@juliaelliott \n1) From https://www.kaggle.com/two-stage-frequently-asked-questions for big files\n> you can instead upload a checksum of your archive file (such as an md5 or sha hash)
\n\nQ: Can I calculate this hash (e.g. md5sum final_package.zip), save it to a txt file, and upload this txt file?
\n\n2) From https://www.kaggle.com/two-stage-frequently-asked-questions\n> you may be permitted to select two submissions for final scoring. If so, don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each.
\n\nQ: Does it mean that I can put all models and code into final_package.zip, including training instruction, and add several (e.g. N>2) instructions for test submission, each of them will be with different test parameters (e.g. instructionA - \"use modelA for submission\", instructionB - \"use modelB for submission\"), and during 2nd stage I will made N submissions for all instructions and select two of them for a final submits? Is it ok?
@juliaelliott \n1) From https://www.kaggle.com/two-stage-frequently-asked-questions for big files\n> you can instead upload a checksum of your archive file (such as an md5 or sha hash)
\n\nQ: Can I calculate this hash (e.g. md5sum final_package.zip), save it to a txt file, and upload this txt file?
\n\n2) From https://www.kaggle.com/two-stage-frequently-asked-questions\n> you may be permitted to select two submissions for final scoring. If so, don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each.
\n\nQ: Does it mean that I can put all models and code into final_package.zip, including training instruction, and add several (e.g. N>2) instructions for test submission, each of them will be with different test parameters (e.g. instructionA - \"use modelA for submission\", instructionB - \"use modelB for submission\"), and during 2nd stage I will made N submissions for all instructions and select two of them for a final submits? Is it ok?
I have a question on the kind of retraining that's allowed.\nWhen I was training during stage 1, I saved several model checkpoints and afterwards used the one with the best cv score as my final model.
\n\nI didn't know we'd be allowed to retrain in stage 2 so I didn't automate the step of selecting the final model, neither did I include a text description (although it's clear which one I took if you look at the checkpoints and the final model which are all included in my upload).
\n\nI would be able to retrain my pipeline on stage 2 without any changes, but I would still need to manually select the best scoring model from my checkpoints. Would that violate the stage 2 rules?
\n\nBest,\nSzymon
", "rawMarkdown": "I have a question on the kind of retraining that's allowed.\nWhen I was training during stage 1, I saved several model checkpoints and afterwards used the one with the best cv score as my final model.\n\nI didn't know we'd be allowed to retrain in stage 2 so I didn't automate the step of selecting the final model, neither did I include a text description (although it's clear which one I took if you look at the checkpoints and the final model which are all included in my upload).\n\nI would be able to retrain my pipeline on stage 2 without any changes, but I would still need to manually select the best scoring model from my checkpoints. Would that violate the stage 2 rules?\n\nBest,\nSzymon", "votes": 1 }, { "id": 665733, "postDate": "2019-11-05T10:37:51.180Z", "content": "Do anyone have absecence of \"Submit predictions\" button in stage 2?
", "rawMarkdown": "Do anyone have absecence of \"Submit predictions\" button in stage 2?", "votes": 1, "replies": [ { "id": 665848, "postDate": "2019-11-05T13:12:01.280Z", "content": "yes.. all till date submission showing up errored..\non Public LB can see only 600 ranks
", "rawMarkdown": "yes.. all till date submission showing up errored..\non Public LB can see only 600 ranks", "votes": 1 } ] }, { "id": 665298, "postDate": "2019-11-04T21:24:11.083Z", "content": "\n\nI just tried to download the uploaded zip archive to make sure everything is fine once again but unable to download it. I clicked download button but it does not start downloading. I have no idea whether this is an expected behavior for 2 stage competition or a bug.
