How about you guys? ;)
", "messages": [ { "id": "542498", "postDate": "06/04/2019 00:29:54", "content": "How about you guys? ;)
", "rawMarkdown": "How about you guys? ;)", "votes": null }, { "id": "542510", "postDate": "06/04/2019 00:40:13", "content": "I submitted a variant of the straight-forward transformer-network from this public kernel (identical up to changing into my CV for sake of convenience).
\n\nI didn't bother researching this further, since it had terrible LB score and not-so-good CV score.
\n\nWould won a gold medal...
", "rawMarkdown": "I submitted a variant of the straight-forward transformer-network from [this public kernel](https://www.kaggle.com/buchan/transformer-network-with-1d-cnn-feature-extraction) (identical up to changing into my CV for sake of convenience).\n\nI didn't bother researching this further, since it had terrible LB score and not-so-good CV score.\n\nWould won a gold medal...", "votes": null }, { "id": "542520", "postDate": "06/04/2019 00:49:27", "content": "Same has happened for me. My first submission in this competition would have taken me 500 places up.
", "rawMarkdown": "Same has happened for me. My first submission in this competition would have taken me 500 places up.", "votes": null }, { "id": "542526", "postDate": "06/04/2019 00:54:49", "content": "My best was more 2 weeks ago. I wasted so much time last 2 weeks, although I didn't select my best submission.
", "rawMarkdown": "My best was more 2 weeks ago. I wasted so much time last 2 weeks, although I didn't select my best submission.", "votes": null }, { "id": "542530", "postDate": "06/04/2019 00:56:53", "content": "My first submission which only with several basic statistical features was the best (2.45179) in the Private LB ;)\nIts public LB score was the worst, though.
", "rawMarkdown": "My first submission which only with several basic statistical features was the best (2.45179) in the Private LB ;)\nIts public LB score was the worst, though.", "votes": null }, { "id": "542531", "postDate": "06/04/2019 00:57:27", "content": "My Best One was one of the earliest submitted, but with public score 1.69472 I would never imagine that it is one to be chosen.
", "rawMarkdown": "My Best One was one of the earliest submitted, but with public score 1.69472 I would never imagine that it is one to be chosen.", "votes": null }, { "id": "542533", "postDate": "06/04/2019 00:59:38", "content": "Totally agree. The test set samples were not random. Public LB sample size was too small. Even the final test set size was too small. I actually compiled CNN_LSTM model very similar to that, but with less dense layers. I didn't do well with public leaderboard. I didn't continue work on it. It was my second best score on the P. LB.
", "rawMarkdown": "Totally agree. The test set samples were not random. Public LB sample size was too small. Even the final test set size was too small. I actually compiled CNN_LSTM model very similar to that, but with less dense layers. I didn't do well with public leaderboard. I didn't continue work on it. It was my second best score on the P. LB.", "votes": null }, { "id": "542574", "postDate": "06/04/2019 01:40:07", "content": "it happens because the mean of Public and Private LB are very different (4 and 6.7)
", "rawMarkdown": "it happens because the mean of Public and Private LB are very different (4 and 6.7)", "votes": null }, { "id": "542680", "postDate": "06/04/2019 03:27:00", "content": "My first day submission using only 1/3 data was 2.3847. However it's public score is 1.91754, so I didn't choose it......\nThis competition seems really weird.
", "rawMarkdown": "My first day submission using only 1/3 data was 2.3847. However it's public score is 1.91754, so I didn't choose it......\nThis competition seems really weird.", "votes": null }, { "id": "542712", "postDate": "06/04/2019 03:54:12", "content": "Private Score: 2.39760\nPublic Score: 1.63585\nAfter 5 folds test_CV = 2.029 | train_mean = 0.870 --> overfit IMHO.\nSo frustrating!!!
", "rawMarkdown": "Private Score: 2.39760\nPublic Score: 1.63585\nAfter 5 folds test_CV = 2.029 | train_mean = 0.870 --> overfit IMHO.\nSo frustrating!!!", "votes": null }, { "id": "542772", "postDate": "06/04/2019 04:47:26", "content": "How do you know Private LB mean?
