![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1068279%2Fa3d0250ef118534fdd3524ceb5f563b5%2FKiss-classic.jpg?generation=1572950430212672&alt=media\")
Keep It Simple Stupid!!! :)
\n\nI'm not sure if I speak for others but i found the structure of this two stage competition quite confusing!
\n\nIt would be wonderful if Kaggle could adhere to the ancient engineering principal of K.I.S.S....
\n\n
...simplicity should be a key goal in design, and unnecessary complexity should be avoided...
\n\nFor the next competition, having clear instructions about exactly what to upload (the model - is that just code, is it the weights, is it the code and the weights what's the definition of a \"model\"... i'm unclear and this seems very difficult to police and i suspect some folks will take advantage)
\n\nPerhaps simply having one stage where the public leader-board is calculated on a subset of the test images is the way to go Or if the organizers are worried about some folks bending the rules then just release a stage 2 test set (no additional labels no additional training set) and everyone would simply need to run the new stage 2 test set through their model(s) and upload the results.
\n\n... anyway just a thought!
", "messages": [ { "id": 665737, "postDate": "2019-11-05T10:42:20.440Z", "content": "Keep It Simple Stupid!!! :)
\n\nI'm not sure if I speak for others but i found the structure of this two stage competition quite confusing!
\n\nIt would be wonderful if Kaggle could adhere to the ancient engineering principal of K.I.S.S....
\n\n
...simplicity should be a key goal in design, and unnecessary complexity should be avoided...
\n\nFor the next competition, having clear instructions about exactly what to upload (the model - is that just code, is it the weights, is it the code and the weights what's the definition of a \"model\"... i'm unclear and this seems very difficult to police and i suspect some folks will take advantage)
\n\nPerhaps simply having one stage where the public leader-board is calculated on a subset of the test images is the way to go Or if the organizers are worried about some folks bending the rules then just release a stage 2 test set (no additional labels no additional training set) and everyone would simply need to run the new stage 2 test set through their model(s) and upload the results.
\n\n... anyway just a thought!
", "rawMarkdown": "Keep It Simple Stupid!!! :)\n\nI'm not sure if I speak for others but i found the structure of this two stage competition quite confusing!\n\nIt would be wonderful if Kaggle could adhere to the ancient engineering principal of K.I.S.S....\n\n\n\n*...simplicity should be a key goal in design, and unnecessary complexity should be avoided...*\n\nFor the next competition, having clear instructions about exactly what to upload (the model - is that just code, is it the weights, is it the code and the weights what's the definition of a \"model\"... i'm unclear and this seems very difficult to police and i suspect some folks will take advantage)\n\nPerhaps simply having one stage where the public leader-board is calculated on a subset of the test images is the way to go Or if the organizers are worried about some folks bending the rules then just release a stage 2 test set (no additional labels no additional training set) and everyone would simply need to run the new stage 2 test set through their model(s) and upload the results.\n\n... anyway just a thought!", "votes": 26 }, { "id": 665925, "postDate": "2019-11-05T14:43:34.410Z", "content": "I am participating in 2 stage competition for the second time. First time, I didn't know what to do, I was not there for the second stage. Now, thanks to my teammates, I am there in the leaderboard. I am hopeful that Kaggle will take this into consideration. Keep certain images as public LB and rest of them for private LB as in other image competitions.
", "rawMarkdown": "I am participating in 2 stage competition for the second time. First time, I didn't know what to do, I was not there for the second stage. Now, thanks to my teammates, I am there in the leaderboard. I am hopeful that Kaggle will take this into consideration. Keep certain images as public LB and rest of them for private LB as in other image competitions. ", "votes": 3, "replies": [ { "id": 666059, "postDate": "2019-11-05T17:20:16.307Z", "content": "Great to hear! there is always benefits to teaming up :)
", "rawMarkdown": "Great to hear! there is always benefits to teaming up :) ", "votes": 2 } ] }, { "id": 666251, "postDate": "2019-11-05T23:46:03.373Z", "content": "We wanna rock all night and code everyday.
", "rawMarkdown": "We wanna rock all night and code everyday.", "votes": 1, "replies": [ { "id": 666645, "postDate": "2019-11-06T10:14:25.300Z", "content": "... \"rock 'n code\" could be a brand new music genre :)
", "rawMarkdown": "... \"rock 'n code\" could be a brand new music genre :)", "votes": 4 } ] }, { "id": 666089, "postDate": "2019-11-05T18:08:11.147Z", "content": "I got demotivated by the possibility of retraining in stage 2 and the huge dataset (although I should have taken both into account earlier)
\n\nThe rules could have been more clear from the start but we should agree that @juliaelliott did her best to answer all the questions and if you really wanted to get to stage 2 by yesterday (no matter what the score) you probably made it.
\n\nAlso I'm wondering what the upsides to having 2 stages are?
\n\nMy guess would be: \nEnsured reproducability for the organizers is the most important one
\n\nbut I'd like to hear opinions on that :)
", "rawMarkdown": "I got demotivated by the possibility of retraining in stage 2 and the huge dataset (although I should have taken both into account earlier)\n\nThe rules could have been more clear from the start but we should agree that @juliaelliott did her best to answer all the questions and if you really wanted to get to stage 2 by yesterday (no matter what the score) you probably made it.\n\nAlso I'm wondering what the upsides to having 2 stages are?\n\nMy guess would be: \n__Ensured reproducability for the organizers is the most important one__\n\nbut I'd like to hear opinions on that :)", "votes": 2, "replies": [ { "id": 666130, "postDate": "2019-11-05T18:58:03.173Z", "content": "I think it is to prevent some people to probe the test set and adjust the hyper-parameters like this, a posteriori (for what I understood, I am seeing sometimes people complaining about this). And for a potential medical application a test set that is really unseen is important (you really need to know how it works on unseen data for real and how it will behave in a other service for example).