", "rawMarkdown": "@juliaelliott\n\nI just tried to download the uploaded zip archive to make sure everything is fine once again but unable to download it. I clicked download button but it does not start downloading. I have no idea whether this is an expected behavior for 2 stage competition or a bug. ", "votes": 1 }, { "id": 665144, "postDate": "2019-11-04T17:13:08.517Z", "content": "@vadiksadik Yes to both questions.
", "rawMarkdown": "@vadiksadik Yes to both questions.\n", "votes": 1, "replies": [ { "id": 665151, "postDate": "2019-11-04T17:20:32.140Z", "content": "Second Yes surprised me, I thought we need to decide on specific logic for both submissions now. Does N from the OP question has to be finite? Does it have to be a countable set?
@juliaelliott thank you for the reply.\nAt the bottom of this thread, you said
\n\n\n\n\nYou should be locking in your 2 best models that you intend to rerun yourself to generate your stage 2 submissions.
\n
But at the same time, you said Yes to my 2nd question where I asked about uploading all of my models and all of my instructions now and selecting two of them for final submission.
\n\nI assume these two are different situations.\nSo, finally, is it still Yes to my 2nd question? 🙂
", "rawMarkdown": "@juliaelliott thank you for the reply.\nAt the bottom of this thread, you said \n> You should be locking in your 2 best models that you intend to rerun yourself to generate your stage 2 submissions.\n\nBut at the same time, you said **Yes** to my 2nd question where I asked about uploading all of my models and all of my instructions now and selecting two of them for final submission.\n\nI assume these two are different situations.\nSo, finally, is it still **Yes** to my 2nd question? 🙂 " }, { "id": 665285, "postDate": "2019-11-04T20:56:36.480Z", "content": "@vadiksadik When you were referring to \"all models,\" I made the incorrect assumption you were referring to the 2 models that represent those that you will run in stage 2. Correction: You should not be uploading >2 models. But you can, in stage 2, make >2 submissions (this is to allow room for debugging or additional submissions required to re-train on the additional train data). And then, in the end, you need to make 2 final submission selections before the stage 2 deadline.
", "rawMarkdown": "@vadiksadik When you were referring to \"all models,\" I made the incorrect assumption you were referring to the 2 models that represent those that you will run in stage 2. Correction: You should not be uploading >2 models. But you can, in stage 2, make >2 submissions (this is to allow room for debugging or additional submissions required to re-train on the additional train data). And then, in the end, you need to make 2 final submission selections before the stage 2 deadline.", "votes": 3 }, { "id": 665289, "postDate": "2019-11-04T21:14:16.807Z", "content": "@juliaelliott \nIf we added 2 models for stage 2 but added instructions.txt file with information on how we plan to improve that is it OK or should be removed? we didn't know at what extent the code can be changed.
\n\nI was referring to your comment to @vadiksadik saying \"Yes to both questions.\" and did that.
\n\nPlease clarify this. I think this will not make a wrong upload for us.
", "rawMarkdown": "@juliaelliott \nIf we added 2 models for stage 2 but added instructions.txt file with information on how we plan to improve that is it OK or should be removed? we didn't know at what extent the code can be changed. \n\nI was referring to your comment to @vadiksadik saying \"Yes to both questions.\" and did that. \n\nPlease clarify this. I think this will not make a wrong upload for us. \n", "votes": 1 }, { "id": 665295, "postDate": "2019-11-04T21:21:37.463Z", "content": "@muhakabartay No, you should not be making code changes beyond what you upload as your stage 1 model. So, you cannot be including instructions on how to change code as a way to circumvent this rule. Choose your 2 models and use those in stage 2 for your submissions,.
