{ "id": 69675, "title": "Stage 2 has begun!", "url": "/competitions/rsna-pneumonia-detection-challenge/discussion/69675", "author_name": "Phil Culliton", "post_date": "2018-10-25T23:43:42.347000", "votes": 3, "comment_count": 19, "views": 0, "content": "

Welcome to stage 2 of the RSNA Pneumonia Detection Challenge!

\n\n

The data has changed. There is a new training set (including stage 1's test set) and a test set comprised of images that are new to the competition. The metric and file formats are the same. Feel free to check out the Data page to see what's different and download the new data.

\n\n

Please also check out the two-stage FAQ, which has some good information about how two-stage competitions work.

\n\n

Edited to add: PLEASE make sure to select a stage 2 submission before the stage ends.

", "messages": [ { "id": 410478, "postDate": "2018-10-26T05:41:58.043Z", "content": "

Why we're not able make submission?

", "rawMarkdown": "Why we're not able make submission? ", "votes": 13, "replies": [ { "id": 410868, "postDate": "2018-10-26T19:32:54.623Z", "content": "

That should be resolved - please let me know if you're still not able to make submissions!

", "rawMarkdown": "That should be resolved - please let me know if you're still not able to make submissions!" }, { "id": 411410, "postDate": "2018-10-28T04:48:43.923Z", "content": "

@Phil: Are we allowed to retrain the model using the stage 1 model that I previously submitted as initial layer/weights for stage 2 training? Thanks!

", "rawMarkdown": "@Phil: Are we allowed to retrain the model using the stage 1 model that I previously submitted as initial layer/weights for stage 2 training? Thanks!", "votes": 1 } ] }, { "id": 410512, "postDate": "2018-10-26T07:29:35.310Z", "content": "

Based on the earlier competition rules I thought there would be only a stage 2 test dataset...that's why we needed to lock down our code and model to prevent tweaking of any parameters....Am I the only one confused about that?

", "rawMarkdown": "Based on the earlier competition rules I thought there would be only a stage 2 test dataset...that's why we needed to lock down our code and model to prevent tweaking of any parameters....Am I the only one confused about that?\n", "votes": 9, "replies": [ { "id": 410590, "postDate": "2018-10-26T10:10:14.917Z", "content": "

No I was thinking the same. Being able to retrain the model and make multiple submissions for stage 2 kind of defeats the point of having 2 stages, doesn't it?

", "rawMarkdown": "No I was thinking the same. Being able to retrain the model and make multiple submissions for stage 2 kind of defeats the point of having 2 stages, doesn't it?" }, { "id": 410599, "postDate": "2018-10-26T10:25:56.380Z", "content": "

We have the same confusion！

", "rawMarkdown": "We have the same confusion！" }, { "id": 410638, "postDate": "2018-10-26T11:39:46.687Z", "content": "

Bu bare in mind that current LB is calculated on 1% of test set.

", "rawMarkdown": "Bu bare in mind that current LB is calculated on 1% of test set." }, { "id": 410680, "postDate": "2018-10-26T12:47:37.833Z", "content": "

Ah yes, right. So that's 30 images, it's not really possible to get good information on the quality of the model over so little amount of data.

", "rawMarkdown": "Ah yes, right. So that's 30 images, it's not really possible to get good information on the quality of the model over so little amount of data.", "votes": 1 } ] }, { "id": 410370, "postDate": "2018-10-25T23:43:42.347Z", "content": "

Welcome to stage 2 of the RSNA Pneumonia Detection Challenge!

\n\n

The data has changed. There is a new training set (including stage 1's test set) and a test set comprised of images that are new to the competition. The metric and file formats are the same. Feel free to check out the Data page to see what's different and download the new data.

\n\n

Please also check out the two-stage FAQ, which has some good information about how two-stage competitions work.

\n\n

Edited to add: PLEASE make sure to select a stage 2 submission before the stage ends.

", "rawMarkdown": "Welcome to stage 2 of the RSNA Pneumonia Detection Challenge!\n\nThe data has changed. There is a new training set (including stage 1's test set) and a test set comprised of images that are new to the competition. The metric and file formats are the same. Feel free to check out the [Data][1] page to see what's different and download the new data.\n\nPlease also check out the [two-stage FAQ][2], which has some good information about how two-stage competitions work.\n\nEdited to add: PLEASE make sure to select a stage 2 submission before the stage ends.\n\n\n [1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/data\n [2]: https://www.kaggle.com/two-stage-frequently-asked-questions", "votes": 3 }, { "id": 410651, "postDate": "2018-10-26T12:01:10.923Z", "content": "

Ok, if I understood corectly, we can retrain our models with the new data, BUT without changing the HYPERPARAMETERS used to train the previous models. So, for example, my team is using 4 models in the ensemble. We can retrain those 4 models with exatly the same hyperparameters and submit the CSV.