", "rawMarkdown": "How do you know Private LB mean?", "votes": null }, { "id": "542786", "postDate": "06/04/2019 04:58:58", "content": "The mean was indirectly disclosed in the forum. link
\n\nYou just need to use the image to calc it.
\n\nAnd that's the key why we survived the shakeup in LB.
", "rawMarkdown": "The mean was indirectly disclosed in the forum. [link](https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/90664)\n\nYou just need to use the image to calc it.\n\nAnd that's the key why we survived the shakeup in LB.", "votes": null }, { "id": "542803", "postDate": "06/04/2019 05:15:46", "content": "@titericz Oh, thanks for the answer!
", "rawMarkdown": "titericz Oh, thanks for the answer!", "votes": null }, { "id": "543074", "postDate": "06/04/2019 10:12:02", "content": "To my huge surprise my best private score was based on solution, which contained mistake in code (which I noticed after submission), what caused data leakage. Private score 2.44809, public score 1.99416
", "rawMarkdown": "To my huge surprise my best private score was based on solution, which contained mistake in code (which I noticed after submission), what caused data leakage. Private score 2.44809, public score 1.99416", "votes": null }, { "id": "543302", "postDate": "06/04/2019 13:02:10", "content": "As @cpmpml says no gambling while chosing the final submission. Because of this we are regretting. This is a part of learning. Best of luck for next time.
", "rawMarkdown": "As @cpmpml says no gambling while chosing the final submission. Because of this we are regretting. This is a part of learning. Best of luck for next time.", "votes": null } ], "comments": [ { "id": 542510, "author_name": "idog90", "author_url": "", "post_date": "06/04/2019 00:40:13", "content": "I submitted a variant of the straight-forward transformer-network from this public kernel (identical up to changing into my CV for sake of convenience).
\n\nI didn't bother researching this further, since it had terrible LB score and not-so-good CV score.
\n\nWould won a gold medal...
", "votes": null, "replies": [ { "id": 542533, "author_name": "joejeo1", "author_url": "", "post_date": "06/04/2019 00:59:38", "content": "Totally agree. The test set samples were not random. Public LB sample size was too small. Even the final test set size was too small. I actually compiled CNN_LSTM model very similar to that, but with less dense layers. I didn't do well with public leaderboard. I didn't continue work on it. It was my second best score on the P. LB.
", "votes": null, "replies": [] } ] }, { "id": 542520, "author_name": "abhinand05", "author_url": "", "post_date": "06/04/2019 00:49:27", "content": "Same has happened for me. My first submission in this competition would have taken me 500 places up.
", "votes": null, "replies": [] }, { "id": 542526, "author_name": "joejeo1", "author_url": "", "post_date": "06/04/2019 00:54:49", "content": "My best was more 2 weeks ago. I wasted so much time last 2 weeks, although I didn't select my best submission.
", "votes": null, "replies": [] }, { "id": 542530, "author_name": "shinsei66", "author_url": "", "post_date": "06/04/2019 00:56:53", "content": "My first submission which only with several basic statistical features was the best (2.45179) in the Private LB ;)\nIts public LB score was the worst, though.
", "votes": null, "replies": [] }, { "id": 542531, "author_name": "elvenmonk", "author_url": "", "post_date": "06/04/2019 00:57:27", "content": "My Best One was one of the earliest submitted, but with public score 1.69472 I would never imagine that it is one to be chosen.
", "votes": null, "replies": [] }, { "id": 542574, "author_name": "titericz", "author_url": "", "post_date": "06/04/2019 01:40:07", "content": "it happens because the mean of Public and Private LB are very different (4 and 6.7)
", "votes": null, "replies": [ { "id": 542772, "author_name": "sergeyzlobin", "author_url": "", "post_date": "06/04/2019 04:47:26", "content": "How do you know Private LB mean?