", "rawMarkdown": "@muhakabartay No, you should not be making code changes beyond what you upload as your stage 1 model. So, you cannot be including instructions on how to change code as a way to circumvent this rule. Choose your 2 models and use those in stage 2 for your submissions,.", "votes": 1 }, { "id": 665303, "postDate": "2019-11-04T21:28:37.717Z", "content": "@juliaelliott fuuuh, re-uploaded ))) fortunately
", "rawMarkdown": "@juliaelliott fuuuh, re-uploaded ))) fortunately ", "votes": 1 }, { "id": 665312, "postDate": "2019-11-04T21:36:11.243Z", "content": "Hi @juliaelliott , \nIf I did not train a particular model for the ensemble, but write the full code for the training and ensemble, and submit the code, can I rerun that code with the stage 2 and stage 1 data for the final submission?
\n\nThanks.
", "rawMarkdown": "Hi @juliaelliott , \nIf I did not train a particular model for the ensemble, but write the full code for the training and ensemble, and submit the code, can I rerun that code with the stage 2 and stage 1 data for the final submission?\n\nThanks.", "votes": 1 }, { "id": 665343, "postDate": "2019-11-04T22:31:24.163Z", "content": "@bopengiowa Yes, that's fine, since you're permitted to retrain on stage 2's training data (which includes both the train and test from stage 1 -- labels to be released upon stage 2's start).
", "rawMarkdown": "@bopengiowa Yes, that's fine, since you're permitted to retrain on stage 2's training data (which includes both the train and test from stage 1 -- labels to be released upon stage 2's start)." }, { "id": 665344, "postDate": "2019-11-04T22:34:54.163Z", "content": "Thanks for the response.
", "rawMarkdown": "Thanks for the response." } ] }, { "id": 665142, "postDate": "2019-11-04T17:12:15.583Z", "content": "@juliaelliott - adding to the above wonderful question, I have one more question. There are 3 models which are performing almost equal in stage 1. Can we upload 3 all codes and then select 2 from those 3 as the best submissions for stage 2. Your inputs wil be highly appreciated ...
", "rawMarkdown": "@juliaelliott - adding to the above wonderful question, I have one more question. There are 3 models which are performing almost equal in stage 1. Can we upload 3 all codes and then select 2 from those 3 as the best submissions for stage 2. Your inputs wil be highly appreciated ...", "votes": 1, "replies": [ { "id": 665148, "postDate": "2019-11-04T17:16:10.767Z", "content": "@manojprabhaakr You should be uploading the 2 models you will rerun to make your stage 2 submissions.
", "rawMarkdown": "@manojprabhaakr You should be uploading the 2 models you will rerun to make your stage 2 submissions.", "votes": 1 }, { "id": 665175, "postDate": "2019-11-04T17:37:43.237Z", "content": "Thanks for the clarification @juliaelliott
", "rawMarkdown": "Thanks for the clarification @juliaelliott ", "votes": 1 } ] }, { "id": 665979, "postDate": "2019-11-05T15:54:04.180Z", "content": "@juliaelliott\nCould you clarify what kind of changes we can make with kernels in stage 2?
\n\nA) We can change path to new data? Allowed? \nB) We can change for example epochs=5 to epochs=10, or lr=0.01 to lr=0.1, or smth like that without new lines of code but change of variables. Allowed?\nC) More than in B) and if allowed, so what exactly.
\n\nThe reason I am asking, you said we can make several submissions and choose that best one which obviously mean that some changes allowed (otherwise, why I need to submit exact same code few times 🤔).
\n\nThank you very much for clarification in advance.
\n\nSorry for disturbing so many times.
\n\nBest, \nMukharbek
", "rawMarkdown": "@juliaelliott\nCould you clarify what kind of changes we can make with kernels in stage 2?\n\nA) We can change path to new data? Allowed? \nB) We can change for example epochs=5 to epochs=10, or lr=0.01 to lr=0.1, or smth like that without new lines of code but change of variables. Allowed?\nC) More than in B) and if allowed, so what exactly.\n\nThe reason I am asking, you said we can make several submissions and choose that best one which obviously mean that some changes allowed (otherwise, why I need to submit exact same code few times 🤔). \n\nThank you very much for clarification in advance.\n\nSorry for disturbing so many times. \n\nBest, \nMukharbek \n\n\n\n", "votes": 2, "replies": [ { "id": 665991, "postDate": "2019-11-05T16:08:08.113Z", "content": "As far as I understand,\n- A is allowed\n- B is not allowed
\n\nI actually see some flaws of this two stage competition format for the reason you mentioned. \nWe are not allowed to tune hyperparameters but we are allowed to prepare hard coded variations of our scripts and make more than 20 or 30 submissions based on different hyperparametes and choose the one which sounds the best based on cv and lb scores.
