looks like this competition will see .9xx ?
", "messages": [ { "id": 459705, "postDate": "2019-01-22T08:03:30.780Z", "content": "For my results, credit goes to this kernel\nhttps://www.kaggle.com/braquino/vsb-power-lstm-attention. i just do the tuning.
\n\nInitially, I started with Lightgbm and the best Lb i achieved was 0.555.
", "rawMarkdown": "For my results, credit goes to this kernel\nhttps://www.kaggle.com/braquino/vsb-power-lstm-attention. i just do the tuning. \n\nInitially, I started with Lightgbm and the best Lb i achieved was 0.555.", "votes": 7, "replies": [ { "id": 459767, "postDate": "2019-01-22T10:28:00.503Z", "content": "Is your model stable?
", "rawMarkdown": "Is your model stable?", "votes": 3 }, { "id": 459768, "postDate": "2019-01-22T10:34:39.450Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 480399, "postDate": "2019-02-28T06:17:58.343Z", "content": "Finally we see 0.8 submission today. I wonder whether it is stable model?
", "rawMarkdown": "Finally we see 0.8 submission today. I wonder whether it is stable model?", "votes": 1, "replies": [ { "id": 480453, "postDate": "2019-02-28T08:00:37.947Z", "content": "Ensembles of models are always stable enough. But don't worry, the ensemble was overfitted on public LB a bit and it very likely will be worse on the private part. Anyway the difference between 0.7 and 0.8 is not so huge as it seems.
", "rawMarkdown": "Ensembles of models are always stable enough. But don't worry, the ensemble was overfitted on public LB a bit and it very likely will be worse on the private part. Anyway the difference between 0.7 and 0.8 is not so huge as it seems.", "votes": 5 }, { "id": 480946, "postDate": "2019-02-28T22:57:11.757Z", "content": "A few FP and FN and score decreases a lot. MCC penalizes each wrong prediction.\nHow many positives do you have? Here around 800 (around 4%) for best LB score. It's far from the 6% of the train ratio.
", "rawMarkdown": "A few FP and FN and score decreases a lot. MCC penalizes each wrong prediction.\nHow many positives do you have? Here around 800 (around 4%) for best LB score. It's far from the 6% of the train ratio. ", "votes": 5 } ] }, { "id": 460086, "postDate": "2019-01-23T01:13:52.943Z", "content": "@ONODERA, Model is not stable yet. Besides, I am unable to reproduce the model which gave 0.724, as i didn't fix the seeds that time. The best reproducible model so far for me is giving 0.684 on lb.
", "rawMarkdown": "@ONODERA, Model is not stable yet. Besides, I am unable to reproduce the model which gave 0.724, as i didn't fix the seeds that time. The best reproducible model so far for me is giving 0.684 on lb.", "votes": 2 }, { "id": 458819, "postDate": "2019-01-20T14:41:36.947Z", "content": "looks like this competition will see .9xx ?
", "rawMarkdown": "looks like this competition will see .9xx ?", "votes": 2 } ], "comments": [ { "id": 459705, "author_name": "HarshitMehta", "author_url": "", "post_date": "2019-01-22T08:03:30.780000", "content": "For my results, credit goes to this kernel\nhttps://www.kaggle.com/braquino/vsb-power-lstm-attention. i just do the tuning.
\n\nInitially, I started with Lightgbm and the best Lb i achieved was 0.555.
", "votes": 7, "replies": [ { "id": 459767, "author_name": "ONODERA", "author_url": "", "post_date": "2019-01-22T10:28:00.503000", "content": "Is your model stable?
", "votes": 3, "replies": [] }, { "id": 459768, "author_name": "", "author_url": "", "post_date": "2019-01-22T10:34:39.450000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 480399, "author_name": "Anna Novikova", "author_url": "", "post_date": "2019-02-28T06:17:58.343000", "content": "Finally we see 0.8 submission today. I wonder whether it is stable model?
", "votes": 1, "replies": [ { "id": 480453, "author_name": "Sergei Fironov", "author_url": "", "post_date": "2019-02-28T08:00:37.947000", "content": "Ensembles of models are always stable enough. But don't worry, the ensemble was overfitted on public LB a bit and it very likely will be worse on the private part. Anyway the difference between 0.7 and 0.8 is not so huge as it seems.
", "votes": 5, "replies": [] }, { "id": 480946, "author_name": "MPWARE", "author_url": "", "post_date": "2019-02-28T22:57:11.757000", "content": "A few FP and FN and score decreases a lot. MCC penalizes each wrong prediction.\nHow many positives do you have? Here around 800 (around 4%) for best LB score. It's far from the 6% of the train ratio.
", "votes": 5, "replies": [] } ] }, { "id": 460086, "author_name": "HarshitMehta", "author_url": "", "post_date": "2019-01-23T01:13:52.943000", "content": "@ONODERA, Model is not stable yet. Besides, I am unable to reproduce the model which gave 0.724, as i didn't fix the seeds that time. The best reproducible model so far for me is giving 0.684 on lb.
", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "459705": "For my results, credit goes to this kernel\nhttps://www.kaggle.com/braquino/vsb-power-lstm-attention. i just do the tuning. \n\nInitially, I started with Lightgbm and the best Lb i achieved was 0.555.", "480399": "Finally we see 0.8 submission today. I wonder whether it is stable model?", "460086": "@ONODERA, Model is not stable yet. Besides, I am unable to reproduce the model which gave 0.724, as i didn't fix the seeds that time. The best reproducible model so far for me is giving 0.684 on lb.", "458819": "looks like this competition will see .9xx ?" } }