I just followed up with all the discussions in last 3 days that has been done by the community to minimize the imbalance in access to (already) shared ideas. (I also do not like unfairness) I was somewhat restricted to act on this issue on my own behalf. Glad to see the work done by the community. ", "votes": 3 }, { "id": 1259525, "postDate": "2021-04-01T13:31:53.020Z", "content": "I saw their code, kept a todo to have a look, only to find out today it was gone. Next time I'll clone before looking…
", "rawMarkdown": "I saw their code, kept a todo to have a look, only to find out today it was gone. Next time I'll clone before looking...", "votes": 3, "replies": [ { "id": 1259669, "postDate": "2021-04-01T15:14:14.610Z", "content": "This comment show something. You are not concern about the people who didn't saw that resources and are at disadvantage, you are mad because you didn't copy.
", "rawMarkdown": "This comment show something. You are not concern about the people who didn't saw that resources and are at disadvantage, you are mad because you didn't copy.", "votes": -18 }, { "id": 1261310, "postDate": "2021-04-02T23:12:28.327Z", "content": "I myself am new to Image Captioning, and before the competition I knew nothing about RNNs or Transformers, I found the open sourced solutions very helpful to me as it is to others and enabled me to base my approach upon them (Instead of starting from scratch).
\n@cpmpml did a great deal by sharing these clones to everyone to keep it fair for newcomers, which is the main point of this topic.
\nI realized that I did fork the second place solution so I am sharing it if helps : https://github.com/hasan-nn/LG_SMILES_2nd.
\n\nThis comment show something. You are not concern about the people who didn't saw that resources and are at disadvantage, you are mad because you didn't copy.
\n
@hiramcho I am glad I didn't read it before now.
\nAnd I am glad to see the community feedback on it.
", "rawMarkdown": "> This comment show something. You are not concern about the people who didn't saw that resources and are at disadvantage, you are mad because you didn't copy.\n\n@hiramcho I am glad I didn't read it before now.\n\nAnd I am glad to see the community feedback on it.", "votes": -1 }, { "id": 1261703, "postDate": "2021-04-03T11:06:25.367Z", "content": "@hasann Thanks, I'll add your link to the post.
", "rawMarkdown": "@hasann Thanks, I'll add your link to the post.", "votes": -1 } ] }, { "id": 1260608, "postDate": "2021-04-02T09:03:39.707Z", "content": "currently the hardware stop newcomers from going forward. of course sharing resource is another issue.
", "rawMarkdown": "currently the hardware stop newcomers from going forward. of course sharing resource is another issue.", "votes": 1 }, { "id": 1259859, "postDate": "2021-04-01T17:49:21.530Z", "content": "Hey all,
\nWe (Kaggle) are not a party to the competition rules and cannot provide legal interpretation on the usability of code generated (directly or indirectly) from outside sources. We were not privy to the license afforded to the Dacon.ai code when it was public, and whether or not its forking is legally viable (e.g. if the code was duplicated and modified at a time when the code was open-sourced, etc).
\nAll we can say is that the winning solutions must satisfy the conditions of winners obligations as outlined in the competition rules, where section 11.b seems the most applicable
\n\n", "rawMarkdown": "Hey all,\n\nWe (Kaggle) are not a party to the competition rules and cannot provide legal interpretation on the usability of code generated (directly or indirectly) from outside sources. We were not privy to the license afforded to the Dacon.ai code when it was public, and whether or not its forking is legally viable (e.g. if the code was duplicated and modified at a time when the code was open-sourced, etc).\n\nAll we can say is that the winning solutions must satisfy the conditions of winners obligations as outlined in the competition rules, where section 11.b seems the most applicable\n\n>As a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:\n\n>b) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license", "votes": 1, "replies": [ { "id": 1259882, "postDate": "2021-04-01T18:06:57.973Z", "content": "\nAs a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:
\nb) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license
\n
The competition sponsors reserve the right to review any solution for submission eligibility
", "rawMarkdown": "The competition sponsors reserve the right to review any solution for submission eligibility", "votes": 2 } ] }, { "id": 1259716, "postDate": "2021-04-01T15:52:02.593Z", "content": "Wonder, why they do that.
