{ "id": 115976, "title": "How to prob the 2nd stage LB with just 2 submissions? ", "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/115976", "author_name": "SeuTao", "post_date": "2019-11-06T10:58:04.520000", "votes": 8, "comment_count": 6, "views": 0, "content": "

\"\"\nJust binary the probs?

", "messages": [ { "id": 666777, "postDate": "2019-11-06T13:37:44.097Z", "content": "

all zeros: 16.353, all ones: 18.185 :D

", "rawMarkdown": "all zeros: 16.353, all ones: 18.185 :D ", "votes": 10, "replies": [ { "id": 666784, "postDate": "2019-11-06T13:50:13.233Z", "content": "

select all zeros as you final submission, - it seems to be better ;)

", "rawMarkdown": "select all zeros as you final submission, - it seems to be better ;)", "votes": 2 } ] }, { "id": 666667, "postDate": "2019-11-06T10:58:04.520Z", "content": "

\"\"\nJust binary the probs?

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1025985%2F807c7d2ca53e1d6706a3cc6f88c51cc5%2F2019-11-06%206.55.57.png?generation=1573037780817380&alt=media)\nJust binary the probs?", "votes": 8 }, { "id": 666689, "postDate": "2019-11-06T11:30:22.930Z", "content": "

This is cool.

\n\n

I think it is possible to use all the scores present currently on the LB to figure out the only possible number of ones. But then still need to \"binary the probs\" actually. Interesting.

", "rawMarkdown": "This is cool. \n\nI think it is possible to use all the scores present currently on the LB to figure out the only possible number of ones. But then still need to \"binary the probs\" actually. Interesting.", "votes": 1 }, { "id": 667671, "postDate": "2019-11-07T14:38:25.560Z", "content": "

sorry for the dumb question, but you guys sound aliens to me... what do you mean with \"binary the probs\" ?

", "rawMarkdown": "sorry for the dumb question, but you guys sound aliens to me... what do you mean with \"binary the probs\" ? ", "replies": [ { "id": 667674, "postDate": "2019-11-07T14:41:10.503Z", "content": "

He meant casting all probabilities greater than 0.5 to 1 and less than or equal to 0.5 to 0. that is what (sub.Label > 0.5).astype(float) does. nosound explained here. Or you can choose another threshold like 0.2 or 0.1. It seems 0.5 isn't working for some people.

", "rawMarkdown": "He meant casting all probabilities greater than 0.5 to 1 and less than or equal to 0.5 to 0. that is what (sub.Label > 0.5).astype(float) does. nosound explained [here](https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/116025). Or you can choose another threshold like 0.2 or 0.1. It seems 0.5 isn't working for some people.", "votes": 1 }, { "id": 667680, "postDate": "2019-11-07T14:46:39.837Z", "content": "

Many thanks for explaining this !

", "rawMarkdown": "Many thanks for explaining this ! " } ] } ], "comments": [ { "id": 666777, "author_name": "Oleg Yaroshevskiy", "author_url": "", "post_date": "2019-11-06T13:37:44.097000", "content": "

all zeros: 16.353, all ones: 18.185 :D

", "votes": 10, "replies": [ { "id": 666784, "author_name": "nosound", "author_url": "", "post_date": "2019-11-06T13:50:13.233000", "content": "

select all zeros as you final submission, - it seems to be better ;)

", "votes": 2, "replies": [] } ] }, { "id": 666689, "author_name": "nosound", "author_url": "", "post_date": "2019-11-06T11:30:22.930000", "content": "

This is cool.

\n\n

I think it is possible to use all the scores present currently on the LB to figure out the only possible number of ones. But then still need to \"binary the probs\" actually. Interesting.

", "votes": 1, "replies": [] }, { "id": 667671, "author_name": "Benjamin Dubreu", "author_url": "", "post_date": "2019-11-07T14:38:25.560000", "content": "

sorry for the dumb question, but you guys sound aliens to me... what do you mean with \"binary the probs\" ?

", "votes": 0, "replies": [ { "id": 667674, "author_name": "Tim Yee", "author_url": "", "post_date": "2019-11-07T14:41:10.503000", "content": "

He meant casting all probabilities greater than 0.5 to 1 and less than or equal to 0.5 to 0. that is what (sub.Label > 0.5).astype(float) does. nosound explained here. Or you can choose another threshold like 0.2 or 0.1. It seems 0.5 isn't working for some people.

", "votes": 1, "replies": [] }, { "id": 667680, "author_name": "Benjamin Dubreu", "author_url": "", "post_date": "2019-11-07T14:46:39.837000", "content": "

Many thanks for explaining this !

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "666777": "all zeros: 16.353, all ones: 18.185 :D ", "666667": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1025985%2F807c7d2ca53e1d6706a3cc6f88c51cc5%2F2019-11-06%206.55.57.png?generation=1573037780817380&alt=media)\nJust binary the probs?", "666689": "This is cool. \n\nI think it is possible to use all the scores present currently on the LB to figure out the only possible number of ones. But then still need to \"binary the probs\" actually. Interesting.", "667671": "sorry for the dumb question, but you guys sound aliens to me... what do you mean with \"binary the probs\" ? " } }