My observations like many others is the difficulty of getting consistent results with NNs. It was really hard to get a good cross validation setup for this problem. As a result I did not chose what turned out to be my best model on teh private LB.
\n\nMy best model on hindsight is a CatBoost model with public LB = 0.648, private LB=0.655, and CV scores 0.719 ± 0.094 but I did not choose it as one of my final submissions. This would have put me in bronze medal at number 80 position.
", "messages": [ { "id": 496197, "postDate": "2019-03-22T00:55:38.237Z", "content": "My observations like many others is the difficulty of getting consistent results with NNs. It was really hard to get a good cross validation setup for this problem. As a result I did not chose what turned out to be my best model on teh private LB.
\n\nMy best model on hindsight is a CatBoost model with public LB = 0.648, private LB=0.655, and CV scores 0.719 ± 0.094 but I did not choose it as one of my final submissions. This would have put me in bronze medal at number 80 position.
", "rawMarkdown": "My observations like many others is the difficulty of getting consistent results with NNs. It was really hard to get a [good cross validation setup for this problem](https://www.kaggle.com/c/vsb-power-line-fault-detection/discussion/85143). As a result I did not chose what turned out to be my best model on teh private LB.\n \nMy best model on hindsight is a [CatBoost model with public LB = 0.648, private LB=0.655, and CV scores 0.719 ± 0.094 ](https://www.kaggle.com/sheriytm/catboost-with-handmade-features?scriptVersionId=10861610)but I did not choose it as one of my final submissions. This would have put me in bronze medal at number 80 position.", "votes": 4 }, { "id": 496211, "postDate": "2019-03-22T01:20:26.883Z", "content": "Hi YaGana ! Sorry that this happened to you again :(\nMy best unselected submission would've got 0.655 on the private LB, and only 0.577 on the pubic LB.
", "rawMarkdown": "Hi YaGana ! Sorry that this happened to you again :(\nMy best unselected submission would've got 0.655 on the private LB, and only 0.577 on the pubic LB.", "votes": 1, "replies": [ { "id": 496218, "postDate": "2019-03-22T01:29:15.300Z", "content": "Thanks @tarunpaparaju, and congratulations on a strong showing here. I was using a lower gap between the best_threshold score and my averaged fold score as a guide for a good submission but I did not consider that with my LGBM and Catboost models because their scores were so low.
\n\nA quick question, did you upvote this post twice? I got an email you upvoted but it does not show anymore.
", "rawMarkdown": "Thanks @tarunpaparaju, and congratulations on a strong showing here. I was using a lower gap between the best_threshold score and my averaged fold score as a guide for a good submission but I did not consider that with my LGBM and Catboost models because their scores were so low.\n\nA quick question, did you upvote this post twice? I got an email you upvoted but it does not show anymore." }, { "id": 496221, "postDate": "2019-03-22T01:31:14.533Z", "content": "Upvoted now :)
", "rawMarkdown": "Upvoted now :)" }, { "id": 496222, "postDate": "2019-03-22T01:34:10.297Z", "content": "Thanks ! You did well too, despite not winning a medal :)
", "rawMarkdown": "Thanks ! You did well too, despite not winning a medal :)", "votes": 1 }, { "id": 496223, "postDate": "2019-03-22T01:34:19.937Z", "rawMarkdown": "", "isDeleted": true }, { "id": 496224, "postDate": "2019-03-22T01:34:53.303Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 496211, "author_name": "Tarun Paparaju", "author_url": "", "post_date": "2019-03-22T01:20:26.883000", "content": "Hi YaGana ! Sorry that this happened to you again :(\nMy best unselected submission would've got 0.655 on the private LB, and only 0.577 on the pubic LB.
", "votes": 1, "replies": [ { "id": 496218, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2019-03-22T01:29:15.300000", "content": "Thanks @tarunpaparaju, and congratulations on a strong showing here. I was using a lower gap between the best_threshold score and my averaged fold score as a guide for a good submission but I did not consider that with my LGBM and Catboost models because their scores were so low.
\n\nA quick question, did you upvote this post twice? I got an email you upvoted but it does not show anymore.
", "votes": 0, "replies": [] }, { "id": 496221, "author_name": "Tarun Paparaju", "author_url": "", "post_date": "2019-03-22T01:31:14.533000", "content": "Upvoted now :)
", "votes": 0, "replies": [] }, { "id": 496222, "author_name": "Tarun Paparaju", "author_url": "", "post_date": "2019-03-22T01:34:10.297000", "content": "Thanks ! You did well too, despite not winning a medal :)
", "votes": 1, "replies": [] }, { "id": 496223, "author_name": "", "author_url": "", "post_date": "2019-03-22T01:34:19.937000", "content": "", "votes": 0, "replies": [] }, { "id": 496224, "author_name": "", "author_url": "", "post_date": "2019-03-22T01:34:53.303000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "496197": "My observations like many others is the difficulty of getting consistent results with NNs. It was really hard to get a [good cross validation setup for this problem](https://www.kaggle.com/c/vsb-power-line-fault-detection/discussion/85143). As a result I did not chose what turned out to be my best model on teh private LB.\n \nMy best model on hindsight is a [CatBoost model with public LB = 0.648, private LB=0.655, and CV scores 0.719 ± 0.094 ](https://www.kaggle.com/sheriytm/catboost-with-handmade-features?scriptVersionId=10861610)but I did not choose it as one of my final submissions. This would have put me in bronze medal at number 80 position.", "496211": "Hi YaGana ! Sorry that this happened to you again :(\nMy best unselected submission would've got 0.655 on the private LB, and only 0.577 on the pubic LB." } }