", "rawMarkdown": "Wonder, why they do that.", "votes": 1, "replies": [ { "id": 1259727, "postDate": "2021-04-01T16:02:01.490Z", "content": "I wonder too.
", "rawMarkdown": "I wonder too.", "votes": 1 } ] }, { "id": 1259691, "postDate": "2021-04-01T15:34:23.113Z", "content": "Thanks @cpmpml. The solution was deleted when I saw the post.
", "rawMarkdown": "Thanks @cpmpml. The solution was deleted when I saw the post. ", "votes": 1 }, { "id": 1267569, "postDate": "2021-04-08T15:44:40.453Z", "content": "@cpmpml , I don't know if this count as unfairness or not , but after working with Competition data it appears that people having access to beast GPUs as in those provided to Nvidia Ambassadors will be able to experiment alot and converge faster .
\nWhereas other competitors relying on Kaggle environments and Google Colab will most probably have to do quantitative experiments and summarize them well , while this is a good learning , Kaggle should have provided some extra GPU hours to participants competing in this competition who doesn't have good GPUs to make it somewhat fair.
Ys, compute resouce can make a difference, but I can tell you I won some prize in deep learning comps with 2 RTX 1080 Ti GPU when others had access to DGX1.
\nIn this competition I don't know how critical this is as I haven't entered yet.
", "rawMarkdown": "Ys, compute resouce can make a difference, but I can tell you I won some prize in deep learning comps with 2 RTX 1080 Ti GPU when others had access to DGX1.\n\nIn this competition I don't know how critical this is as I haven't entered yet.", "votes": 2 }, { "id": 1267601, "postDate": "2021-04-08T16:10:18.130Z", "content": "@cpmpml seems like learning to do experiments in scarcity of compute resources is the key :) .
", "rawMarkdown": "@cpmpml seems like learning to do experiments in scarcity of compute resources is the key :) .", "votes": 2 }, { "id": 1267604, "postDate": "2021-04-08T16:13:21.203Z", "content": "Always. Too much resource can lead you into some brute force search instead of thinking hard about how to solve the problem.
\nSure, too little resources can be too constraining.
", "rawMarkdown": "Always. Too much resource can lead you into some brute force search instead of thinking hard about how to solve the problem.\n\nSure, too little resources can be too constraining.", "votes": 8 }, { "id": 1267608, "postDate": "2021-04-08T16:22:09.710Z", "content": "@cpmpml Thanks will take note of this points in upcoming experiments :).
", "rawMarkdown": "@cpmpml Thanks will take note of this points in upcoming experiments :).", "votes": 1 } ] }, { "id": 1259718, "postDate": "2021-04-01T15:52:57.970Z", "content": "Agreed…
\nI didn't take closer look to the code and the write up unfortunately. I couldn't imagine that would be deleted either.
", "rawMarkdown": "Agreed...\n\nI didn't take closer look to the code and the write up unfortunately. I couldn't imagine that would be deleted either.", "votes": 2 }, { "id": 1314967, "postDate": "2021-05-19T13:24:59.457Z", "content": "I will remove my smiles-solutions dataset after this competition ends.
", "rawMarkdown": "I will remove my smiles-solutions dataset after this competition ends. ", "votes": -1 }, { "id": 1280243, "postDate": "2021-04-21T18:16:09.037Z", "content": "Never a dull moment on Kaggle. I had a quick scan of the code and it doesnt seem revolutionary. Only solution 2 is still available. From looking at this thread, seems there is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified. It always strikes me with Kaggle, anyway, that if you want to be ahead of the pack, you have to do something different to everyone else, or you will just end up in the middle of the pack.
