Hi everyone!
\n\nThis competition has been a great learning experience for me so far. I have tried a lot of different routes and I think I found a valid CV strategy as my CV and LB score is well aligned.
\n\nLGBM 3Kfold Not Shuffled\nCV: 2.084, LB: 1.526\nCV: 2.052, LB: 1.490\nCV: 2.014, LB: 1.433\nCV: 1.996, LB: 1.416 (only 6 features)
\n\nWhile adding lots of features in the beginning, in the last step I reduced the features by a large margin and was able to increase the model accuracy.
\n\nMy current highest LB score is a blend with a public kernel. As this seems to work well, I am now looking for a team up to blend our models and engineered features.
\n\nIf you are interested, please post your validation strategy and corresponding CV/LB scores. \nI'd also love to get in touch with more experienced kagglers to guide me in the last 24 days of this competition.
\n\nIlu
", "messages": [ { "id": "529977", "postDate": "05/11/2019 11:17:12", "content": "Hi everyone!
\n\nThis competition has been a great learning experience for me so far. I have tried a lot of different routes and I think I found a valid CV strategy as my CV and LB score is well aligned.
\n\nLGBM 3Kfold Not Shuffled\nCV: 2.084, LB: 1.526\nCV: 2.052, LB: 1.490\nCV: 2.014, LB: 1.433\nCV: 1.996, LB: 1.416 (only 6 features)
\n\nWhile adding lots of features in the beginning, in the last step I reduced the features by a large margin and was able to increase the model accuracy.
\n\nMy current highest LB score is a blend with a public kernel. As this seems to work well, I am now looking for a team up to blend our models and engineered features.
\n\nIf you are interested, please post your validation strategy and corresponding CV/LB scores. \nI'd also love to get in touch with more experienced kagglers to guide me in the last 24 days of this competition.
\n\nIlu
", "rawMarkdown": "Hi everyone!\n\nThis competition has been a great learning experience for me so far. I have tried a lot of different routes and I think I found a valid CV strategy as my CV and LB score is well aligned.\n\nLGBM 3Kfold Not Shuffled\nCV: 2.084, LB: 1.526\nCV: 2.052, LB: 1.490\nCV: 2.014, LB: 1.433\nCV: 1.996, LB: 1.416 (only 6 features)\n\nWhile adding lots of features in the beginning, in the last step I reduced the features by a large margin and was able to increase the model accuracy. \n\nMy current highest LB score is a blend with a public kernel. As this seems to work well, I am now looking for a team up to blend our models and engineered features. \n\nIf you are interested, please post your validation strategy and corresponding CV/LB scores. \nI'd also love to get in touch with more experienced kagglers to guide me in the last 24 days of this competition. \n\nIlu", "votes": null }, { "id": "529978", "postDate": "05/11/2019 11:24:41", "content": "\n\n\n(only 6 features)
\n
Nice!
", "rawMarkdown": "> (only 6 features)\n\nNice!", "votes": null }, { "id": "529983", "postDate": "05/11/2019 11:42:03", "content": "Thanks. I took your advice, uncle @cpmpml \nYou are probably not looking for an apprentice?
", "rawMarkdown": "Thanks. I took your advice, uncle @cpmpml \nYou are probably not looking for an apprentice?", "votes": null }, { "id": "529988", "postDate": "05/11/2019 11:55:19", "content": "I have already agreed to team, sorry. I hope you'll find good team mates.
", "rawMarkdown": "I have already agreed to team, sorry. I hope you'll find good team mates.", "votes": null }, { "id": "530045", "postDate": "05/11/2019 15:06:52", "content": "The more features I feed to my model, the better it gets (both CV and LB). I'm not surprised that you can achieve a very good score with few features, but surprised that your models are not further improving by adding more features.
", "rawMarkdown": "The more features I feed to my model, the better it gets (both CV and LB). I'm not surprised that you can achieve a very good score with few features, but surprised that your models are not further improving by adding more features.", "votes": null }, { "id": "530069", "postDate": "05/11/2019 16:13:34", "content": "Congratulations on building a robust model with only few features. I agree with your approach
", "rawMarkdown": "Congratulations on building a robust model with only few features. I agree with your approach", "votes": null }, { "id": "530106", "postDate": "05/11/2019 17:56:46", "content": "Thank you, to everyone who sent me an invitation. I feel very flattered ;)\nI have decided to join @pukkinming and his team.
