I have been trying to train for fc for feature engineering. I am using 50 columns as features and 5 folds with lgb model. But my kernel dies after 8 hours or so. No luck so far. any tips? HELP!!!
", "messages": [ { "id": "568314", "postDate": "07/04/2019 17:41:06", "content": "I have been trying to train for fc for feature engineering. I am using 50 columns as features and 5 folds with lgb model. But my kernel dies after 8 hours or so. No luck so far. any tips? HELP!!!
", "rawMarkdown": "I have been trying to train for fc for feature engineering. I am using 50 columns as features and 5 folds with lgb model. But my kernel dies after 8 hours or so. No luck so far. any tips? HELP!!!", "votes": null }, { "id": "568321", "postDate": "07/04/2019 17:49:11", "content": "At the bottom of the kernel, you should be able to see the error that occurred. You probably ran out of time - you only get 9 hours per kernel session. Try reducing the n_estimators for lgbm.
", "rawMarkdown": "At the bottom of the kernel, you should be able to see the error that occurred. You probably ran out of time - you only get 9 hours per kernel session. Try reducing the n_estimators for lgbm.", "votes": null }, { "id": "568367", "postDate": "07/04/2019 19:40:34", "content": "LGB with 50 features should be finished in a flash - even for all data at a time\nEdit: didn't see about estimators at first. Yea it's possible. I think almost everyone are using somehow modified Andrew's model, and his n_estimators by default was 20k if I remember correctly
", "rawMarkdown": "LGB with 50 features should be finished in a flash - even for all data at a time\nEdit: didn't see about estimators at first. Yea it's possible. I think almost everyone are using somehow modified Andrew's model, and his n_estimators by default was 20k if I remember correctly", "votes": null }, { "id": "569470", "postDate": "07/06/2019 17:53:38", "content": "I tried reducing n_estimators and fold 4, my training completed in around 3 hours. Thanks for the help.
", "rawMarkdown": "I tried reducing n_estimators and fold 4, my training completed in around 3 hours. Thanks for the help.", "votes": null }, { "id": "569471", "postDate": "07/06/2019 17:55:18", "content": "@stanislavblinov How LGB can be finished in a flash? Mine is taking hours with n_estimator 3000. Maybe I am doing something wrong?
", "rawMarkdown": "stanislavblinov How LGB can be finished in a flash? Mine is taking hours with n_estimator 3000. Maybe I am doing something wrong?", "votes": null }, { "id": "569503", "postDate": "07/06/2019 19:16:07", "content": "I mean metaphorical \"flash\" of course. One of my models uses a lot of features and takes >40 hours.
", "rawMarkdown": "I mean metaphorical \"flash\" of course. One of my models uses a lot of features and takes >40 hours.", "votes": null } ], "comments": [ { "id": 568321, "author_name": "petersk20", "author_url": "", "post_date": "07/04/2019 17:49:11", "content": "At the bottom of the kernel, you should be able to see the error that occurred. You probably ran out of time - you only get 9 hours per kernel session. Try reducing the n_estimators for lgbm.
", "votes": null, "replies": [ { "id": 568367, "author_name": "stanislavblinov", "author_url": "", "post_date": "07/04/2019 19:40:34", "content": "LGB with 50 features should be finished in a flash - even for all data at a time\nEdit: didn't see about estimators at first. Yea it's possible. I think almost everyone are using somehow modified Andrew's model, and his n_estimators by default was 20k if I remember correctly
", "votes": null, "replies": [] }, { "id": 569470, "author_name": "iamsiraj", "author_url": "", "post_date": "07/06/2019 17:53:38", "content": "I tried reducing n_estimators and fold 4, my training completed in around 3 hours. Thanks for the help.
", "votes": null, "replies": [] }, { "id": 569471, "author_name": "iamsiraj", "author_url": "", "post_date": "07/06/2019 17:55:18", "content": "@stanislavblinov How LGB can be finished in a flash? Mine is taking hours with n_estimator 3000. Maybe I am doing something wrong?
", "votes": null, "replies": [] }, { "id": 569503, "author_name": "stanislavblinov", "author_url": "", "post_date": "07/06/2019 19:16:07", "content": "I mean metaphorical \"flash\" of course. One of my models uses a lot of features and takes >40 hours.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "568314": "I have been trying to train for fc for feature engineering. I am using 50 columns as features and 5 folds with lgb model. But my kernel dies after 8 hours or so. No luck so far. any tips? HELP!!!", "568321": "At the bottom of the kernel, you should be able to see the error that occurred. You probably ran out of time - you only get 9 hours per kernel session. Try reducing the n_estimators for lgbm.", "568367": "LGB with 50 features should be finished in a flash - even for all data at a time\nEdit: didn't see about estimators at first. Yea it's possible. I think almost everyone are using somehow modified Andrew's model, and his n_estimators by default was 20k if I remember correctly", "569470": "I tried reducing n_estimators and fold 4, my training completed in around 3 hours. Thanks for the help.", "569471": "stanislavblinov How LGB can be finished in a flash? Mine is taking hours with n_estimator 3000. Maybe I am doing something wrong?", "569503": "I mean metaphorical \"flash\" of course. One of my models uses a lot of features and takes >40 hours." }, "source": "meta" }