I am new to stage competitions\nif some one can explain me few things below\n1) Is stage 2 data score is private lb score ?\n2) How stage 2 score is computed ,is it through the model that we upload ?
", "messages": [ { "id": 660543, "postDate": "2019-10-29T09:52:09.423Z", "content": "I am new to stage competitions\nif some one can explain me few things below\n1) Is stage 2 data score is private lb score ?\n2) How stage 2 score is computed ,is it through the model that we upload ?
", "rawMarkdown": "I am new to stage competitions\nif some one can explain me few things below\n1) Is stage 2 data score is private lb score ?\n2) How stage 2 score is computed ,is it through the model that we upload ?", "votes": 6 }, { "id": 660789, "postDate": "2019-10-29T15:57:10.477Z", "content": "1) Yes.\n2) The stage 2 score is computed based on the predictions you upload in stage 2, based on the stage 2 dataset that is provided in 1 week's time by November 5th. This is not a code competition, so we are not automatically running your code. However, you must upload the model(s) you will use in stage 2 by the end of stage 1 in order to be eligible to remain on the final leaderboard.
", "rawMarkdown": "1) Yes.\n2) The stage 2 score is computed based on the predictions you upload in stage 2, based on the stage 2 dataset that is provided in [1 week's time](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/timeline) by November 5th. This is not a code competition, so we are not automatically running your code. However, you must upload the model(s) you will use in stage 2 by the end of stage 1 in order to be eligible to remain on the final leaderboard.", "votes": 4, "replies": [ { "id": 661305, "postDate": "2019-10-30T05:37:06.210Z", "content": "@juliat thanks\n1) So i should upload the model wts used in stage1 ?\n2) is it ok to further train the model after we upload in stage 1 . For example by end of stage1 my models that i m going to use for stage 2 are not yet well trained enough to produce best result. SO can i still train in a week time to get best result ?
", "rawMarkdown": "@juliat thanks\n1) So i should upload the model wts used in stage1 ?\n2) is it ok to further train the model after we upload in stage 1 . For example by end of stage1 my models that i m going to use for stage 2 are not yet well trained enough to produce best result. SO can i still train in a week time to get best result ?\n", "votes": -1 }, { "id": 661531, "postDate": "2019-10-30T12:39:33.763Z", "content": "Sorry if this is outlined or somewhere else.
\n\nBut I am confused as to whether or not we are allowed to change any of our code besides data paths after stage 1. Like can I refine my ensembling methods with the models that are uploaded. Or is the competition virtually over once stage 1 ends except for running our uploaded models on the new data?
", "rawMarkdown": "Sorry if this is outlined or somewhere else. \n\nBut I am confused as to whether or not we are allowed to change any of our code besides data paths after stage 1. Like can I refine my ensembling methods with the models that are uploaded. Or is the competition virtually over once stage 1 ends except for running our uploaded models on the new data?" }, { "id": 661622, "postDate": "2019-10-30T14:17:58.413Z", "content": "Can you please clarify for us whether stage1 labels will be released and whether it's allowed to fine-tune our models on them to make predictions on stage2? Thanks.
\n\nUPD: Yes we can: https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/112145#latest-660134
", "rawMarkdown": "Can you please clarify for us whether stage1 labels will be released and whether it's allowed to fine-tune our models on them to make predictions on stage2? Thanks.\n\n**UPD**: Yes we can: https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/112145#latest-660134\n" }, { "id": 661979, "postDate": "2019-10-31T00:11:32.250Z", "content": "@jaideepvalani - Review the two-stage FAQ. You may re-train your model on stage 2's updated train set (that will add stage 1's test set), but you should not be making any code changes or manual parameter tuning.
\n\n@governor - No, you cannot change any of your code after stage 1.
\n\n@bloodaxe - The stage 1 test set labels will be released. You are permitted to retrain on this additional data, but you cannot fine-tune them manually.
