Here are a few of the features that im using: https://www.kaggle.com/abhishek/quite-a-few-features-1-51
\n\nI think you can just improve on the score by using better parameters
", "messages": [ { "id": "520565", "postDate": "04/21/2019 10:05:37", "content": "Here are a few of the features that im using: https://www.kaggle.com/abhishek/quite-a-few-features-1-51
\n\nI think you can just improve on the score by using better parameters
", "rawMarkdown": "Here are a few of the features that im using: https://www.kaggle.com/abhishek/quite-a-few-features-1-51 \n\nI think you can just improve on the score by using better parameters", "votes": null }, { "id": "520571", "postDate": "04/21/2019 10:16:49", "content": "Thanks for sharing.
", "rawMarkdown": "Thanks for sharing.", "votes": null }, { "id": "520595", "postDate": "04/21/2019 11:31:08", "content": "Thanks for sharing! :-)
", "rawMarkdown": "Thanks for sharing! :-)", "votes": null }, { "id": "520599", "postDate": "04/21/2019 11:44:49", "content": "Thanks for sharing! Which ones you found had a better predictive value?
", "rawMarkdown": "Thanks for sharing! Which ones you found had a better predictive value?", "votes": null }, { "id": "520604", "postDate": "04/21/2019 11:57:26", "content": "Thanks for sharing! Lets improve it.
", "rawMarkdown": "Thanks for sharing! Lets improve it.", "votes": null }, { "id": "520621", "postDate": "04/21/2019 12:37:34", "content": "You may try it for yourself by using the feature importance generated by xgboost (you may add it in existing code)and experiment which one is better.. :-)
", "rawMarkdown": "You may try it for yourself by using the feature importance generated by xgboost (you may add it in existing code)and experiment which one is better.. :-)", "votes": null }, { "id": "520684", "postDate": "04/21/2019 15:28:05", "content": "@abhishek, thanks for sharing. I see a few features I have not used and will be testing them.
", "rawMarkdown": "abhishek, thanks for sharing. I see a few features I have not used and will be testing them.", "votes": null }, { "id": "520709", "postDate": "04/21/2019 16:09:11", "content": "TQSM for sharing @abhishek Sir.
", "rawMarkdown": "**TQSM for sharing @abhishek Sir.**", "votes": null }, { "id": "520793", "postDate": "04/21/2019 18:56:37", "content": "is your approach different?
", "rawMarkdown": "is your approach different?", "votes": null }, { "id": "520824", "postDate": "04/21/2019 20:08:13", "content": "My approach is kind of similar. Since I am in this competition for a while, I am at the feature selection stage right now. There are features I was thinking about adding that I now see in your kernel. So I have tested them but they did not improve my CV hence not good for my setup. Perharps they do not interact well with my set of features. I will revisit that later.
", "rawMarkdown": "My approach is kind of similar. Since I am in this competition for a while, I am at the feature selection stage right now. There are features I was thinking about adding that I now see in your kernel. So I have tested them but they did not improve my CV hence not good for my setup. Perharps they do not interact well with my set of features. I will revisit that later.", "votes": null }, { "id": "520826", "postDate": "04/21/2019 20:09:22", "content": "Have you tried sk-learns \"f regression\" or \"mutual info regression\" for this task? I'm using them right now to check if features are good and how to tweak them, but yours might be better.
", "rawMarkdown": "Have you tried sk-learns \"f regression\" or \"mutual info regression\" for this task? I'm using them right now to check if features are good and how to tweak them, but yours might be better.", "votes": null }, { "id": "520957", "postDate": "04/22/2019 04:12:56", "content": "Thanks for sharing it helps to improve my score on LB
", "rawMarkdown": "Thanks for sharing it helps to improve my score on LB", "votes": null }, { "id": "520989", "postDate": "04/22/2019 06:14:23", "content": "Indeed, I'll go through the analysis next weekend and post the results here ;)
", "rawMarkdown": "Indeed, I'll go through the analysis next weekend and post the results here ;)", "votes": null }, { "id": "521039", "postDate": "04/22/2019 08:21:54", "content": "a very well done project
", "rawMarkdown": "a very well done project", "votes": null }, { "id": "521581", "postDate": "04/23/2019 04:18:38", "content": "Thanks for sharing.
