Are we allowed to use ipynb files from GitHub? I'm not sure if the lisense is OK for doing this since there are different kinds of open source lisense for these GitHub repos. And the notebooks on Kaggle also has its lisense here.
\n\nupdate:
\n\nTrain on any external dataset may cause the leak of both CV and LB. So I think I'm not gonna do this until to the last week of this game.
", "messages": [ { "id": 1818956, "postDate": "2022-06-13T10:16:00.890Z", "content": "Are we allowed to use ipynb files from GitHub? I'm not sure if the lisense is OK for doing this since there are different kinds of open source lisense for these GitHub repos. And the notebooks on Kaggle also has its lisense here.
\n\nupdate:
\n\nTrain on any external dataset may cause the leak of both CV and LB. So I think I'm not gonna do this until to the last week of this game.
", "rawMarkdown": "Are we allowed to use ipynb files from GitHub? I'm not sure if the lisense is OK for doing this since there are different kinds of open source lisense for these GitHub repos. And the notebooks on Kaggle also has its lisense here.\n\n~~Anyway I'm going to try the [KGTorrent](https://www.kaggle.com/competitions/AI4Code/discussion/326520) dataset first.~~\n\n---------\nupdate:\n\n~~Like @paweljankiewicz comments, the private set has a high chance coming from public sources. I'll not train any public dataset until the organizers have some comments on this.~~\n\nTrain on any external dataset may cause the leak of both CV and LB. So I think I'm not gonna do this until to the last week of this game.", "votes": 5 }, { "id": 1819110, "postDate": "2022-06-13T13:13:23.650Z", "content": "I'm not sure to be honest about this. I'd say that a chance that the same notebook appears in the public dataset and test notebooks is pretty high. I think every dataset that has notebooks need to be cleared and that includes also other kaggle datasets with notebooks.
", "rawMarkdown": "I'm not sure to be honest about this. I'd say that a chance that the same notebook appears in the public dataset and test notebooks is pretty high. I think every dataset that has notebooks need to be cleared and that includes also other kaggle datasets with notebooks.", "replies": [ { "id": 1819208, "postDate": "2022-06-13T14:27:05.307Z", "content": "I think the same thing. I guess the notebooks in the test set could be the private ones on Kaggle, and private is the default option. User on kaggle has a high chance to make a notebook private on Kaggle but public on GitHub or any other places.
", "rawMarkdown": "I think the same thing. I guess the notebooks in the test set could be the private ones on Kaggle, and private is the default option. User on kaggle has a high chance to make a notebook private on Kaggle but public on GitHub or any other places.", "votes": 1 }, { "id": 1819226, "postDate": "2022-06-13T14:50:21.370Z", "content": "I don't think that they can be private because private notebooks could have far fewer comments and structure in the markdown. I mean there are notebooks in the train data which are like books with index and every comment has a number: 1,2,3 etc. You wouldn't do this in a private notebook. So in short private notebooks can have a different distribution of features.
", "rawMarkdown": "I don't think that they can be private because private notebooks could have far fewer comments and structure in the markdown. I mean there are notebooks in the train data which are like books with index and every comment has a number: 1,2,3 etc. You wouldn't do this in a private notebook. So in short private notebooks can have a different distribution of features.", "votes": 1 }, { "id": 1819241, "postDate": "2022-06-13T15:11:04.197Z", "content": "Ah you are right.. but if the test set is truly coming from the public, then maybe it's best to disable external dataset then. It'll be great if the organizers could have some comments on this one.
", "rawMarkdown": "Ah you are right.. but if the test set is truly coming from the public, then maybe it's best to disable external dataset then. It'll be great if the organizers could have some comments on this one." }, { "id": 1819310, "postDate": "2022-06-13T16:20:12.433Z", "content": "As stated in the oveview and data pages, the private set will be newly gathered for 3 months after the deadline. There is still a chance to leak, but I don't think it will be a big problem.
", "rawMarkdown": "As stated in the oveview and data pages, the private set will be newly gathered for 3 months after the deadline. There is still a chance to leak, but I don't think it will be a big problem.", "votes": 4 }, { "id": 1819358, "postDate": "2022-06-13T17:11:23.427Z", "content": "Datasets are pinned to their version when we rerun notebooks in the 2nd stage. This eliminates any leakage/cheating that you would have from updating an external dataset while the 2nd stage is running.