", "rawMarkdown": "As far as I understand,\n- A is allowed\n- B is not allowed\n\nI actually see some flaws of this two stage competition format for the reason you mentioned. \nWe are not allowed to tune hyperparameters but we are allowed to prepare hard coded variations of our scripts and make more than 20 or 30 submissions based on different hyperparametes and choose the one which sounds the best based on cv and lb scores.", "votes": 2 } ] }, { "id": 665305, "postDate": "2019-11-04T21:30:05.917Z", "content": "@appian I have the same issue when tried to download to check that was a correct file, but it just jumps to up of the page without downloading.
", "rawMarkdown": "@appian I have the same issue when tried to download to check that was a correct file, but it just jumps to up of the page without downloading. ", "votes": 2, "replies": [ { "id": 665309, "postDate": "2019-11-04T21:33:16.127Z", "content": "Exactly. This makes me worried because I can't make sure upload was successful or not.
", "rawMarkdown": "Exactly. This makes me worried because I can't make sure upload was successful or not.", "votes": 1 } ] }, { "id": 665137, "postDate": "2019-11-04T17:08:21.137Z", "content": "\n\nSame question from me. I don't understand this part
\n\n\"In order to be eligible for Stage 2, each team's Stage 1 submission must include the model uploaded, via Team -> Your Model. This model should match that which will be used to generate the 2 final submissions you will select for scoring in Stage 2. Be aware that if you do not select 2 final submissions (via 'My Submissions'), the platform will auto-select your best-scoring model on the public leaderboard. The deadline for model upload is firmly the end of Stage 1.\"
\n\nCould you write more explicitly whether we do selection of 2 best our models now and zip it or we add all we wish (4-5 for example) and do selection at stage 2 (as pointed by @vadiksadik above in part 2) )??
", "rawMarkdown": "@juliaelliott \n\nSame question from me. I don't understand this part\n\n\"In order to be eligible for Stage 2, each team's Stage 1 submission must include the model uploaded, via Team -> Your Model. This model should match that which will be used to generate the 2 final submissions you will select for scoring in Stage 2. Be aware that if you do not select 2 final submissions (via 'My Submissions'), the platform will auto-select your best-scoring model on the public leaderboard. The deadline for model upload is firmly the end of Stage 1.\"\n\nCould you write more explicitly whether we do selection of 2 best our models now and zip it or we add all we wish (4-5 for example) and do selection at stage 2 (as pointed by @vadiksadik above in part 2) )??\n", "votes": 2, "replies": [ { "id": 665146, "postDate": "2019-11-04T17:14:02.883Z", "content": "@muhakabartay You should be locking in your 2 best models that you intend to rerun yourself to generate your stage 2 submissions.
", "rawMarkdown": "@muhakabartay You should be locking in your 2 best models that you intend to rerun yourself to generate your stage 2 submissions.", "votes": 1 }, { "id": 665181, "postDate": "2019-11-04T17:46:31.413Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 666113, "author_name": "Szymon", "author_url": "", "post_date": "2019-11-05T18:30:46.293000", "content": "I have a question on the kind of retraining that's allowed.\nWhen I was training during stage 1, I saved several model checkpoints and afterwards used the one with the best cv score as my final model.
\n\nI didn't know we'd be allowed to retrain in stage 2 so I didn't automate the step of selecting the final model, neither did I include a text description (although it's clear which one I took if you look at the checkpoints and the final model which are all included in my upload).
