![\"3](\"https://indipest.files.wordpress.com/2021/03/bw6d5zz.gif\")
Hello kagglers,
\nYou all must have seen many notebooks which gives 0.999 score. Since the submission dealine is nearing, I thought to make a topic to assist everyone in selection. Many here are experience kagglers while some of them are beginners. So this topic is mainly precautionary topic for them who are thinking to take 0.999 as their final submission.
Some steps you must remember while selecting last submission:
\nDon't and never go with 0.999 score notebooks or if i say, nbs greater than 0.9 will not give you a medal or a good rank.
The crypto market, as @michau96 said in his thread, is fluctuating and going down. The dataset given to us, is of the time when the prices of most of the cryptos were going up. But now, according to the latest trend, it's value is decreasing.
The crypto market is always unpredictable even if you ask about it to experts or any investor too. So just see that your model is able to handle that unpredictability.
Try Hybrid Models rather than single model. A single model will not be able to handle the unpredictability. Hybrid models, having a bit different approach, will be able to handle the performance of market giving you an edge over others.
From all the above statements I mentioned, we can conclude that an hybrid model with 0.4 to 0.65 score will be able to fetch a medal, though exceptions may occur. This is all my opinion I learned from the notebooks and discussions and from knowledge of markets. I would highly like to know your point of view on these points and your suggestions.
\nThanks
\nAditya
Hello kagglers,
\nYou all must have seen many notebooks which gives 0.999 score. Since the submission dealine is nearing, I thought to make a topic to assist everyone in selection. Many here are experience kagglers while some of them are beginners. So this topic is mainly precautionary topic for them who are thinking to take 0.999 as their final submission.
Some steps you must remember while selecting last submission:
\nDon't and never go with 0.999 score notebooks or if i say, nbs greater than 0.9 will not give you a medal or a good rank.
The crypto market, as @michau96 said in his thread, is fluctuating and going down. The dataset given to us, is of the time when the prices of most of the cryptos were going up. But now, according to the latest trend, it's value is decreasing.
The crypto market is always unpredictable even if you ask about it to experts or any investor too. So just see that your model is able to handle that unpredictability.
Try Hybrid Models rather than single model. A single model will not be able to handle the unpredictability. Hybrid models, having a bit different approach, will be able to handle the performance of market giving you an edge over others.
From all the above statements I mentioned, we can conclude that an hybrid model with 0.4 to 0.65 score will be able to fetch a medal, though exceptions may occur. This is all my opinion I learned from the notebooks and discussions and from knowledge of markets. I would highly like to know your point of view on these points and your suggestions.
\nThanks
\nAditya
\n\n0.4 to 0.65
\n
I don't think anyone will be able to get near these scores, I believe top scores will be < 0.1
", "rawMarkdown": "> 0.4 to 0.65\n\nI don't think anyone will be able to get near these scores, I believe top scores will be < 0.1", "votes": 1, "replies": [ { "id": 1669483, "postDate": "2022-01-30T18:07:03.990Z", "content": "It is a possibility but i said the score in regards with CV not private lb score.
", "rawMarkdown": "It is a possibility but i said the score in regards with CV not private lb score.", "votes": -1 }, { "id": 1669654, "postDate": "2022-01-30T21:19:50.553Z", "content": "Public leaderboard perhaps?
\nI doubt anyone can get a score of .4 in their CV, if they've done CV correctly
", "rawMarkdown": "Public leaderboard perhaps?\n\nI doubt anyone can get a score of .4 in their CV, if they've done CV correctly" }, { "id": 1669732, "postDate": "2022-01-30T22:29:32.950Z", "content": "\n\nThis is all my opinion I learned from the notebooks and discussions and from knowledge of markets
\n
Did your learning start last week? 0.4 is not possible without major over-fitting.
\nHistoric top scores for this in equities markets is 0.04… If you are getting \"0.4 to 0.65\" in your CV \"hybrid models\" then you are in the same group of people you are trying to advise..