", "rawMarkdown": "Never a dull moment on Kaggle. I had a quick scan of the code and it doesnt seem revolutionary. Only solution 2 is still available. From looking at this thread, seems there is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified. It always strikes me with Kaggle, anyway, that if you want to be ahead of the pack, you have to do something different to everyone else, or you will just end up in the middle of the pack. ", "replies": [ { "id": 1280265, "postDate": "2021-04-21T19:07:15.510Z", "content": "All 3 solutions are available, just read the post till the end…
", "rawMarkdown": "All 3 solutions are available, just read the post till the end...", "votes": -1 }, { "id": 1280275, "postDate": "2021-04-21T19:23:51.087Z", "content": "\n\nthere is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified
\n
On what ground? Which rule would be violated?
\nI do think it is the reverse: if I had not made them visible again, then sharing them and removing them afterwards is private sharing and it could disqualify all teams that reused them. Now the risk is gone.
\nI did this for fairness, I don't know if these solutions are of any interest or not as I haven't entered this competition yet,
", "rawMarkdown": "> there is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified\n\nOn what ground? Which rule would be violated?\n\nI do think it is the reverse: if I had not made them visible again, then sharing them and removing them afterwards is private sharing and it could disqualify all teams that reused them. Now the risk is gone.\n\nI did this for fairness, I don't know if these solutions are of any interest or not as I haven't entered this competition yet,", "votes": -1 }, { "id": 1280280, "postDate": "2021-04-21T19:31:51.070Z", "content": "\n\nOn what ground? Which rule would be violated?
\n
Well, at least the top solutions github repository had no licence file. No idea, what the legal implications are in the U.S. (for commercial usage).
\nEDIT: Article about that scenario
Your article starts with \"implications of it being embedded in commercial apps\"
\nI haven't done such thing hence I didn't read the rest.
\nI have been involved in quite a few legal actions regarding IP, and the only thing I am sure about is that all advice except from an IP attorney are irrelevant and not worth understanding.
\nGithub T&Cs explicitly state that:
\n\"if you publish your source code in a public repository on GitHub, you have accepted the Terms of Service, by which you allow others to view and fork your repository\"
\nThat's what the people I took the code form did. If they want to come after me, fine, I'll definitely listen to what their attorney have to say. For the record they are:
\nhttps://github.com/HamdiTarek/LG_SMILES_1st\nhttps://github.com/testmailtt/LG_SMILES_2nd\nhttps://github.com/snapbuy/LG_SMILES_3rd\nIf I have to remove the Kaggle dataset then I'll share the forks I did on github. Which I have the right to do per the above provision.
", "rawMarkdown": "Your article starts with \"implications of it being embedded in commercial apps\"\n\nI haven't done such thing hence I didn't read the rest.\n\nI have been involved in quite a few legal actions regarding IP, and the only thing I am sure about is that all advice except from an IP attorney are irrelevant and not worth understanding.\n\nGithub T&Cs explicitly state that:\n\n\"if you publish your source code in a public repository on GitHub, you have accepted the Terms of Service, by which you allow others to view and fork your repository\"\n\nThat's what the people I took the code form did. If they want to come after me, fine, I'll definitely listen to what their attorney have to say. For the record they are:\n\n https://github.com/HamdiTarek/LG_SMILES_1st\n https://github.com/testmailtt/LG_SMILES_2nd\n https://github.com/snapbuy/LG_SMILES_3rd\n\nIf I have to remove the Kaggle dataset then I'll share the forks I did on github. Which I have the right to do per the above provision.", "votes": -1 }, { "id": 1280392, "postDate": "2021-04-21T21:36:05.997Z", "content": "\n\nI haven't done such thing hence I didn't read the rest.
\n
Correct, but the competition hosts might. So it is relevant for your submissions usage if you make place 1-3
\nThe article author is head of their companies legal department. I guess he should know the matter.
Nothing against you, by the way. I find it great that you shared it. I just wanted to point out possible complications.
", "rawMarkdown": "> I haven't done such thing hence I didn't read the rest.\n\nCorrect, but the competition hosts might. So it is relevant for your submissions usage if you make place 1-3\nThe article author is head of their companies legal department. I guess he should know the matter.\n\nNothing against you, by the way. I find it great that you shared it. I just wanted to point out *possible* complications.", "votes": 1 }, { "id": 1280398, "postDate": "2021-04-21T22:01:48.113Z", "content": "If you are looking for all legal implications then you should look at the liabilities we agree to. I have not read the rules in detail enough to be sure it is still the case, but at a point we aere asked to make sure we don't infringe any legal IP rights in our solution code. If we use something preinstalled in Kaggle kernel that infringes some IP rights then we are the ones liable, not Kaggle. Nice isn't it?