", "rawMarkdown": "Thank you, to everyone who sent me an invitation. I feel very flattered ;)\nI have decided to join @pukkinming and his team.", "votes": null }, { "id": "530129", "postDate": "05/11/2019 19:41:57", "content": "I've had several offers but wanted to see how well I can do in this competition working alone. Maybe I'll regret my decision!
", "rawMarkdown": "I've had several offers but wanted to see how well I can do in this competition working alone. Maybe I'll regret my decision!", "votes": null }, { "id": "530131", "postDate": "05/11/2019 19:58:04", "content": "First of all, you still have plenty of time to join a team. And doing something alone doesn't have to be bad. I did that the last 3 comps, but this time I decided to go the team route and learn something new :)
", "rawMarkdown": "First of all, you still have plenty of time to join a team. And doing something alone doesn't have to be bad. I did that the last 3 comps, but this time I decided to go the team route and learn something new :)", "votes": null }, { "id": "530133", "postDate": "05/11/2019 20:03:08", "content": "There are plenty of people I'd be very happy to work with (I think you're in a fantastic team!) but I think for this competition I'll see what I can achieve by myself. Fingers crossed!
", "rawMarkdown": "There are plenty of people I'd be very happy to work with (I think you're in a fantastic team!) but I think for this competition I'll see what I can achieve by myself. Fingers crossed!", "votes": null }, { "id": "530604", "postDate": "05/13/2019 09:12:01", "content": "only 6 features!\nCan I ask what feature you use?
", "rawMarkdown": "only 6 features!\nCan I ask what feature you use?", "votes": null }, { "id": "530643", "postDate": "05/13/2019 10:47:12", "content": "You may ask, but I obviously am not telling you before the competition end. Sorry.
", "rawMarkdown": "You may ask, but I obviously am not telling you before the competition end. Sorry.", "votes": null }, { "id": "530991", "postDate": "05/14/2019 05:10:07", "content": "mine used similar umber of features, or even less, achieved 1.5LB using pure RNN. I usually like more on playing with model, not tricky features.
", "rawMarkdown": "mine used similar umber of features, or even less, achieved 1.5LB using pure RNN. I usually like more on playing with model, not tricky features.", "votes": null }, { "id": "531312", "postDate": "05/14/2019 16:23:46", "content": "Yes,I will do the same.
", "rawMarkdown": "Yes,I will do the same.", "votes": null } ], "comments": [ { "id": 529978, "author_name": "cpmpml", "author_url": "", "post_date": "05/11/2019 11:24:41", "content": "\n\n\n(only 6 features)
\n
Nice!
", "votes": null, "replies": [ { "id": 529983, "author_name": "ilu000", "author_url": "", "post_date": "05/11/2019 11:42:03", "content": "Thanks. I took your advice, uncle @cpmpml \nYou are probably not looking for an apprentice?
", "votes": null, "replies": [] }, { "id": 529988, "author_name": "cpmpml", "author_url": "", "post_date": "05/11/2019 11:55:19", "content": "I have already agreed to team, sorry. I hope you'll find good team mates.
", "votes": null, "replies": [] }, { "id": 530045, "author_name": "amjad85", "author_url": "", "post_date": "05/11/2019 15:06:52", "content": "The more features I feed to my model, the better it gets (both CV and LB). I'm not surprised that you can achieve a very good score with few features, but surprised that your models are not further improving by adding more features.
", "votes": null, "replies": [] } ] }, { "id": 530069, "author_name": "stecasasso", "author_url": "", "post_date": "05/11/2019 16:13:34", "content": "Congratulations on building a robust model with only few features. I agree with your approach
", "votes": null, "replies": [] }, { "id": 530106, "author_name": "ilu000", "author_url": "", "post_date": "05/11/2019 17:56:46", "content": "Thank you, to everyone who sent me an invitation. I feel very flattered ;)\nI have decided to join @pukkinming and his team.