", "rawMarkdown": "Datasets are pinned to their version when we rerun notebooks in the 2nd stage. This eliminates any leakage/cheating that you would have from updating an external dataset while the 2nd stage is running.", "votes": 5 }, { "id": 1819610, "postDate": "2022-06-14T02:16:02.963Z", "content": "Oh I got it. So if include external dataset there will be a high chance to make CV and LB leak, but not for private LB leak. This is good to know..
\nThanks for answering.
", "rawMarkdown": "Oh I got it. So if include external dataset there will be a high chance to make CV and LB leak, but not for private LB leak. This is good to know..\n\nThanks for answering." } ] }, { "id": 1855567, "postDate": "2022-07-14T17:41:27.897Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1819110, "author_name": "Paweł Jankiewicz", "author_url": "", "post_date": "2022-06-13T13:13:23.650000", "content": "I'm not sure to be honest about this. I'd say that a chance that the same notebook appears in the public dataset and test notebooks is pretty high. I think every dataset that has notebooks need to be cleared and that includes also other kaggle datasets with notebooks.
", "votes": 0, "replies": [ { "id": 1819208, "author_name": "Chenglu", "author_url": "", "post_date": "2022-06-13T14:27:05.307000", "content": "I think the same thing. I guess the notebooks in the test set could be the private ones on Kaggle, and private is the default option. User on kaggle has a high chance to make a notebook private on Kaggle but public on GitHub or any other places.
", "votes": 1, "replies": [] }, { "id": 1819226, "author_name": "Paweł Jankiewicz", "author_url": "", "post_date": "2022-06-13T14:50:21.370000", "content": "I don't think that they can be private because private notebooks could have far fewer comments and structure in the markdown. I mean there are notebooks in the train data which are like books with index and every comment has a number: 1,2,3 etc. You wouldn't do this in a private notebook. So in short private notebooks can have a different distribution of features.
", "votes": 1, "replies": [] }, { "id": 1819241, "author_name": "Chenglu", "author_url": "", "post_date": "2022-06-13T15:11:04.197000", "content": "Ah you are right.. but if the test set is truly coming from the public, then maybe it's best to disable external dataset then. It'll be great if the organizers could have some comments on this one.
", "votes": 0, "replies": [] }, { "id": 1819310, "author_name": "yufuin", "author_url": "", "post_date": "2022-06-13T16:20:12.433000", "content": "As stated in the oveview and data pages, the private set will be newly gathered for 3 months after the deadline. There is still a chance to leak, but I don't think it will be a big problem.
", "votes": 4, "replies": [] }, { "id": 1819358, "author_name": "Will Cukierski", "author_url": "", "post_date": "2022-06-13T17:11:23.427000", "content": "Datasets are pinned to their version when we rerun notebooks in the 2nd stage. This eliminates any leakage/cheating that you would have from updating an external dataset while the 2nd stage is running.
", "votes": 5, "replies": [] }, { "id": 1819610, "author_name": "Chenglu", "author_url": "", "post_date": "2022-06-14T02:16:02.963000", "content": "Oh I got it. So if include external dataset there will be a high chance to make CV and LB leak, but not for private LB leak. This is good to know..
\nThanks for answering.
", "votes": 0, "replies": [] } ] }, { "id": 1855567, "author_name": "", "author_url": "", "post_date": "2022-07-14T17:41:27.897000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1818956": "Are we allowed to use ipynb files from GitHub? I'm not sure if the lisense is OK for doing this since there are different kinds of open source lisense for these GitHub repos. And the notebooks on Kaggle also has its lisense here.\n\n~~Anyway I'm going to try the [KGTorrent](https://www.kaggle.com/competitions/AI4Code/discussion/326520) dataset first.~~\n\n---------\nupdate:\n\n~~Like @paweljankiewicz comments, the private set has a high chance coming from public sources. I'll not train any public dataset until the organizers have some comments on this.~~\n\nTrain on any external dataset may cause the leak of both CV and LB. So I think I'm not gonna do this until to the last week of this game.", "1819110": "I'm not sure to be honest about this. I'd say that a chance that the same notebook appears in the public dataset and test notebooks is pretty high. I think every dataset that has notebooks need to be cleared and that includes also other kaggle datasets with notebooks.", "1855567": "" } }