\nThis is to say that looking for the worst case scenario leads to nightmarish nights.
\nBack to the topic, I never reuse code as is, I read it and borrow ideas if possible. Ideas are not subject to copyright, hence they can be reused (unless patented). That's what I would do with these SMILES solutions: read them, reuse ideas, but not the code itself.
", "rawMarkdown": "If you are looking for all legal implications then you should look at the liabilities we agree to. I have not read the rules in detail enough to be sure it is still the case, but at a point we aere asked to make sure we don't infringe any legal IP rights in our solution code. If we use something preinstalled in Kaggle kernel that infringes some IP rights then we are the ones liable, not Kaggle. Nice isn't it?\n\nThis is to say that looking for the worst case scenario leads to nightmarish nights.\n\nBack to the topic, I never reuse code as is, I read it and borrow ideas if possible. Ideas are not subject to copyright, hence they can be reused (unless patented). That's what I would do with these SMILES solutions: read them, reuse ideas, but not the code itself." }, { "id": 1280951, "postDate": "2021-04-22T14:06:54.007Z", "content": "you should look at the liabilities we agree to
\nI knew of this rule section, that's why I commented in the first place. Yep still in there, see section 4.
read them, reuse ideas, but not the code itself.
\nThat's what I generally do too. It is also the better way to learn stuff.
hence they can be reused (unless patented)
\nThere is more then just patents, see rule section 4 for a lot of examples.
\nBut I guess, it is a bit overkill and unhealthy to search for all kind of possible legal traps here.
Thanks for this chat. You obviously know about legal matters. I think I will remove my dataset at the end of the competition, no need to have any issue in the future.
", "rawMarkdown": "Thanks for this chat. You obviously know about legal matters. I think I will remove my dataset at the end of the competition, no need to have any issue in the future.\n" }, { "id": 1281107, "postDate": "2021-04-22T16:23:29.460Z", "content": "You are welcome, thank you too.
", "rawMarkdown": "You are welcome, thank you too.", "votes": 1 } ] }, { "id": 1271932, "postDate": "2021-04-13T04:41:34.807Z", "content": "Github Solution 1 is deleted. xD
\nI was expecting something like this.
I shared the code in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions
", "rawMarkdown": "I shared the code in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions\n", "votes": 3 } ] }, { "id": 1267661, "postDate": "2021-04-08T17:20:07.570Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1259780, "author_name": "Y.Nakama", "author_url": "", "post_date": "2021-04-01T16:51:51.700000", "content": "Dacon removed the codes from github, but it seems the codes are still public from other users github.
\nI guess they forked them before the these got deleted.
\nThat's why it's nonsense to remove open sourced solution.
", "votes": 7, "replies": [] }, { "id": 1259979, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-01T18:46:04.120000", "content": "@yasufuminakama Thanks for sharing! I will update my post.
", "votes": 2, "replies": [] } ] }, { "id": 1259730, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-01T16:03:31.413000", "content": "It is funny to get downvoted on this one. I take it as a way to count who is reusing this top solution approach.
\nSorry guys, I dislike unfairness so much that I act against my own interest by sharing what I know.
", "votes": 19, "replies": [] }, { "id": 1262441, "author_name": "nofreewill42", "author_url": "", "post_date": "2021-04-04T09:11:09.253000", "content": "If I remember correctly, the 1st solution only used 120k images from all the training images. They calculated some weights to all the images and chose the 120k based on that; for calculating (summing up?) weights for molecules included things like:
\nI also don't remember the exact specifics, most probably the code can tell more.
", "votes": 5, "replies": [] }, { "id": 1265880, "author_name": "SDSTony", "author_url": "", "post_date": "2021-04-07T09:21:29.597000", "content": "I just found out that this Youtuber had took a recording of my post before it got deleted. People interested in reading the summary of the solutions can check it out.