", "rawMarkdown": "> This is all my opinion I learned from the notebooks and discussions and from knowledge of markets\n\nDid your learning start last week? 0.4 is not possible without major over-fitting.\n\nHistoric top scores for this in equities markets is 0.04... If you are getting \"0.4 to 0.65\" in your CV \"hybrid models\" then you are in the same group of people you are trying to advise..", "votes": 1 }, { "id": 1669999, "postDate": "2022-01-31T05:59:50.220Z", "content": "Actually I have started this competition from starting and have seen market and it is my assumption of getting it near 0.4, CV or public lb. I do agree that market is unpredictable and it is a possibility that <0.1 may be the winning score. But i am getting a feeling like that. And thanks why i have mentioned to get your alls point of view as there are many experts and experienced person and my assumption can be wrong or very wrong too. Thanks for the comment @julianmukaj
", "rawMarkdown": "Actually I have started this competition from starting and have seen market and it is my assumption of getting it near 0.4, CV or public lb. I do agree that market is unpredictable and it is a possibility that <0.1 may be the winning score. But i am getting a feeling like that. And thanks why i have mentioned to get your alls point of view as there are many experts and experienced person and my assumption can be wrong or very wrong too. Thanks for the comment @julianmukaj " } ] }, { "id": 1670602, "postDate": "2022-01-31T17:00:46.230Z", "content": "Date range of sample test data and the notebook with 0.999 score and are confusing people. I agree the strict model is good to evaluate and select the model. But it should not be used for submission, because it will restrict the data range for learning. As my conclusion, I never select the notebook with 0.999 and the notebook with a strict model itself.
", "rawMarkdown": "Date range of sample test data and the notebook with 0.999 score and are confusing people. I agree the strict model is good to evaluate and select the model. But it should not be used for submission, because it will restrict the data range for learning. As my conclusion, I never select the notebook with 0.999 and the notebook with a strict model itself.", "votes": 2 }, { "id": 1668207, "postDate": "2022-01-29T12:48:31.247Z", "content": "
\n\n0.4 to 0.65
\n
I don't think anyone will be able to get near these scores, I believe top scores will be < 0.1
", "votes": 1, "replies": [ { "id": 1669483, "author_name": "Aditya Sharma", "author_url": "", "post_date": "2022-01-30T18:07:03.990000", "content": "It is a possibility but i said the score in regards with CV not private lb score.
", "votes": -1, "replies": [] }, { "id": 1669654, "author_name": "Marco Gorelli", "author_url": "", "post_date": "2022-01-30T21:19:50.553000", "content": "Public leaderboard perhaps?
\nI doubt anyone can get a score of .4 in their CV, if they've done CV correctly
", "votes": 0, "replies": [] }, { "id": 1669732, "author_name": "JM", "author_url": "", "post_date": "2022-01-30T22:29:32.950000", "content": "\n\nThis is all my opinion I learned from the notebooks and discussions and from knowledge of markets
\n
Did your learning start last week? 0.4 is not possible without major over-fitting.
\nHistoric top scores for this in equities markets is 0.04… If you are getting \"0.4 to 0.65\" in your CV \"hybrid models\" then you are in the same group of people you are trying to advise..
", "votes": 1, "replies": [] }, { "id": 1669999, "author_name": "Aditya Sharma", "author_url": "", "post_date": "2022-01-31T05:59:50.220000", "content": "Actually I have started this competition from starting and have seen market and it is my assumption of getting it near 0.4, CV or public lb. I do agree that market is unpredictable and it is a possibility that <0.1 may be the winning score. But i am getting a feeling like that. And thanks why i have mentioned to get your alls point of view as there are many experts and experienced person and my assumption can be wrong or very wrong too. Thanks for the comment @julianmukaj
", "votes": 0, "replies": [] } ] }, { "id": 1670602, "author_name": "stpete_ishii", "author_url": "", "post_date": "2022-01-31T17:00:46.230000", "content": "Date range of sample test data and the notebook with 0.999 score and are confusing people. I agree the strict model is good to evaluate and select the model. But it should not be used for submission, because it will restrict the data range for learning. As my conclusion, I never select the notebook with 0.999 and the notebook with a strict model itself.
", "votes": 2, "replies": [] }, { "id": 1668207, "author_name": "JM", "author_url": "", "post_date": "2022-01-29T12:48:31.247000", "content": "