\n\n

Have I misundertood anything?

", "rawMarkdown": "Ok, if I understood corectly, we can retrain our models with the new data, BUT without changing the HYPERPARAMETERS used to train the previous models. So, for example, my team is using 4 models in the ensemble. We can retrain those 4 models with exatly the same hyperparameters and submit the CSV.\n\nHave I misundertood anything?", "votes": 1, "replies": [ { "id": 410869, "postDate": "2018-10-26T19:33:34.547Z", "content": "

Nope! You have it correct.

", "rawMarkdown": "Nope! You have it correct." } ] }, { "id": 414658, "postDate": "2018-11-03T09:59:09.863Z", "content": "

Any chance to reveal stage 2 labels?

", "rawMarkdown": "Any chance to reveal stage 2 labels? " }, { "id": 414174, "postDate": "2018-11-02T09:03:57.647Z", "content": "

There seems to be something weird going on with the thresholds for medal awards.

", "rawMarkdown": "There seems to be [something weird going on][1] with the thresholds for medal awards.\n\n [1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70210" }, { "id": 410801, "postDate": "2018-10-26T17:28:53.877Z", "content": "

I have questions.

", "rawMarkdown": "I have [questions][1].\n\n [1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/69745", "replies": [ { "id": 410838, "postDate": "2018-10-26T18:20:00.447Z", "content": "

Those could have been my questions ;-) Great List!

", "rawMarkdown": "Those could have been my questions ;-) Great List!" } ] }, { "id": 410448, "postDate": "2018-10-26T04:46:39.453Z", "content": "

As I understand we can retrain our models using stage 2 train data set. What about number od submissions available?\nShall that be just one or can we do more than that?

", "rawMarkdown": "As I understand we can retrain our models using stage 2 train data set. What about number od submissions available?\nShall that be just one or can we do more than that?", "replies": [ { "id": 410463, "postDate": "2018-10-26T05:18:44.730Z", "content": "

Yeah, I have the same questions.\nHow many submissions are allowed? How does that combine with Model Upload?\nIf there are several submissions, is it possible to choose?

", "rawMarkdown": "Yeah, I have the same questions.\nHow many submissions are allowed? How does that combine with Model Upload?\nIf there are several submissions, is it possible to choose?" }, { "id": 410505, "postDate": "2018-10-26T06:55:02.487Z", "content": "

Based on my experience in the \"Data Science Bowl 2018\", I would say yes we would need to choose final submission. And the number of submissions per day is the same as in stage-1.

", "rawMarkdown": "Based on my experience in the \"Data Science Bowl 2018\", I would say yes we would need to choose final submission. And the number of submissions per day is the same as in stage-1.", "votes": 1 }, { "id": 410867, "postDate": "2018-10-26T19:32:31.340Z", "content": "

You can make the normal number of submissions per day until stage 2 ends.

", "rawMarkdown": "You can make the normal number of submissions per day until stage 2 ends." } ] }, { "id": 410851, "postDate": "2018-10-26T18:53:05.750Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 410478, "author_name": "", "author_url": "", "post_date": "2018-10-26T05:41:58.043000", "content": "

Why we're not able make submission?

", "votes": 13, "replies": [ { "id": 410868, "author_name": "Phil Culliton", "author_url": "", "post_date": "2018-10-26T19:32:54.623000", "content": "

That should be resolved - please let me know if you're still not able to make submissions!

", "votes": 0, "replies": [] }, { "id": 411410, "author_name": "", "author_url": "", "post_date": "2018-10-28T04:48:43.923000", "content": "

@Phil: Are we allowed to retrain the model using the stage 1 model that I previously submitted as initial layer/weights for stage 2 training? Thanks!

", "votes": 1, "replies": [] } ] }, { "id": 410512, "author_name": "Robin Smits", "author_url": "", "post_date": "2018-10-26T07:29:35.310000", "content": "

Based on the earlier competition rules I thought there would be only a stage 2 test dataset...that's why we needed to lock down our code and model to prevent tweaking of any parameters....Am I the only one confused about that?

", "votes": 9, "replies": [ { "id": 410590, "author_name": "Romain Destenay", "author_url": "", "post_date": "2018-10-26T10:10:14.917000", "content": "

No I was thinking the same. Being able to retrain the model and make multiple submissions for stage 2 kind of defeats the point of having 2 stages, doesn't it?

", "votes": 0, "replies": [] }, { "id": 410599, "author_name": "Li di", "author_url": "", "post_date": "2018-10-26T10:25:56.380000", "content": "

We have the same confusion！

", "votes": 0, "replies": [] }, { "id": 410638, "author_name": "Michal Gdak", "author_url": "", "post_date": "2018-10-26T11:39:46.687000", "content": "

Bu bare in mind that current LB is calculated on 1% of test set.