", "votes": 6, "replies": [] }, { "id": 1262296, "author_name": "SDSTony", "author_url": "", "post_date": "2021-04-04T04:42:42.980000", "content": "@cpmpml Thank you for bringing this topic up! I just followed up with all the discussions in last 3 days that has been done by the community to minimize the imbalance in access to (already) shared ideas. (I also do not like unfairness) I was somewhat restricted to act on this issue on my own behalf. Glad to see the work done by the community.
", "votes": 3, "replies": [] }, { "id": 1259525, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-01T13:31:53.020000", "content": "I saw their code, kept a todo to have a look, only to find out today it was gone. Next time I'll clone before looking…
", "votes": 3, "replies": [ { "id": 1259669, "author_name": "Hiram Coria 🧬", "author_url": "", "post_date": "2021-04-01T15:14:14.610000", "content": "This comment show something. You are not concern about the people who didn't saw that resources and are at disadvantage, you are mad because you didn't copy.
", "votes": -18, "replies": [] }, { "id": 1261310, "author_name": "Hasan N", "author_url": "", "post_date": "2021-04-02T23:12:28.327000", "content": "I myself am new to Image Captioning, and before the competition I knew nothing about RNNs or Transformers, I found the open sourced solutions very helpful to me as it is to others and enabled me to base my approach upon them (Instead of starting from scratch).
\n@cpmpml did a great deal by sharing these clones to everyone to keep it fair for newcomers, which is the main point of this topic.
\nI realized that I did fork the second place solution so I am sharing it if helps : https://github.com/hasan-nn/LG_SMILES_2nd.
\n\nThis comment show something. You are not concern about the people who didn't saw that resources and are at disadvantage, you are mad because you didn't copy.
\n
@hiramcho I am glad I didn't read it before now.
\nAnd I am glad to see the community feedback on it.
", "votes": -1, "replies": [] }, { "id": 1261703, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-03T11:06:25.367000", "content": "@hasann Thanks, I'll add your link to the post.
", "votes": -1, "replies": [] } ] }, { "id": 1260608, "author_name": "dragon zhang", "author_url": "", "post_date": "2021-04-02T09:03:39.707000", "content": "currently the hardware stop newcomers from going forward. of course sharing resource is another issue.
", "votes": 1, "replies": [] }, { "id": 1259859, "author_name": "Addison Howard", "author_url": "", "post_date": "2021-04-01T17:49:21.530000", "content": "Hey all,
\nWe (Kaggle) are not a party to the competition rules and cannot provide legal interpretation on the usability of code generated (directly or indirectly) from outside sources. We were not privy to the license afforded to the Dacon.ai code when it was public, and whether or not its forking is legally viable (e.g. if the code was duplicated and modified at a time when the code was open-sourced, etc).
\nAll we can say is that the winning solutions must satisfy the conditions of winners obligations as outlined in the competition rules, where section 11.b seems the most applicable
\n\n", "votes": 1, "replies": [ { "id": 1259882, "author_name": "Sirish Somanchi", "author_url": "", "post_date": "2021-04-01T18:06:57.973000", "content": "\nAs a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:
\nb) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license
\n
The competition sponsors reserve the right to review any solution for submission eligibility
", "votes": 2, "replies": [] } ] }, { "id": 1259716, "author_name": "tugstugi", "author_url": "", "post_date": "2021-04-01T15:52:02.593000", "content": "Wonder, why they do that.
", "votes": 1, "replies": [ { "id": 1259727, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-01T16:02:01.490000", "content": "I wonder too.
", "votes": 1, "replies": [] } ] }, { "id": 1259691, "author_name": "toxu", "author_url": "", "post_date": "2021-04-01T15:34:23.113000", "content": "Thanks @cpmpml. The solution was deleted when I saw the post.
", "votes": 1, "replies": [] }, { "id": 1267569, "author_name": "Athar Sayed", "author_url": "", "post_date": "2021-04-08T15:44:40.453000", "content": "@cpmpml , I don't know if this count as unfairness or not , but after working with Competition data it appears that people having access to beast GPUs as in those provided to Nvidia Ambassadors will be able to experiment alot and converge faster .