", "rawMarkdown": "@jaideepvalani - Review the [two-stage FAQ](https://www.kaggle.com/two-stage-frequently-asked-questions). You may re-train your model on stage 2's updated train set (that will add stage 1's test set), but you should not be making any code changes or manual parameter tuning.\n\n@governor - No, you cannot change any of your code after stage 1.\n\n@bloodaxe - The stage 1 test set labels will be released. You are permitted to retrain on this additional data, but you cannot fine-tune them manually." } ] }, { "id": 663571, "postDate": "2019-11-02T10:32:33.777Z", "content": "So, to get things straight (please excuse me, it is my first time in 2 stage contest) :\n- Stage 2 will release a train set and a test set. \n- We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\n- To be eligible, we need to upload the code for stage 1
\n\ncan we upload a .ipynb file to fulfill this requirement?
", "rawMarkdown": "So, to get things straight (please excuse me, it is my first time in 2 stage contest) :\n- Stage 2 will release a train set and a test set. \n- We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\n- To be eligible, we need to upload the code for stage 1\n\ncan we upload a .ipynb file to fulfill this requirement?", "votes": 2, "replies": [ { "id": 663728, "postDate": "2019-11-02T15:44:16.307Z", "content": "@dcstang Yes, all correct. If an .ipynb is what generates your CSV submission, then yes, that is adequate for your upload.
", "rawMarkdown": "@dcstang Yes, all correct. If an .ipynb is what generates your CSV submission, then yes, that is adequate for your upload.", "votes": 1 }, { "id": 663889, "postDate": "2019-11-02T20:08:59.013Z", "content": "Thank you for the reply and clarification!
", "rawMarkdown": "Thank you for the reply and clarification! ", "votes": 1 }, { "id": 664691, "postDate": "2019-11-04T04:20:07.193Z", "content": "@dcstang @juliaelliott \nPlease correct me if I am wrong:\n\"We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\"\n=====> You may retrain your model again using new training set for stage2 (stage_1_train + stage_1_test set)
", "rawMarkdown": "@dcstang @juliaelliott \nPlease correct me if I am wrong:\n\"We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\"\n=====> **You may retrain your model again using new training set for stage2 (stage_1_train + stage_1_test set)**" } ] }, { "id": 660909, "postDate": "2019-10-29T19:09:59.090Z", "content": "Thanks for asking this: So basically we need to upload a Pytorch model.pth or Keras model.h5 before the end of stage 1, or November 5.
\n\nThe two-stage FAQs from Kaggle, referenced in the RSNA reminder email is then, inaccurate to mention that we need to upload all the code to generate the submission csv file.\nThis is what is written in the FAQs:\n\nWhat should I upload?\nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those.\n
@dcstang - The two-stage FAQ is accurate. What you've copied & pasted there is what is expected of your stage 1 upload.
\n\nIf the entirety of your code to generate the competition's .CSV predictions is represented in a .pth or .h5, then sure, that is sufficient to upload as your model.
\n\nThe end of stage 1 is November 4th @ 11:59pm UTC. See the timeline page.
", "rawMarkdown": "@dcstang - The two-stage FAQ **is** accurate. What you've copied & pasted there is what is expected of your stage 1 upload. \n\nIf the entirety of your code to generate the competition's .CSV predictions is represented in a .pth or .h5, then sure, that is sufficient to upload as your model. \n\nThe end of stage 1 is November 4th @ 11:59pm UTC. See the [timeline page](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/timeline).", "votes": 1 }, { "id": 662335, "postDate": "2019-10-31T12:45:12.177Z", "content": "@juliaelliott \nwhat if we have to use different kernel to generate csv file using the weights from other kernel where we train the model.\nThese small small things may not be there in faqs so many people who are new would end up getting delisted from lb because of not fully complying with the stage 2 rules.
", "rawMarkdown": "@juliaelliott \nwhat if we have to use different kernel to generate csv file using the weights from other kernel where we train the model.\nThese small small things may not be there in faqs so many people who are new would end up getting delisted from lb because of not fully complying with the stage 2 rules.\n\n" }, { "id": 662540, "postDate": "2019-10-31T17:04:57.303Z", "content": "@jaideepvalani That's fine - just be sure to upload all of the code required to generate your stage 1 predictions.
", "rawMarkdown": "@jaideepvalani That's fine - just be sure to upload all of the code required to generate your stage 1 predictions." }, { "id": 662618, "postDate": "2019-10-31T19:06:38.437Z", "content": "Ok thanks for clarifying that @juliaelliott!