", "votes": null, "replies": [] }, { "id": 530129, "author_name": "bigironsphere", "author_url": "", "post_date": "05/11/2019 19:41:57", "content": "I've had several offers but wanted to see how well I can do in this competition working alone. Maybe I'll regret my decision!
", "votes": null, "replies": [ { "id": 530131, "author_name": "ilu000", "author_url": "", "post_date": "05/11/2019 19:58:04", "content": "First of all, you still have plenty of time to join a team. And doing something alone doesn't have to be bad. I did that the last 3 comps, but this time I decided to go the team route and learn something new :)
", "votes": null, "replies": [] }, { "id": 530133, "author_name": "bigironsphere", "author_url": "", "post_date": "05/11/2019 20:03:08", "content": "There are plenty of people I'd be very happy to work with (I think you're in a fantastic team!) but I think for this competition I'll see what I can achieve by myself. Fingers crossed!
", "votes": null, "replies": [] } ] }, { "id": 530604, "author_name": "zjp2origin", "author_url": "", "post_date": "05/13/2019 09:12:01", "content": "only 6 features!\nCan I ask what feature you use?
", "votes": null, "replies": [ { "id": 530643, "author_name": "ilu000", "author_url": "", "post_date": "05/13/2019 10:47:12", "content": "You may ask, but I obviously am not telling you before the competition end. Sorry.
", "votes": null, "replies": [] }, { "id": 531312, "author_name": "zjp2origin", "author_url": "", "post_date": "05/14/2019 16:23:46", "content": "Yes,I will do the same.
", "votes": null, "replies": [] } ] }, { "id": 530991, "author_name": "asterisk", "author_url": "", "post_date": "05/14/2019 05:10:07", "content": "mine used similar umber of features, or even less, achieved 1.5LB using pure RNN. I usually like more on playing with model, not tricky features.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "529977": "Hi everyone!\n\nThis competition has been a great learning experience for me so far. I have tried a lot of different routes and I think I found a valid CV strategy as my CV and LB score is well aligned.\n\nLGBM 3Kfold Not Shuffled\nCV: 2.084, LB: 1.526\nCV: 2.052, LB: 1.490\nCV: 2.014, LB: 1.433\nCV: 1.996, LB: 1.416 (only 6 features)\n\nWhile adding lots of features in the beginning, in the last step I reduced the features by a large margin and was able to increase the model accuracy. \n\nMy current highest LB score is a blend with a public kernel. As this seems to work well, I am now looking for a team up to blend our models and engineered features. \n\nIf you are interested, please post your validation strategy and corresponding CV/LB scores. \nI'd also love to get in touch with more experienced kagglers to guide me in the last 24 days of this competition. \n\nIlu", "529978": "> (only 6 features)\n\nNice!", "529983": "Thanks. I took your advice, uncle @cpmpml \nYou are probably not looking for an apprentice?", "529988": "I have already agreed to team, sorry. I hope you'll find good team mates.", "530045": "The more features I feed to my model, the better it gets (both CV and LB). I'm not surprised that you can achieve a very good score with few features, but surprised that your models are not further improving by adding more features.", "530069": "Congratulations on building a robust model with only few features. I agree with your approach", "530106": "Thank you, to everyone who sent me an invitation. I feel very flattered ;)\nI have decided to join @pukkinming and his team.", "530129": "I've had several offers but wanted to see how well I can do in this competition working alone. Maybe I'll regret my decision!", "530131": "First of all, you still have plenty of time to join a team. And doing something alone doesn't have to be bad. I did that the last 3 comps, but this time I decided to go the team route and learn something new :)", "530133": "There are plenty of people I'd be very happy to work with (I think you're in a fantastic team!) but I think for this competition I'll see what I can achieve by myself. Fingers crossed!", "530604": "only 6 features!\nCan I ask what feature you use?", "530643": "You may ask, but I obviously am not telling you before the competition end. Sorry.", "530991": "mine used similar umber of features, or even less, achieved 1.5LB using pure RNN. I usually like more on playing with model, not tricky features.", "531312": "Yes,I will do the same." }, "source": "meta" }