\nWhereas other competitors relying on Kaggle environments and Google Colab will most probably have to do quantitative experiments and summarize them well , while this is a good learning , Kaggle should have provided some extra GPU hours to participants competing in this competition who doesn't have good GPUs to make it somewhat fair.
Ys, compute resouce can make a difference, but I can tell you I won some prize in deep learning comps with 2 RTX 1080 Ti GPU when others had access to DGX1.
\nIn this competition I don't know how critical this is as I haven't entered yet.
", "votes": 2, "replies": [] }, { "id": 1267601, "author_name": "Athar Sayed", "author_url": "", "post_date": "2021-04-08T16:10:18.130000", "content": "@cpmpml seems like learning to do experiments in scarcity of compute resources is the key :) .
", "votes": 2, "replies": [] }, { "id": 1267604, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-08T16:13:21.203000", "content": "Always. Too much resource can lead you into some brute force search instead of thinking hard about how to solve the problem.
\nSure, too little resources can be too constraining.
", "votes": 8, "replies": [] }, { "id": 1267608, "author_name": "Athar Sayed", "author_url": "", "post_date": "2021-04-08T16:22:09.710000", "content": "@cpmpml Thanks will take note of this points in upcoming experiments :).
", "votes": 1, "replies": [] } ] }, { "id": 1259718, "author_name": "Serigne ", "author_url": "", "post_date": "2021-04-01T15:52:57.970000", "content": "Agreed…
\nI didn't take closer look to the code and the write up unfortunately. I couldn't imagine that would be deleted either.
", "votes": 2, "replies": [] }, { "id": 1314967, "author_name": "CPMP", "author_url": "", "post_date": "2021-05-19T13:24:59.457000", "content": "I will remove my smiles-solutions dataset after this competition ends.
", "votes": -1, "replies": [] }, { "id": 1280243, "author_name": "anonemaus", "author_url": "", "post_date": "2021-04-21T18:16:09.037000", "content": "Never a dull moment on Kaggle. I had a quick scan of the code and it doesnt seem revolutionary. Only solution 2 is still available. From looking at this thread, seems there is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified. It always strikes me with Kaggle, anyway, that if you want to be ahead of the pack, you have to do something different to everyone else, or you will just end up in the middle of the pack.
", "votes": 0, "replies": [ { "id": 1280265, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-21T19:07:15.510000", "content": "All 3 solutions are available, just read the post till the end…
", "votes": -1, "replies": [] }, { "id": 1280275, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-21T19:23:51.087000", "content": "\n\nthere is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified
\n
On what ground? Which rule would be violated?
\nI do think it is the reverse: if I had not made them visible again, then sharing them and removing them afterwards is private sharing and it could disqualify all teams that reused them. Now the risk is gone.
\nI did this for fairness, I don't know if these solutions are of any interest or not as I haven't entered this competition yet,
", "votes": -1, "replies": [] }, { "id": 1280280, "author_name": "Gabriel Lindenmaier", "author_url": "", "post_date": "2021-04-21T19:31:51.070000", "content": "\n\nOn what ground? Which rule would be violated?
\n
Well, at least the top solutions github repository had no licence file. No idea, what the legal implications are in the U.S. (for commercial usage).
\nEDIT: Article about that scenario
Your article starts with \"implications of it being embedded in commercial apps\"
\nI haven't done such thing hence I didn't read the rest.
\nI have been involved in quite a few legal actions regarding IP, and the only thing I am sure about is that all advice except from an IP attorney are irrelevant and not worth understanding.
\nGithub T&Cs explicitly state that:
\n\"if you publish your source code in a public repository on GitHub, you have accepted the Terms of Service, by which you allow others to view and fork your repository\"
\nThat's what the people I took the code form did. If they want to come after me, fine, I'll definitely listen to what their attorney have to say. For the record they are:
\nhttps://github.com/HamdiTarek/LG_SMILES_1st\nhttps://github.com/testmailtt/LG_SMILES_2nd\nhttps://github.com/snapbuy/LG_SMILES_3rd\nIf I have to remove the Kaggle dataset then I'll share the forks I did on github. Which I have the right to do per the above provision.