", "rawMarkdown": "Ok thanks for clarifying that @juliaelliott!" }, { "id": 663026, "postDate": "2019-11-01T10:58:06.387Z", "content": "@juliaelliott
\n1) will there be stage-2-train set also or only stage 2 test \n2) Is there possible overlap of patient id between any train and test ,which i think is unlikely
The stage 2 train set will be a combination of the stage 1 train set and the stage 1 test set (which we will provide labels for). You may retrain on this in stage 2, and will need to predict on the stage 2 test set. Otherwise, no additional training data.
\n\nThere will be no patient overlap with the stage 2 test set.
", "rawMarkdown": "@jaideepvalani \n\nThe stage 2 train set will be a combination of the stage 1 train set and the stage 1 test set (which we will provide labels for). You may retrain on this in stage 2, and will need to predict on the stage 2 test set. Otherwise, no additional training data.\n\nThere will be no patient overlap with the stage 2 test set." }, { "id": 663235, "postDate": "2019-11-01T16:44:19.290Z", "content": "Thanks..\n@juliaelliott sorry to keep you engaged..\nPLB score will be based on stage 2 predictions or it would be only private lb at the end ?
", "rawMarkdown": "Thanks..\n@juliaelliott sorry to keep you engaged..\nPLB score will be based on stage 2 predictions or it would be only private lb at the end ?" }, { "id": 663275, "postDate": "2019-11-01T18:16:12.127Z", "content": "@jaideepvalani - Private leaderboard and final ranking is based only on the stage 2 test set.
", "rawMarkdown": "@jaideepvalani - Private leaderboard and final ranking is based *only* on the stage 2 test set." }, { "id": 663491, "postDate": "2019-11-02T07:05:26.227Z", "content": "okie .. @juliaelliott \nfine\n1) Who is going to refer uploaded code and model.. is it for check by host team for comparison? or it is going to be accessible by us to run the stage 2 \n2) Do you consider below as a change in code \n a) People are using meta data stage 1 train set which is saved as dataset ,in stage 2 we will have test images also of stage 1 ,to include the those test images a change will be needed to kernel in order to retrain\n b) If suppose there is inclusion of additional augmentation on the top of what was uploaded,is this a change ?
", "rawMarkdown": "okie .. @juliaelliott \nfine\n1) Who is going to refer uploaded code and model.. is it for check by host team for comparison? or it is going to be accessible by us to run the stage 2 \n2) Do you consider below as a change in code \n a) People are using meta data stage 1 train set which is saved as dataset ,in stage 2 we will have test images also of stage 1 ,to include the those test images a change will be needed to kernel in order to retrain\n b) If suppose there is inclusion of additional augmentation on the top of what was uploaded,is this a change ?" }, { "id": 663737, "postDate": "2019-11-02T15:56:32.290Z", "content": "@jaideepvalani Please read my prior responses and the FAQ carefully before asking additional questions - these have all been answered already.\n1. Yes, the host will validate this. You will run your own code in stage 2 to make CSV submissions on the stage 2 test set.\n2. Augmentation is not allowed. Retraining on the additional stage 2 train data — stage 1 test set that becomes part of eligible training data in stage 2 — is allowed. Minor adjustments solely for the purpose of loading the updated train set are permitted. Otherwise, code changes are not permitted.
", "rawMarkdown": "@jaideepvalani Please read my prior responses and the FAQ carefully before asking additional questions - these have all been answered already.\n1. Yes, the host will validate this. You will run your own code in stage 2 to make CSV submissions on the stage 2 test set.\n2. Augmentation is **not allowed**. Retraining on the additional stage 2 train data — stage 1 test set that becomes part of eligible training data in stage 2 — is allowed. Minor adjustments solely for the purpose of loading the updated train set are permitted. Otherwise, code changes are **not permitted**. " } ] }, { "id": 663724, "postDate": "2019-11-02T15:38:44.723Z", "content": "@juliaelliott \nI also have some questions.\n1. How many times can we submit on stage2?\n2. Can we see scores of our submissions during stage2 week?\n3. Can we choose 2 submissions from our several submissions?
