How are you planning to select your final submissions ? Highest LB or highest CV ? Or any other strategies ? Also, what do you think about shakeups ? Could it be huge or there would be a nice correlation between private and public scores ?
", "messages": [ { "id": 939007, "postDate": "2020-07-22T00:43:53.417Z", "content": "Yes this very import decision. Based on my past competition experience I use this and this method. Good luck =)
", "rawMarkdown": "Yes this very import decision. Based on my past competition experience I use [this](https://www.youtube.com/watch?v=oHg5SJYRHA0) and [this](https://www.youtube.com/watch?v=oHg5SJYRHA0) method. Good luck =) ", "votes": 10, "replies": [ { "id": 939018, "postDate": "2020-07-22T01:13:29.520Z", "content": "Ooh, I was completely wrong with. this strategy then :p
", "rawMarkdown": "Ooh, I was completely wrong with. [this](https://m.youtube.com/watch?v=eVTXPUF4Oz4) strategy then :p ", "votes": 2 }, { "id": 939155, "postDate": "2020-07-22T04:12:22.663Z", "content": "this may be because all your score could be more or less around each other :)
\ngood look with money and gold :)
I never considered either of those approaches.
", "rawMarkdown": "I never considered either of those approaches. ", "votes": 1 } ] }, { "id": 938993, "postDate": "2020-07-22T00:00:58.867Z", "content": "How are you planning to select your final submissions ? Highest LB or highest CV ? Or any other strategies ? Also, what do you think about shakeups ? Could it be huge or there would be a nice correlation between private and public scores ?
", "rawMarkdown": "How are you planning to select your final submissions ? Highest LB or highest CV ? Or any other strategies ? Also, what do you think about shakeups ? Could it be huge or there would be a nice correlation between private and public scores ? ", "votes": 5 }, { "id": 939922, "postDate": "2020-07-22T15:08:41.043Z", "content": "I usually pick 1 that overfits public LB and a more conservative one based on CV, public LB, and my hypotheses about private LB and CV/LB similarity.
", "rawMarkdown": "I usually pick 1 that overfits public LB and a more conservative one based on CV, public LB, and my hypotheses about private LB and CV/LB similarity.", "votes": 4 }, { "id": 939375, "postDate": "2020-07-22T07:38:56.857Z", "content": "I guess we can expect a serious shakeup because the sample size of the test set is quite small. At the same time, the test set images are graded by multiple pathologists, which makes their labels more robust compared to the training ones. Will see soon which effect prevails!
", "rawMarkdown": "I guess we can expect a serious shakeup because the sample size of the test set is quite small. At the same time, the test set images are graded by multiple pathologists, which makes their labels more robust compared to the training ones. Will see soon which effect prevails!", "votes": 1 } ], "comments": [ { "id": 939007, "author_name": "DrHB", "author_url": "", "post_date": "2020-07-22T00:43:53.417000", "content": "Yes this very import decision. Based on my past competition experience I use this and this method. Good luck =)
", "votes": 10, "replies": [ { "id": 939018, "author_name": "Indranil Bhattacharya", "author_url": "", "post_date": "2020-07-22T01:13:29.520000", "content": "Ooh, I was completely wrong with. this strategy then :p
", "votes": 2, "replies": [] }, { "id": 939155, "author_name": "Jaideep", "author_url": "", "post_date": "2020-07-22T04:12:22.663000", "content": "this may be because all your score could be more or less around each other :)
\ngood look with money and gold :)
I never considered either of those approaches.
", "votes": 1, "replies": [] } ] }, { "id": 939922, "author_name": "Ian Pan", "author_url": "", "post_date": "2020-07-22T15:08:41.043000", "content": "I usually pick 1 that overfits public LB and a more conservative one based on CV, public LB, and my hypotheses about private LB and CV/LB similarity.
", "votes": 4, "replies": [] }, { "id": 939375, "author_name": "Nikita Kozodoi", "author_url": "", "post_date": "2020-07-22T07:38:56.857000", "content": "I guess we can expect a serious shakeup because the sample size of the test set is quite small. At the same time, the test set images are graded by multiple pathologists, which makes their labels more robust compared to the training ones. Will see soon which effect prevails!
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "939007": "Yes this very import decision. Based on my past competition experience I use [this](https://www.youtube.com/watch?v=oHg5SJYRHA0) and [this](https://www.youtube.com/watch?v=oHg5SJYRHA0) method. Good luck =) ", "938993": "How are you planning to select your final submissions ? Highest LB or highest CV ? Or any other strategies ? Also, what do you think about shakeups ? Could it be huge or there would be a nice correlation between private and public scores ? ", "939922": "I usually pick 1 that overfits public LB and a more conservative one based on CV, public LB, and my hypotheses about private LB and CV/LB similarity.", "939375": "I guess we can expect a serious shakeup because the sample size of the test set is quite small. At the same time, the test set images are graded by multiple pathologists, which makes their labels more robust compared to the training ones. Will see soon which effect prevails!" } }