", "votes": -1, "replies": [] }, { "id": 1280392, "author_name": "Gabriel Lindenmaier", "author_url": "", "post_date": "2021-04-21T21:36:05.997000", "content": "\n\nI haven't done such thing hence I didn't read the rest.
\n
Correct, but the competition hosts might. So it is relevant for your submissions usage if you make place 1-3
\nThe article author is head of their companies legal department. I guess he should know the matter.
Nothing against you, by the way. I find it great that you shared it. I just wanted to point out possible complications.
", "votes": 1, "replies": [] }, { "id": 1280398, "author_name": "CPMP", "author_url": "", "post_date": "2021-04-21T22:01:48.113000", "content": "If you are looking for all legal implications then you should look at the liabilities we agree to. I have not read the rules in detail enough to be sure it is still the case, but at a point we aere asked to make sure we don't infringe any legal IP rights in our solution code. If we use something preinstalled in Kaggle kernel that infringes some IP rights then we are the ones liable, not Kaggle. Nice isn't it?
\nThis is to say that looking for the worst case scenario leads to nightmarish nights.
\nBack to the topic, I never reuse code as is, I read it and borrow ideas if possible. Ideas are not subject to copyright, hence they can be reused (unless patented). That's what I would do with these SMILES solutions: read them, reuse ideas, but not the code itself.
", "votes": 0, "replies": [] }, { "id": 1280951, "author_name": "Gabriel Lindenmaier", "author_url": "", "post_date": "2021-04-22T14:06:54.007000", "content": "you should look at the liabilities we agree to
\nI knew of this rule section, that's why I commented in the first place. Yep still in there, see section 4.
read them, reuse ideas, but not the code itself.
\nThat's what I generally do too. It is also the better way to learn stuff.
hence they can be reused (unless patented)
\nThere is more then just patents, see rule section 4 for a lot of examples.
\nBut I guess, it is a bit overkill and unhealthy to search for all kind of possible legal traps here.
Thanks for this chat. You obviously know about legal matters. I think I will remove my dataset at the end of the competition, no need to have any issue in the future.
", "votes": 0, "replies": [] }, { "id": 1281107, "author_name": "Gabriel Lindenmaier", "author_url": "", "post_date": "2021-04-22T16:23:29.460000", "content": "You are welcome, thank you too.
", "votes": 1, "replies": [] } ] }, { "id": 1271932, "author_name": "Salman", "author_url": "", "post_date": "2021-04-13T04:41:34.807000", "content": "Github Solution 1 is deleted. xD
\nI was expecting something like this.
I shared the code in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions
", "votes": 3, "replies": [] } ] }, { "id": 1267661, "author_name": "", "author_url": "", "post_date": "2021-04-08T17:20:07.570000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1259515": "@addisonhoward \n\nA similar competition wqas recently held by Dacon.ai from Korea. [ Links to top solution code](https://www.kaggle.com/c/bms-molecular-translation/discussion/223381) has been shared in this competition forum, and some Kaggle public notebooks were created from that.\n\nNow this is all gone for a non valid reason (copyright). Linking to external content cannot be a copyright violation. Also, Dacon removed the code from github as well as slides form where they were publicly available.\n\nSome of these solutions seem very relevant to this competition. They are most probably used by current top teams who could download the code before it has been removed. Issue is that newcomers cannot do the same.\n\nI guess people who downloaded the code cannot republish it because of copyright issues. But the summary of the solutions that were posted in this forum should be put back to level this unfairness a bit IMHO.\n\nEdit: I shared a summary of Dacon top solution as a first step to fix this unfairness: https://www.kaggle.com/c/bms-molecular-translation/discussion/229984\n\nEdit2: @yasufuminakama has found clones of the erased code and kindly shared in a comment below. I reproduce the links here:\n\nhttps://github.com/HamdiTarek/LG_SMILES_1st\n\nhttps://github.com/testmailtt/LG_SMILES_2nd\n\nhttps://github.com/snapbuy/LG_SMILES_3rd\n\nWe are still missing the slides but this is good enough to close the issue IMHO.