\n\nI would be able to retrain my pipeline on stage 2 without any changes, but I would still need to manually select the best scoring model from my checkpoints. Would that violate the stage 2 rules?
\n\nBest,\nSzymon
", "votes": 1, "replies": [] }, { "id": 665733, "author_name": "Alexey Kachalov", "author_url": "", "post_date": "2019-11-05T10:37:51.180000", "content": "Do anyone have absecence of \"Submit predictions\" button in stage 2?
", "votes": 1, "replies": [ { "id": 665848, "author_name": "Jaideep", "author_url": "", "post_date": "2019-11-05T13:12:01.280000", "content": "yes.. all till date submission showing up errored..\non Public LB can see only 600 ranks
", "votes": 1, "replies": [] } ] }, { "id": 665298, "author_name": "Appian", "author_url": "", "post_date": "2019-11-04T21:24:11.083000", "content": "\n\nI just tried to download the uploaded zip archive to make sure everything is fine once again but unable to download it. I clicked download button but it does not start downloading. I have no idea whether this is an expected behavior for 2 stage competition or a bug.
", "votes": 1, "replies": [] }, { "id": 665144, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T17:13:08.517000", "content": "@vadiksadik Yes to both questions.
", "votes": 1, "replies": [ { "id": 665151, "author_name": "nosound", "author_url": "", "post_date": "2019-11-04T17:20:32.140000", "content": "Second Yes surprised me, I thought we need to decide on specific logic for both submissions now. Does N from the OP question has to be finite? Does it have to be a countable set?
@juliaelliott thank you for the reply.\nAt the bottom of this thread, you said
\n\n\n\n\nYou should be locking in your 2 best models that you intend to rerun yourself to generate your stage 2 submissions.
\n
But at the same time, you said Yes to my 2nd question where I asked about uploading all of my models and all of my instructions now and selecting two of them for final submission.
\n\nI assume these two are different situations.\nSo, finally, is it still Yes to my 2nd question? 🙂
", "votes": 0, "replies": [] }, { "id": 665285, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T20:56:36.480000", "content": "@vadiksadik When you were referring to \"all models,\" I made the incorrect assumption you were referring to the 2 models that represent those that you will run in stage 2. Correction: You should not be uploading >2 models. But you can, in stage 2, make >2 submissions (this is to allow room for debugging or additional submissions required to re-train on the additional train data). And then, in the end, you need to make 2 final submission selections before the stage 2 deadline.
", "votes": 3, "replies": [] }, { "id": 665289, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2019-11-04T21:14:16.807000", "content": "@juliaelliott \nIf we added 2 models for stage 2 but added instructions.txt file with information on how we plan to improve that is it OK or should be removed? we didn't know at what extent the code can be changed.
\n\nI was referring to your comment to @vadiksadik saying \"Yes to both questions.\" and did that.
\n\nPlease clarify this. I think this will not make a wrong upload for us.
", "votes": 1, "replies": [] }, { "id": 665295, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T21:21:37.463000", "content": "@muhakabartay No, you should not be making code changes beyond what you upload as your stage 1 model. So, you cannot be including instructions on how to change code as a way to circumvent this rule. Choose your 2 models and use those in stage 2 for your submissions,.
", "votes": 1, "replies": [] }, { "id": 665303, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2019-11-04T21:28:37.717000", "content": "@juliaelliott fuuuh, re-uploaded ))) fortunately
", "votes": 1, "replies": [] }, { "id": 665312, "author_name": "Bo Peng", "author_url": "", "post_date": "2019-11-04T21:36:11.243000", "content": "Hi @juliaelliott , \nIf I did not train a particular model for the ensemble, but write the full code for the training and ensemble, and submit the code, can I rerun that code with the stage 2 and stage 1 data for the final submission?
\n\nThanks.