", "votes": 0, "replies": [] }, { "id": 410680, "author_name": "Romain Destenay", "author_url": "", "post_date": "2018-10-26T12:47:37.833000", "content": "

Ah yes, right. So that's 30 images, it's not really possible to get good information on the quality of the model over so little amount of data.

", "votes": 1, "replies": [] } ] }, { "id": 410651, "author_name": "Fernando Camargo", "author_url": "", "post_date": "2018-10-26T12:01:10.923000", "content": "

Ok, if I understood corectly, we can retrain our models with the new data, BUT without changing the HYPERPARAMETERS used to train the previous models. So, for example, my team is using 4 models in the ensemble. We can retrain those 4 models with exatly the same hyperparameters and submit the CSV.

\n\n

Have I misundertood anything?

", "votes": 1, "replies": [ { "id": 410869, "author_name": "Phil Culliton", "author_url": "", "post_date": "2018-10-26T19:33:34.547000", "content": "

Nope! You have it correct.

", "votes": 0, "replies": [] } ] }, { "id": 414658, "author_name": "raddar", "author_url": "", "post_date": "2018-11-03T09:59:09.863000", "content": "

Any chance to reveal stage 2 labels?

", "votes": 0, "replies": [] }, { "id": 414174, "author_name": "Andy Harless", "author_url": "", "post_date": "2018-11-02T09:03:57.647000", "content": "

There seems to be something weird going on with the thresholds for medal awards.

", "votes": 0, "replies": [] }, { "id": 410801, "author_name": "Andy Harless", "author_url": "", "post_date": "2018-10-26T17:28:53.877000", "content": "

I have questions.

", "votes": 0, "replies": [ { "id": 410838, "author_name": "Robin Smits", "author_url": "", "post_date": "2018-10-26T18:20:00.447000", "content": "

Those could have been my questions ;-) Great List!

", "votes": 0, "replies": [] } ] }, { "id": 410448, "author_name": "Michal Gdak", "author_url": "", "post_date": "2018-10-26T04:46:39.453000", "content": "

As I understand we can retrain our models using stage 2 train data set. What about number od submissions available?\nShall that be just one or can we do more than that?

", "votes": 0, "replies": [ { "id": 410463, "author_name": "Sergey Zlobin", "author_url": "", "post_date": "2018-10-26T05:18:44.730000", "content": "

Yeah, I have the same questions.\nHow many submissions are allowed? How does that combine with Model Upload?\nIf there are several submissions, is it possible to choose?

", "votes": 0, "replies": [] }, { "id": 410505, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2018-10-26T06:55:02.487000", "content": "

Based on my experience in the \"Data Science Bowl 2018\", I would say yes we would need to choose final submission. And the number of submissions per day is the same as in stage-1.

", "votes": 1, "replies": [] }, { "id": 410867, "author_name": "Phil Culliton", "author_url": "", "post_date": "2018-10-26T19:32:31.340000", "content": "

You can make the normal number of submissions per day until stage 2 ends.

", "votes": 0, "replies": [] } ] }, { "id": 410851, "author_name": "", "author_url": "", "post_date": "2018-10-26T18:53:05.750000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "410478": "Why we're not able make submission? ", "410512": "Based on the earlier competition rules I thought there would be only a stage 2 test dataset...that's why we needed to lock down our code and model to prevent tweaking of any parameters....Am I the only one confused about that?\n", "410370": "Welcome to stage 2 of the RSNA Pneumonia Detection Challenge!\n\nThe data has changed. There is a new training set (including stage 1's test set) and a test set comprised of images that are new to the competition. The metric and file formats are the same. Feel free to check out the [Data][1] page to see what's different and download the new data.\n\nPlease also check out the [two-stage FAQ][2], which has some good information about how two-stage competitions work.\n\nEdited to add: PLEASE make sure to select a stage 2 submission before the stage ends.\n\n\n [1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/data\n [2]: https://www.kaggle.com/two-stage-frequently-asked-questions", "410651": "Ok, if I understood corectly, we can retrain our models with the new data, BUT without changing the HYPERPARAMETERS used to train the previous models. So, for example, my team is using 4 models in the ensemble. We can retrain those 4 models with exatly the same hyperparameters and submit the CSV.\n\nHave I misundertood anything?", "414658": "Any chance to reveal stage 2 labels? ", "414174": "There seems to be [something weird going on][1] with the thresholds for medal awards.\n\n [1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/70210", "410801": "I have [questions][1].\n\n [1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/discussion/69745", "410448": "As I understand we can retrain our models using stage 2 train data set. What about number od submissions available?\nShall that be just one or can we do more than that?", "410851": "" } }