", "rawMarkdown": "@juliaelliott \nI also have some questions.\n1. How many times can we submit on stage2?\n2. Can we see scores of our submissions during stage2 week?\n3. Can we choose 2 submissions from our several submissions?", "replies": [ { "id": 663730, "postDate": "2019-11-02T15:49:31.387Z", "content": "@shirakia \n1. Same 5 daily submissions max.\n2. No. A very small portion of the stage 2 test set will return a score on the public leaderboard, but this is purely intended to ensure the CSV submission does not return basic errors. But otherwise, the private score is completely hidden until the stage 2 final submission deadline.\n3. Yes, you need to select your 2 final submissions by the stage 2 deadline.
", "rawMarkdown": "@shirakia \n1. Same 5 daily submissions max.\n2. No. A very small portion of the stage 2 test set will return a score on the public leaderboard, but this is purely intended to ensure the CSV submission does not return basic errors. But otherwise, the private score is completely hidden until the stage 2 final submission deadline.\n3. Yes, you need to select your 2 final submissions by the stage 2 deadline.", "votes": 1 }, { "id": 663732, "postDate": "2019-11-02T15:51:55.570Z", "content": "Thank you for a very quick reply! All clear to me.
", "rawMarkdown": "Thank you for a very quick reply! All clear to me." } ] }, { "id": 661500, "postDate": "2019-10-30T11:55:50.630Z", "content": "How many images are on stage 2?
", "rawMarkdown": "How many images are on stage 2?", "replies": [ { "id": 661980, "postDate": "2019-10-31T00:11:58.977Z", "content": "The stage 2 test set is about 27 GB compressed.
", "rawMarkdown": "The stage 2 test set is about 27 GB compressed.", "votes": 4 }, { "id": 662080, "postDate": "2019-10-31T04:17:45.933Z", "content": "@juliaelliott Thanks for you reply!!
", "rawMarkdown": "@juliaelliott Thanks for you reply!!" } ] } ], "comments": [ { "id": 660789, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-10-29T15:57:10.477000", "content": "1) Yes.\n2) The stage 2 score is computed based on the predictions you upload in stage 2, based on the stage 2 dataset that is provided in 1 week's time by November 5th. This is not a code competition, so we are not automatically running your code. However, you must upload the model(s) you will use in stage 2 by the end of stage 1 in order to be eligible to remain on the final leaderboard.
", "votes": 4, "replies": [ { "id": 661305, "author_name": "Jaideep", "author_url": "", "post_date": "2019-10-30T05:37:06.210000", "content": "@juliat thanks\n1) So i should upload the model wts used in stage1 ?\n2) is it ok to further train the model after we upload in stage 1 . For example by end of stage1 my models that i m going to use for stage 2 are not yet well trained enough to produce best result. SO can i still train in a week time to get best result ?
", "votes": -1, "replies": [] }, { "id": 661531, "author_name": "bright", "author_url": "", "post_date": "2019-10-30T12:39:33.763000", "content": "Sorry if this is outlined or somewhere else.
\n\nBut I am confused as to whether or not we are allowed to change any of our code besides data paths after stage 1. Like can I refine my ensembling methods with the models that are uploaded. Or is the competition virtually over once stage 1 ends except for running our uploaded models on the new data?
", "votes": 0, "replies": [] }, { "id": 661622, "author_name": "Eugene Khvedchenya", "author_url": "", "post_date": "2019-10-30T14:17:58.413000", "content": "Can you please clarify for us whether stage1 labels will be released and whether it's allowed to fine-tune our models on them to make predictions on stage2? Thanks.
\n\nUPD: Yes we can: https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/112145#latest-660134
", "votes": 0, "replies": [] }, { "id": 661979, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-10-31T00:11:32.250000", "content": "@jaideepvalani - Review the two-stage FAQ. You may re-train your model on stage 2's updated train set (that will add stage 1's test set), but you should not be making any code changes or manual parameter tuning.
\n\n@governor - No, you cannot change any of your code after stage 1.
\n\n@bloodaxe - The stage 1 test set labels will be released. You are permitted to retrain on this additional data, but you cannot fine-tune them manually.
", "votes": 0, "replies": [] } ] }, { "id": 663571, "author_name": "David Tang", "author_url": "", "post_date": "2019-11-02T10:32:33.777000", "content": "So, to get things straight (please excuse me, it is my first time in 2 stage contest) :\n- Stage 2 will release a train set and a test set. \n- We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\n- To be eligible, we need to upload the code for stage 1
\n\ncan we upload a .ipynb file to fulfill this requirement?