", "rawMarkdown": "Thanks for sharing.", "votes": null }, { "id": "521803", "postDate": "04/23/2019 13:00:48", "content": "I love all of the hard work you have done @abhishek , and I have upvoted your topic and Kernel, but please forgive me for being a ML purist. I definitely lean towards the theory that we should skip feature selection and let the Machine Learning figure features out on it's own. In other words, let an algorithm like Convolutional Neural Networks (CNNs) create the appropriate wavelet-ish decomposition.
\n\nI made an unrelated public kernel using this technique - not a great LB score but you get the idea - please upvote the Kernel if you find it useful, and good luck in the competition!
\n\nhttps://www.kaggle.com/pnussbaum/earthquake-pred-cnn-medical-analogy-v07
\n\nP.S. The above Kernel is really focused on the new Food and Drug Administration advice for AI/ML Medical Devices. I wanted to encourage contest participants working on an \"Earthquake Warning System\" to also consider careers in \"Medical Warning Devices\" and all of the overlap in disciplines - so enjoy!
", "rawMarkdown": "I love all of the hard work you have done @abhishek , and I have upvoted your topic and Kernel, but please forgive me for being a ML purist. I definitely lean towards the theory that we should **skip feature selection and let the Machine Learning figure features out on it's own.** In other words, let an algorithm like Convolutional Neural Networks (CNNs) create the appropriate wavelet-ish decomposition. \n\nI made an unrelated public kernel using this technique - not a great LB score but you get the idea - please upvote the Kernel if you find it useful, and good luck in the competition!\n\nhttps://www.kaggle.com/pnussbaum/earthquake-pred-cnn-medical-analogy-v07\n\nP.S. The above Kernel is really focused on the new Food and Drug Administration advice for AI/ML Medical Devices. I wanted to encourage contest participants working on an \"Earthquake Warning System\" to also consider careers in \"Medical Warning Devices\" and all of the overlap in disciplines - so enjoy!", "votes": null } ], "comments": [ { "id": 520571, "author_name": "cpmpml", "author_url": "", "post_date": "04/21/2019 10:16:49", "content": "Thanks for sharing.
", "votes": null, "replies": [] }, { "id": 520595, "author_name": "projdev", "author_url": "", "post_date": "04/21/2019 11:31:08", "content": "Thanks for sharing! :-)
", "votes": null, "replies": [] }, { "id": 520599, "author_name": "ricarddelgado", "author_url": "", "post_date": "04/21/2019 11:44:49", "content": "Thanks for sharing! Which ones you found had a better predictive value?
", "votes": null, "replies": [ { "id": 520621, "author_name": "projdev", "author_url": "", "post_date": "04/21/2019 12:37:34", "content": "You may try it for yourself by using the feature importance generated by xgboost (you may add it in existing code)and experiment which one is better.. :-)
", "votes": null, "replies": [] }, { "id": 520826, "author_name": "svenhinderer", "author_url": "", "post_date": "04/21/2019 20:09:22", "content": "Have you tried sk-learns \"f regression\" or \"mutual info regression\" for this task? I'm using them right now to check if features are good and how to tweak them, but yours might be better.
", "votes": null, "replies": [] }, { "id": 520989, "author_name": "ricarddelgado", "author_url": "", "post_date": "04/22/2019 06:14:23", "content": "Indeed, I'll go through the analysis next weekend and post the results here ;)
", "votes": null, "replies": [] } ] }, { "id": 520604, "author_name": "dhaqui", "author_url": "", "post_date": "04/21/2019 11:57:26", "content": "Thanks for sharing! Lets improve it.
", "votes": null, "replies": [] }, { "id": 520684, "author_name": "sheriytm", "author_url": "", "post_date": "04/21/2019 15:28:05", "content": "@abhishek, thanks for sharing. I see a few features I have not used and will be testing them.
", "votes": null, "replies": [ { "id": 520793, "author_name": "abhishek", "author_url": "", "post_date": "04/21/2019 18:56:37", "content": "is your approach different?
", "votes": null, "replies": [] }, { "id": 520824, "author_name": "sheriytm", "author_url": "", "post_date": "04/21/2019 20:08:13", "content": "My approach is kind of similar. Since I am in this competition for a while, I am at the feature selection stage right now. There are features I was thinking about adding that I now see in your kernel. So I have tested them but they did not improve my CV hence not good for my setup. Perharps they do not interact well with my set of features. I will revisit that later.