", "votes": 1, "replies": [] }, { "id": 665343, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T22:31:24.163000", "content": "@bopengiowa Yes, that's fine, since you're permitted to retrain on stage 2's training data (which includes both the train and test from stage 1 -- labels to be released upon stage 2's start).
", "votes": 0, "replies": [] }, { "id": 665344, "author_name": "Bo Peng", "author_url": "", "post_date": "2019-11-04T22:34:54.163000", "content": "Thanks for the response.
", "votes": 0, "replies": [] } ] }, { "id": 665142, "author_name": "Manoj Prabhakar", "author_url": "", "post_date": "2019-11-04T17:12:15.583000", "content": "@juliaelliott - adding to the above wonderful question, I have one more question. There are 3 models which are performing almost equal in stage 1. Can we upload 3 all codes and then select 2 from those 3 as the best submissions for stage 2. Your inputs wil be highly appreciated ...
", "votes": 1, "replies": [ { "id": 665148, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T17:16:10.767000", "content": "@manojprabhaakr You should be uploading the 2 models you will rerun to make your stage 2 submissions.
", "votes": 1, "replies": [] }, { "id": 665175, "author_name": "Manoj Prabhakar", "author_url": "", "post_date": "2019-11-04T17:37:43.237000", "content": "Thanks for the clarification @juliaelliott
", "votes": 1, "replies": [] } ] }, { "id": 665979, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2019-11-05T15:54:04.180000", "content": "@juliaelliott\nCould you clarify what kind of changes we can make with kernels in stage 2?
\n\nA) We can change path to new data? Allowed? \nB) We can change for example epochs=5 to epochs=10, or lr=0.01 to lr=0.1, or smth like that without new lines of code but change of variables. Allowed?\nC) More than in B) and if allowed, so what exactly.
\n\nThe reason I am asking, you said we can make several submissions and choose that best one which obviously mean that some changes allowed (otherwise, why I need to submit exact same code few times 🤔).
\n\nThank you very much for clarification in advance.
\n\nSorry for disturbing so many times.
\n\nBest, \nMukharbek
", "votes": 2, "replies": [ { "id": 665991, "author_name": "Appian", "author_url": "", "post_date": "2019-11-05T16:08:08.113000", "content": "As far as I understand,\n- A is allowed\n- B is not allowed
\n\nI actually see some flaws of this two stage competition format for the reason you mentioned. \nWe are not allowed to tune hyperparameters but we are allowed to prepare hard coded variations of our scripts and make more than 20 or 30 submissions based on different hyperparametes and choose the one which sounds the best based on cv and lb scores.
", "votes": 2, "replies": [] } ] }, { "id": 665305, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2019-11-04T21:30:05.917000", "content": "@appian I have the same issue when tried to download to check that was a correct file, but it just jumps to up of the page without downloading.
", "votes": 2, "replies": [ { "id": 665309, "author_name": "Appian", "author_url": "", "post_date": "2019-11-04T21:33:16.127000", "content": "Exactly. This makes me worried because I can't make sure upload was successful or not.
", "votes": 1, "replies": [] } ] }, { "id": 665137, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2019-11-04T17:08:21.137000", "content": "\n\nSame question from me. I don't understand this part
\n\n\"In order to be eligible for Stage 2, each team's Stage 1 submission must include the model uploaded, via Team -> Your Model. This model should match that which will be used to generate the 2 final submissions you will select for scoring in Stage 2. Be aware that if you do not select 2 final submissions (via 'My Submissions'), the platform will auto-select your best-scoring model on the public leaderboard. The deadline for model upload is firmly the end of Stage 1.\"
\n\nCould you write more explicitly whether we do selection of 2 best our models now and zip it or we add all we wish (4-5 for example) and do selection at stage 2 (as pointed by @vadiksadik above in part 2) )??
", "votes": 2, "replies": [ { "id": 665146, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-04T17:14:02.883000", "content": "@muhakabartay You should be locking in your 2 best models that you intend to rerun yourself to generate your stage 2 submissions.
