If you didn't see it yet, in this kernel I have generated 831 features. Some of the code is commented out because otherwise the kernel would hit time limit.\nSome of ideas of these features were mine, some were borrowed from other public kernels (I hope I was able to reference all of them).\nAll the generated features (including those, which were commented) are here: https://www.kaggle.com/artgor/lanl-features
\n\nAlso I saw this interesting kernel with signal denoising: https://www.kaggle.com/tarunpaparaju/lanl-earthquake-prediction-signal-denoising\nSo I also denoised all the data and created the same features for it.\nAs a result there are 1900+ features in total. Obviously using all of them in one model isn't a good idea. It is necessary to select the most important features or use some subsets of features in separate models.\nIn this kernel you can see some approaches to feature selection.\nI wanted to use SHAP, but it takes too much time on such number of features :)
", "messages": [ { "id": 523870, "postDate": "2019-04-27T09:05:40.413Z", "content": "If you didn't see it yet, in this kernel I have generated 831 features. Some of the code is commented out because otherwise the kernel would hit time limit.\nSome of ideas of these features were mine, some were borrowed from other public kernels (I hope I was able to reference all of them).\nAll the generated features (including those, which were commented) are here: https://www.kaggle.com/artgor/lanl-features
\n\nAlso I saw this interesting kernel with signal denoising: https://www.kaggle.com/tarunpaparaju/lanl-earthquake-prediction-signal-denoising\nSo I also denoised all the data and created the same features for it.\nAs a result there are 1900+ features in total. Obviously using all of them in one model isn't a good idea. It is necessary to select the most important features or use some subsets of features in separate models.\nIn this kernel you can see some approaches to feature selection.\nI wanted to use SHAP, but it takes too much time on such number of features :)
", "rawMarkdown": "If you didn't see it yet, in this [kernel](https://www.kaggle.com/artgor/even-more-features) I have generated 831 features. Some of the code is commented out because otherwise the kernel would hit time limit.\nSome of ideas of these features were mine, some were borrowed from other public kernels (I hope I was able to reference all of them).\nAll the generated features (including those, which were commented) are here: https://www.kaggle.com/artgor/lanl-features\n\nAlso I saw this interesting kernel with signal denoising: https://www.kaggle.com/tarunpaparaju/lanl-earthquake-prediction-signal-denoising\nSo I also denoised all the data and created the same features for it.\nAs a result there are 1900+ features in total. Obviously using all of them in one model isn't a good idea. It is necessary to select the most important features or use some subsets of features in separate models.\nIn this [kernel](https://www.kaggle.com/artgor/feature-selection-model-interpretation-and-more) you can see some approaches to feature selection.\nI wanted to use SHAP, but it takes too much time on such number of features :)", "votes": 63 } ], "comments": [], "raw_markdown_by_id": { "523870": "If you didn't see it yet, in this [kernel](https://www.kaggle.com/artgor/even-more-features) I have generated 831 features. Some of the code is commented out because otherwise the kernel would hit time limit.\nSome of ideas of these features were mine, some were borrowed from other public kernels (I hope I was able to reference all of them).\nAll the generated features (including those, which were commented) are here: https://www.kaggle.com/artgor/lanl-features\n\nAlso I saw this interesting kernel with signal denoising: https://www.kaggle.com/tarunpaparaju/lanl-earthquake-prediction-signal-denoising\nSo I also denoised all the data and created the same features for it.\nAs a result there are 1900+ features in total. Obviously using all of them in one model isn't a good idea. It is necessary to select the most important features or use some subsets of features in separate models.\nIn this [kernel](https://www.kaggle.com/artgor/feature-selection-model-interpretation-and-more) you can see some approaches to feature selection.\nI wanted to use SHAP, but it takes too much time on such number of features :)" } }