\n\nThat being said I think the competitions with two stages would be more easy if the ergonomy of the kaggle website was think for it. In kernel only competition you cannot broke the rules by accident, it is actually a simpler competition : you have to put your solution in a kernel to submit, and it will be rerun. Here you are constantly thinking about \"do I have the right to do this?\" \"what should I upload ?\" etc.
", "rawMarkdown": "I think it is to prevent some people to probe the test set and adjust the hyper-parameters like this, *a posteriori* (for what I understood, I am seeing sometimes people complaining about this). And for a potential medical application a test set that is really unseen is important (you really need to know how it works on unseen data for real and how it will behave in a other service for example).\n\nThat being said I think the competitions with two stages would be more easy if the ergonomy of the kaggle website was think for it. In kernel only competition you cannot broke the rules by accident, it is actually a simpler competition : you have to put your solution in a kernel to submit, and it will be rerun. Here you are constantly thinking about \"do I have the right to do this?\" \"what should I upload ?\" etc.", "votes": 2 } ] } ], "comments": [ { "id": 665925, "author_name": "Manoj Prabhakar", "author_url": "", "post_date": "2019-11-05T14:43:34.410000", "content": "I am participating in 2 stage competition for the second time. First time, I didn't know what to do, I was not there for the second stage. Now, thanks to my teammates, I am there in the leaderboard. I am hopeful that Kaggle will take this into consideration. Keep certain images as public LB and rest of them for private LB as in other image competitions.
", "votes": 3, "replies": [ { "id": 666059, "author_name": "Daithí O'Manacháin", "author_url": "", "post_date": "2019-11-05T17:20:16.307000", "content": "Great to hear! there is always benefits to teaming up :)
", "votes": 2, "replies": [] } ] }, { "id": 666251, "author_name": "Marília Prata", "author_url": "", "post_date": "2019-11-05T23:46:03.373000", "content": "We wanna rock all night and code everyday.
", "votes": 1, "replies": [ { "id": 666645, "author_name": "Daithí O'Manacháin", "author_url": "", "post_date": "2019-11-06T10:14:25.300000", "content": "... \"rock 'n code\" could be a brand new music genre :)
", "votes": 4, "replies": [] } ] }, { "id": 666089, "author_name": "Cozy Doomer", "author_url": "", "post_date": "2019-11-05T18:08:11.147000", "content": "I got demotivated by the possibility of retraining in stage 2 and the huge dataset (although I should have taken both into account earlier)
\n\nThe rules could have been more clear from the start but we should agree that @juliaelliott did her best to answer all the questions and if you really wanted to get to stage 2 by yesterday (no matter what the score) you probably made it.
\n\nAlso I'm wondering what the upsides to having 2 stages are?
\n\nMy guess would be: \nEnsured reproducability for the organizers is the most important one
\n\nbut I'd like to hear opinions on that :)
", "votes": 2, "replies": [ { "id": 666130, "author_name": "Etienne R", "author_url": "", "post_date": "2019-11-05T18:58:03.173000", "content": "I think it is to prevent some people to probe the test set and adjust the hyper-parameters like this, a posteriori (for what I understood, I am seeing sometimes people complaining about this). And for a potential medical application a test set that is really unseen is important (you really need to know how it works on unseen data for real and how it will behave in a other service for example).
\n\nThat being said I think the competitions with two stages would be more easy if the ergonomy of the kaggle website was think for it. In kernel only competition you cannot broke the rules by accident, it is actually a simpler competition : you have to put your solution in a kernel to submit, and it will be rerun. Here you are constantly thinking about \"do I have the right to do this?\" \"what should I upload ?\" etc.
", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "665737": "Keep It Simple Stupid!!! :)\n\nI'm not sure if I speak for others but i found the structure of this two stage competition quite confusing!\n\nIt would be wonderful if Kaggle could adhere to the ancient engineering principal of K.I.S.S....\n\n\n\n*...simplicity should be a key goal in design, and unnecessary complexity should be avoided...*\n\nFor the next competition, having clear instructions about exactly what to upload (the model - is that just code, is it the weights, is it the code and the weights what's the definition of a \"model\"... i'm unclear and this seems very difficult to police and i suspect some folks will take advantage)\n\nPerhaps simply having one stage where the public leader-board is calculated on a subset of the test images is the way to go Or if the organizers are worried about some folks bending the rules then just release a stage 2 test set (no additional labels no additional training set) and everyone would simply need to run the new stage 2 test set through their model(s) and upload the results.\n\n... anyway just a thought!", "665925": "I am participating in 2 stage competition for the second time. First time, I didn't know what to do, I was not there for the second stage. Now, thanks to my teammates, I am there in the leaderboard. I am hopeful that Kaggle will take this into consideration. Keep certain images as public LB and rest of them for private LB as in other image competitions. ", "666251": "We wanna rock all night and code everyday.", "666089": "I got demotivated by the possibility of retraining in stage 2 and the huge dataset (although I should have taken both into account earlier)\n\nThe rules could have been more clear from the start but we should agree that @juliaelliott did her best to answer all the questions and if you really wanted to get to stage 2 by yesterday (no matter what the score) you probably made it.\n\nAlso I'm wondering what the upsides to having 2 stages are?\n\nMy guess would be: \n__Ensured reproducability for the organizers is the most important one__\n\nbut I'd like to hear opinions on that :)" } }