", "votes": 1, "replies": [] }, { "id": 665181, "author_name": "", "author_url": "", "post_date": "2019-11-04T17:46:31.413000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "664952": "@juliaelliott \n1) From https://www.kaggle.com/two-stage-frequently-asked-questions for big files\n> you can instead upload a checksum of your archive file (such as an md5 or sha hash)\n\nQ: Can I calculate this hash (e.g. md5sum final_package.zip), save it to a txt file, and upload this txt file?\n\n___\n\n2) From https://www.kaggle.com/two-stage-frequently-asked-questions\n> you may be permitted to select two submissions for final scoring. If so, don't forget to include the code/instructions for reproducing both! It can be totally different code, or it can be the same code with instructions about the modifications you would make to generate each.\n\nQ: Does it mean that I can put **all** models and code into final_package.zip, including training instruction, and add **several** (e.g. `N>2`) instructions for test submission, each of them will be with different test parameters (e.g. *instructionA* - *\"use modelA for submission\"*, instructionB - *\"use modelB for submission\"*), and during 2nd stage I will made **N** submissions for all instructions and select **two** of them for a final submits? Is it ok?", "666113": "I have a question on the kind of retraining that's allowed.\nWhen I was training during stage 1, I saved several model checkpoints and afterwards used the one with the best cv score as my final model.\n\nI didn't know we'd be allowed to retrain in stage 2 so I didn't automate the step of selecting the final model, neither did I include a text description (although it's clear which one I took if you look at the checkpoints and the final model which are all included in my upload).\n\nI would be able to retrain my pipeline on stage 2 without any changes, but I would still need to manually select the best scoring model from my checkpoints. Would that violate the stage 2 rules?\n\nBest,\nSzymon", "665733": "Do anyone have absecence of \"Submit predictions\" button in stage 2?", "665298": "@juliaelliott\n\nI just tried to download the uploaded zip archive to make sure everything is fine once again but unable to download it. I clicked download button but it does not start downloading. I have no idea whether this is an expected behavior for 2 stage competition or a bug. ", "665144": "@vadiksadik Yes to both questions.\n", "665142": "@juliaelliott - adding to the above wonderful question, I have one more question. There are 3 models which are performing almost equal in stage 1. Can we upload 3 all codes and then select 2 from those 3 as the best submissions for stage 2. Your inputs wil be highly appreciated ...", "665979": "@juliaelliott\nCould you clarify what kind of changes we can make with kernels in stage 2?\n\nA) We can change path to new data? Allowed? \nB) We can change for example epochs=5 to epochs=10, or lr=0.01 to lr=0.1, or smth like that without new lines of code but change of variables. Allowed?\nC) More than in B) and if allowed, so what exactly.\n\nThe reason I am asking, you said we can make several submissions and choose that best one which obviously mean that some changes allowed (otherwise, why I need to submit exact same code few times 🤔). \n\nThank you very much for clarification in advance.\n\nSorry for disturbing so many times. \n\nBest, \nMukharbek \n\n\n\n", "665305": "@appian I have the same issue when tried to download to check that was a correct file, but it just jumps to up of the page without downloading. ", "665137": "@juliaelliott \n\nSame question from me. I don't understand this part\n\n\"In order to be eligible for Stage 2, each team's Stage 1 submission must include the model uploaded, via Team -> Your Model. This model should match that which will be used to generate the 2 final submissions you will select for scoring in Stage 2. Be aware that if you do not select 2 final submissions (via 'My Submissions'), the platform will auto-select your best-scoring model on the public leaderboard. The deadline for model upload is firmly the end of Stage 1.\"\n\nCould you write more explicitly whether we do selection of 2 best our models now and zip it or we add all we wish (4-5 for example) and do selection at stage 2 (as pointed by @vadiksadik above in part 2) )??\n" } }