\n\nEdit2: @hasann also cloned one of the solutions and shared it in a comment below. Reproducing it here for convenience: https://github.com/hasan-nn/LG_SMILES_2nd\n\nEdit3: I shared the code for these competitions in this dataset https://www.kaggle.com/cpmpml/lg-smiles-solutions", "1259780": "Dacon removed the codes from github, but it seems the codes are still public from other users github.\n- https://github.com/HamdiTarek/LG_SMILES_1st\n- https://github.com/testmailtt/LG_SMILES_2nd\n- https://github.com/snapbuy/LG_SMILES_3rd", "1259730": "It is funny to get downvoted on this one. I take it as a way to count who is reusing this top solution approach.\n\nSorry guys, I dislike unfairness so much that I act against my own interest by sharing what I know.", "1262441": "If I remember correctly, the 1st solution only used 120k images from all the training images. They calculated some weights to all the images and chose the 120k based on that; for calculating (summing up?) weights for molecules included things like:\n\n- does it contain rare atoms\n- how complex they are (i.e. are there atoms that are close to each other)\n- how many atoms they contain\n- and there could be more, but I don't remember\n\nI also don't remember the exact specifics, most probably the code can tell more.", "1265880": "I just found out that [this Youtuber](https://youtu.be/dmg4iFCVW5s?t=576) had took a recording of my post before it got deleted. People interested in reading the summary of the solutions can check it out.", "1262296": "@cpmpml Thank you for bringing this topic up! I just followed up with all the discussions in last 3 days that has been done by the community to minimize the imbalance in access to (already) shared ideas. (I also do not like unfairness) I was somewhat restricted to act on this issue on my own behalf. Glad to see the work done by the community. ", "1259525": "I saw their code, kept a todo to have a look, only to find out today it was gone. Next time I'll clone before looking...", "1260608": "currently the hardware stop newcomers from going forward. of course sharing resource is another issue.", "1259859": "Hey all,\n\nWe (Kaggle) are not a party to the competition rules and cannot provide legal interpretation on the usability of code generated (directly or indirectly) from outside sources. We were not privy to the license afforded to the Dacon.ai code when it was public, and whether or not its forking is legally viable (e.g. if the code was duplicated and modified at a time when the code was open-sourced, etc).\n\nAll we can say is that the winning solutions must satisfy the conditions of winners obligations as outlined in the competition rules, where section 11.b seems the most applicable\n\n>As a condition to being awarded a Prize, a Prize winner must fulfill the following obligations:\n\n>b) grant to the Competition Sponsor the license to the winning Submission stated in the Competition Specific Rules above, and represent that you have the unrestricted right to grant that license", "1259716": "Wonder, why they do that.", "1259691": "Thanks @cpmpml. The solution was deleted when I saw the post. ", "1267569": "@cpmpml , I don't know if this count as unfairness or not , but after working with Competition data it appears that people having access to beast GPUs as in those provided to Nvidia Ambassadors will be able to experiment alot and converge faster .\nWhereas other competitors relying on Kaggle environments and Google Colab will most probably have to do quantitative experiments and summarize them well , while this is a good learning , Kaggle should have provided some extra GPU hours to participants competing in this competition who doesn't have good GPUs to make it somewhat fair.", "1259718": "Agreed...\n\nI didn't take closer look to the code and the write up unfortunately. I couldn't imagine that would be deleted either.", "1314967": "I will remove my smiles-solutions dataset after this competition ends. ", "1280243": "Never a dull moment on Kaggle. I had a quick scan of the code and it doesnt seem revolutionary. Only solution 2 is still available. From looking at this thread, seems there is a reasonable risk that if the Dacon solution is referenced, then the solution will be at risk of being disqualified. It always strikes me with Kaggle, anyway, that if you want to be ahead of the pack, you have to do something different to everyone else, or you will just end up in the middle of the pack. ", "1271932": "Github Solution 1 is deleted. xD\nI was expecting something like this.", "1267661": "" } }