", "votes": null, "replies": [] }, { "id": 542786, "author_name": "titericz", "author_url": "", "post_date": "06/04/2019 04:58:58", "content": "The mean was indirectly disclosed in the forum. link
\n\nYou just need to use the image to calc it.
\n\nAnd that's the key why we survived the shakeup in LB.
", "votes": null, "replies": [] }, { "id": 542803, "author_name": "sergeyzlobin", "author_url": "", "post_date": "06/04/2019 05:15:46", "content": "@titericz Oh, thanks for the answer!
", "votes": null, "replies": [] } ] }, { "id": 542680, "author_name": "qiqisang", "author_url": "", "post_date": "06/04/2019 03:27:00", "content": "My first day submission using only 1/3 data was 2.3847. However it's public score is 1.91754, so I didn't choose it......\nThis competition seems really weird.
", "votes": null, "replies": [] }, { "id": 542712, "author_name": "tocha4", "author_url": "", "post_date": "06/04/2019 03:54:12", "content": "Private Score: 2.39760\nPublic Score: 1.63585\nAfter 5 folds test_CV = 2.029 | train_mean = 0.870 --> overfit IMHO.\nSo frustrating!!!
", "votes": null, "replies": [] }, { "id": 543074, "author_name": "rafzy15", "author_url": "", "post_date": "06/04/2019 10:12:02", "content": "To my huge surprise my best private score was based on solution, which contained mistake in code (which I noticed after submission), what caused data leakage. Private score 2.44809, public score 1.99416
", "votes": null, "replies": [] }, { "id": 543302, "author_name": "karanjakhar", "author_url": "", "post_date": "06/04/2019 13:02:10", "content": "As @cpmpml says no gambling while chosing the final submission. Because of this we are regretting. This is a part of learning. Best of luck for next time.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "542498": "How about you guys? ;)", "542510": "I submitted a variant of the straight-forward transformer-network from [this public kernel](https://www.kaggle.com/buchan/transformer-network-with-1d-cnn-feature-extraction) (identical up to changing into my CV for sake of convenience).\n\nI didn't bother researching this further, since it had terrible LB score and not-so-good CV score.\n\nWould won a gold medal...", "542520": "Same has happened for me. My first submission in this competition would have taken me 500 places up.", "542526": "My best was more 2 weeks ago. I wasted so much time last 2 weeks, although I didn't select my best submission.", "542530": "My first submission which only with several basic statistical features was the best (2.45179) in the Private LB ;)\nIts public LB score was the worst, though.", "542531": "My Best One was one of the earliest submitted, but with public score 1.69472 I would never imagine that it is one to be chosen.", "542533": "Totally agree. The test set samples were not random. Public LB sample size was too small. Even the final test set size was too small. I actually compiled CNN_LSTM model very similar to that, but with less dense layers. I didn't do well with public leaderboard. I didn't continue work on it. It was my second best score on the P. LB.", "542574": "it happens because the mean of Public and Private LB are very different (4 and 6.7)", "542680": "My first day submission using only 1/3 data was 2.3847. However it's public score is 1.91754, so I didn't choose it......\nThis competition seems really weird.", "542712": "Private Score: 2.39760\nPublic Score: 1.63585\nAfter 5 folds test_CV = 2.029 | train_mean = 0.870 --> overfit IMHO.\nSo frustrating!!!", "542772": "How do you know Private LB mean?", "542786": "The mean was indirectly disclosed in the forum. [link](https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/90664)\n\nYou just need to use the image to calc it.\n\nAnd that's the key why we survived the shakeup in LB.", "542803": "titericz Oh, thanks for the answer!", "543074": "To my huge surprise my best private score was based on solution, which contained mistake in code (which I noticed after submission), what caused data leakage. Private score 2.44809, public score 1.99416", "543302": "As @cpmpml says no gambling while chosing the final submission. Because of this we are regretting. This is a part of learning. Best of luck for next time." }, "source": "meta" }