", "votes": 2, "replies": [ { "id": 663728, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-02T15:44:16.307000", "content": "@dcstang Yes, all correct. If an .ipynb is what generates your CSV submission, then yes, that is adequate for your upload.
", "votes": 1, "replies": [] }, { "id": 663889, "author_name": "David Tang", "author_url": "", "post_date": "2019-11-02T20:08:59.013000", "content": "Thank you for the reply and clarification!
", "votes": 1, "replies": [] }, { "id": 664691, "author_name": "Mohammad Azam Khan", "author_url": "", "post_date": "2019-11-04T04:20:07.193000", "content": "@dcstang @juliaelliott \nPlease correct me if I am wrong:\n\"We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\"\n=====> You may retrain your model again using new training set for stage2 (stage_1_train + stage_1_test set)
", "votes": 0, "replies": [] } ] }, { "id": 660909, "author_name": "David Tang", "author_url": "", "post_date": "2019-10-29T19:09:59.090000", "content": "Thanks for asking this: So basically we need to upload a Pytorch model.pth or Keras model.h5 before the end of stage 1, or November 5.
\n\nThe two-stage FAQs from Kaggle, referenced in the RSNA reminder email is then, inaccurate to mention that we need to upload all the code to generate the submission csv file.\nThis is what is written in the FAQs:\n\nWhat should I upload?\nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those.\n
@dcstang - The two-stage FAQ is accurate. What you've copied & pasted there is what is expected of your stage 1 upload.
\n\nIf the entirety of your code to generate the competition's .CSV predictions is represented in a .pth or .h5, then sure, that is sufficient to upload as your model.
\n\nThe end of stage 1 is November 4th @ 11:59pm UTC. See the timeline page.
", "votes": 1, "replies": [] }, { "id": 662335, "author_name": "Jaideep", "author_url": "", "post_date": "2019-10-31T12:45:12.177000", "content": "@juliaelliott \nwhat if we have to use different kernel to generate csv file using the weights from other kernel where we train the model.\nThese small small things may not be there in faqs so many people who are new would end up getting delisted from lb because of not fully complying with the stage 2 rules.
", "votes": 0, "replies": [] }, { "id": 662540, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-10-31T17:04:57.303000", "content": "@jaideepvalani That's fine - just be sure to upload all of the code required to generate your stage 1 predictions.
", "votes": 0, "replies": [] }, { "id": 662618, "author_name": "David Tang", "author_url": "", "post_date": "2019-10-31T19:06:38.437000", "content": "Ok thanks for clarifying that @juliaelliott!
", "votes": 0, "replies": [] }, { "id": 663026, "author_name": "Jaideep", "author_url": "", "post_date": "2019-11-01T10:58:06.387000", "content": "@juliaelliott
\n1) will there be stage-2-train set also or only stage 2 test \n2) Is there possible overlap of patient id between any train and test ,which i think is unlikely
The stage 2 train set will be a combination of the stage 1 train set and the stage 1 test set (which we will provide labels for). You may retrain on this in stage 2, and will need to predict on the stage 2 test set. Otherwise, no additional training data.
\n\nThere will be no patient overlap with the stage 2 test set.
", "votes": 0, "replies": [] }, { "id": 663235, "author_name": "Jaideep", "author_url": "", "post_date": "2019-11-01T16:44:19.290000", "content": "Thanks..\n@juliaelliott sorry to keep you engaged..\nPLB score will be based on stage 2 predictions or it would be only private lb at the end ?
", "votes": 0, "replies": [] }, { "id": 663275, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-01T18:16:12.127000", "content": "@jaideepvalani - Private leaderboard and final ranking is based only on the stage 2 test set.
", "votes": 0, "replies": [] }, { "id": 663491, "author_name": "Jaideep", "author_url": "", "post_date": "2019-11-02T07:05:26.227000", "content": "okie .. @juliaelliott \nfine\n1) Who is going to refer uploaded code and model.. is it for check by host team for comparison? or it is going to be accessible by us to run the stage 2 \n2) Do you consider below as a change in code \n a) People are using meta data stage 1 train set which is saved as dataset ,in stage 2 we will have test images also of stage 1 ,to include the those test images a change will be needed to kernel in order to retrain\n b) If suppose there is inclusion of additional augmentation on the top of what was uploaded,is this a change ?