", "votes": null, "replies": [] } ] }, { "id": 520709, "author_name": "akashravichandran", "author_url": "", "post_date": "04/21/2019 16:09:11", "content": "TQSM for sharing @abhishek Sir.
", "votes": null, "replies": [] }, { "id": 520957, "author_name": "himusoni", "author_url": "", "post_date": "04/22/2019 04:12:56", "content": "Thanks for sharing it helps to improve my score on LB
", "votes": null, "replies": [] }, { "id": 521039, "author_name": "muslekh12", "author_url": "", "post_date": "04/22/2019 08:21:54", "content": "a very well done project
", "votes": null, "replies": [] }, { "id": 521581, "author_name": "matsumotoshintaro", "author_url": "", "post_date": "04/23/2019 04:18:38", "content": "Thanks for sharing.
", "votes": null, "replies": [] }, { "id": 521803, "author_name": "pnussbaum", "author_url": "", "post_date": "04/23/2019 13:00:48", "content": "I love all of the hard work you have done @abhishek , and I have upvoted your topic and Kernel, but please forgive me for being a ML purist. I definitely lean towards the theory that we should skip feature selection and let the Machine Learning figure features out on it's own. In other words, let an algorithm like Convolutional Neural Networks (CNNs) create the appropriate wavelet-ish decomposition.
\n\nI made an unrelated public kernel using this technique - not a great LB score but you get the idea - please upvote the Kernel if you find it useful, and good luck in the competition!
\n\nhttps://www.kaggle.com/pnussbaum/earthquake-pred-cnn-medical-analogy-v07
\n\nP.S. The above Kernel is really focused on the new Food and Drug Administration advice for AI/ML Medical Devices. I wanted to encourage contest participants working on an \"Earthquake Warning System\" to also consider careers in \"Medical Warning Devices\" and all of the overlap in disciplines - so enjoy!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "520565": "Here are a few of the features that im using: https://www.kaggle.com/abhishek/quite-a-few-features-1-51 \n\nI think you can just improve on the score by using better parameters", "520571": "Thanks for sharing.", "520595": "Thanks for sharing! :-)", "520599": "Thanks for sharing! Which ones you found had a better predictive value?", "520604": "Thanks for sharing! Lets improve it.", "520621": "You may try it for yourself by using the feature importance generated by xgboost (you may add it in existing code)and experiment which one is better.. :-)", "520684": "abhishek, thanks for sharing. I see a few features I have not used and will be testing them.", "520709": "**TQSM for sharing @abhishek Sir.**", "520793": "is your approach different?", "520824": "My approach is kind of similar. Since I am in this competition for a while, I am at the feature selection stage right now. There are features I was thinking about adding that I now see in your kernel. So I have tested them but they did not improve my CV hence not good for my setup. Perharps they do not interact well with my set of features. I will revisit that later.", "520826": "Have you tried sk-learns \"f regression\" or \"mutual info regression\" for this task? I'm using them right now to check if features are good and how to tweak them, but yours might be better.", "520957": "Thanks for sharing it helps to improve my score on LB", "520989": "Indeed, I'll go through the analysis next weekend and post the results here ;)", "521039": "a very well done project", "521581": "Thanks for sharing.", "521803": "I love all of the hard work you have done @abhishek , and I have upvoted your topic and Kernel, but please forgive me for being a ML purist. I definitely lean towards the theory that we should **skip feature selection and let the Machine Learning figure features out on it's own.** In other words, let an algorithm like Convolutional Neural Networks (CNNs) create the appropriate wavelet-ish decomposition. \n\nI made an unrelated public kernel using this technique - not a great LB score but you get the idea - please upvote the Kernel if you find it useful, and good luck in the competition!\n\nhttps://www.kaggle.com/pnussbaum/earthquake-pred-cnn-medical-analogy-v07\n\nP.S. The above Kernel is really focused on the new Food and Drug Administration advice for AI/ML Medical Devices. I wanted to encourage contest participants working on an \"Earthquake Warning System\" to also consider careers in \"Medical Warning Devices\" and all of the overlap in disciplines - so enjoy!" }, "source": "meta" }