", "votes": 0, "replies": [] }, { "id": 663737, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-02T15:56:32.290000", "content": "@jaideepvalani Please read my prior responses and the FAQ carefully before asking additional questions - these have all been answered already.\n1. Yes, the host will validate this. You will run your own code in stage 2 to make CSV submissions on the stage 2 test set.\n2. Augmentation is not allowed. Retraining on the additional stage 2 train data — stage 1 test set that becomes part of eligible training data in stage 2 — is allowed. Minor adjustments solely for the purpose of loading the updated train set are permitted. Otherwise, code changes are not permitted.
", "votes": 0, "replies": [] } ] }, { "id": 663724, "author_name": "shirakia", "author_url": "", "post_date": "2019-11-02T15:38:44.723000", "content": "@juliaelliott \nI also have some questions.\n1. How many times can we submit on stage2?\n2. Can we see scores of our submissions during stage2 week?\n3. Can we choose 2 submissions from our several submissions?
", "votes": 0, "replies": [ { "id": 663730, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-11-02T15:49:31.387000", "content": "@shirakia \n1. Same 5 daily submissions max.\n2. No. A very small portion of the stage 2 test set will return a score on the public leaderboard, but this is purely intended to ensure the CSV submission does not return basic errors. But otherwise, the private score is completely hidden until the stage 2 final submission deadline.\n3. Yes, you need to select your 2 final submissions by the stage 2 deadline.
", "votes": 1, "replies": [] }, { "id": 663732, "author_name": "shirakia", "author_url": "", "post_date": "2019-11-02T15:51:55.570000", "content": "Thank you for a very quick reply! All clear to me.
", "votes": 0, "replies": [] } ] }, { "id": 661500, "author_name": "kambarakun", "author_url": "", "post_date": "2019-10-30T11:55:50.630000", "content": "How many images are on stage 2?
", "votes": 0, "replies": [ { "id": 661980, "author_name": "Julia Elliott", "author_url": "", "post_date": "2019-10-31T00:11:58.977000", "content": "The stage 2 test set is about 27 GB compressed.
", "votes": 4, "replies": [] }, { "id": 662080, "author_name": "kambarakun", "author_url": "", "post_date": "2019-10-31T04:17:45.933000", "content": "@juliaelliott Thanks for you reply!!
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "660543": "I am new to stage competitions\nif some one can explain me few things below\n1) Is stage 2 data score is private lb score ?\n2) How stage 2 score is computed ,is it through the model that we upload ?", "660789": "1) Yes.\n2) The stage 2 score is computed based on the predictions you upload in stage 2, based on the stage 2 dataset that is provided in [1 week's time](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview/timeline) by November 5th. This is not a code competition, so we are not automatically running your code. However, you must upload the model(s) you will use in stage 2 by the end of stage 1 in order to be eligible to remain on the final leaderboard.", "663571": "So, to get things straight (please excuse me, it is my first time in 2 stage contest) :\n- Stage 2 will release a train set and a test set. \n- We will need to run an inference for the stage 2 test set (using the previous code) and make another 'submission.csv'.\n- To be eligible, we need to upload the code for stage 1\n\ncan we upload a .ipynb file to fulfill this requirement?", "660909": "Thanks for asking this: So basically we need to upload a **Pytorch model.pth** or Keras model.h5 before the end of stage 1, or November 5.\n\nThe [two-stage FAQs from Kaggle, referenced in the RSNA reminder email](https://www.kaggle.com/two-stage-frequently-asked-questions) is then, inaccurate to mention that we need to upload all the code to generate the submission csv file.\nThis is what is written in the FAQs:\n```\nWhat should I upload?\nWhen you upload a model, you pack all the code that you are eventually going to use to generate your submission csv file. If your models generate some output files containing the weights, for example, ‘.caffemodel’ or ‘.tfmodel’ files, you are NOT required to submit those.\n```", "663724": "@juliaelliott \nI also have some questions.\n1. How many times can we submit on stage2?\n2. Can we see scores of our submissions during stage2 week?\n3. Can we choose 2 submissions from our several submissions?", "661500": "How many images are on stage 2?" } }