Hey, We are currently having a best model's score of 3.676 on LB and local CV of 3.646 without post processing.\nOur current best model has 250+ features after feature selection.
\n\nLooking for one competitor with strong CV and LB score. Interested one's can ping me here.
", "messages": [ { "id": 462991, "postDate": "2019-01-29T07:47:16.453Z", "content": "Hey, We are currently having a best model's score of 3.676 on LB and local CV of 3.646 without post processing.\nOur current best model has 250+ features after feature selection.
\n\nLooking for one competitor with strong CV and LB score. Interested one's can ping me here.
", "rawMarkdown": "Hey, We are currently having a best model's score of **3.676 on LB** and **local CV of 3.646** without post processing.\nOur current best model has **250+ features** after feature selection.\n\nLooking for one competitor with strong CV and LB score. Interested one's can ping me here.", "votes": 3 }, { "id": 466895, "postDate": "2019-02-06T06:51:32.020Z", "content": "I have rank 170, with local CV as 3.655 and LB as 3.683, We are 2 people. Let me know if you are interested. Also, we are aiming for gold
", "rawMarkdown": "I have rank 170, with local CV as 3.655 and LB as 3.683, We are 2 people. Let me know if you are interested. Also, we are aiming for gold", "votes": 1, "replies": [ { "id": 466906, "postDate": "2019-02-06T07:23:12.367Z", "content": "@rahulin Can you send me a merge request? I can't track you on team search. What's your team name?
", "rawMarkdown": "@rahulin Can you send me a merge request? I can't track you on team search. What's your team name?" }, { "id": 467091, "postDate": "2019-02-06T12:44:19.807Z", "content": "@manu845 @finetuned\nMy team name is RR
", "rawMarkdown": "@manu845 @finetuned\nMy team name is RR", "votes": 1 }, { "id": 467113, "postDate": "2019-02-06T13:27:46.517Z", "content": "@rahulin \nis your 3.683 score a single model? or blend of outlier\\non outlier\\other models?\n with\\without post processing?
\n\nI have a pipeline that blends my single 3.695 model to 3.679 LB without post processing.\nI worked almost only on blending pipeline, and I'm looking for strong single modelers.
\n\nThanks
", "rawMarkdown": "@rahulin \nis your 3.683 score a single model? or blend of outlier\\non outlier\\other models?\n with\\without post processing?\n\nI have a pipeline that blends my single 3.695 model to 3.679 LB without post processing.\nI worked almost only on blending pipeline, and I'm looking for strong single modelers.\n\nThanks" }, { "id": 467162, "postDate": "2019-02-06T15:10:53.457Z", "content": "@AmirH @tpthegreat
\n\n3.683 is with post processing. Single model score was 3.688 (with 5 fold CV)
", "rawMarkdown": "@AmirH @tpthegreat\n\n3.683 is with post processing. Single model score was 3.688 (with 5 fold CV)", "votes": 1 }, { "id": 468673, "postDate": "2019-02-09T12:29:46.253Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 470493, "postDate": "2019-02-13T04:24:29.647Z", "content": "cv 3.63680\nlb 3.684 \na single model, 89 features after feature selection\ncould we have a talk
", "rawMarkdown": "cv 3.63680\nlb 3.684 \na single model, 89 features after feature selection\ncould we have a talk", "replies": [ { "id": 470536, "postDate": "2019-02-13T06:21:01.863Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 463496, "postDate": "2019-01-30T04:51:02.687Z", "content": "@manu845 How did you choose the features?
", "rawMarkdown": "@manu845 How did you choose the features?", "replies": [ { "id": 463521, "postDate": "2019-01-30T05:57:53.670Z", "content": "I used few feature selection methods described in the 'Less is more' discussion thread like boruta and some other personal tricks and methods majorly based on split and gain.
", "rawMarkdown": "I used few feature selection methods described in the 'Less is more' discussion thread like boruta and some other personal tricks and methods majorly based on split and gain. " } ] }, { "id": 463105, "postDate": "2019-01-29T12:09:49.173Z", "content": "I am behind, so you will not want to team up with me, but I have a question, that you might be willing to help me with.
\n\nYour LB-CV difference 0.032, but mine is is 0.44 (3.696 -3.652). I have appr. 160 features so that may be part of the problem. My lgb - parameters is:
\n\nparam = {'num_leaves': 32, \n 'min_data_in_leaf': 150, \n 'objective':'regression', \n 'max_depth': 5, \n 'learning_rate': 0.02, \n \"boosting\": \"gbdt\", \n \"feature_fraction\": 0.7522, \n \"bagging_freq\": 1, \n \"bagging_fraction\": 0.7083 , \n \"bagging_seed\": 11, \n \"metric\": 'rmse', \n \"lambda_l1\": 0.2634, \n \"random_state\": 133, \n \"nthread\": 6, \n \"verbosity\": -1}\nWould you suggest me to change any of these?
", "rawMarkdown": "I am behind, so you will not want to team up with me, but I have a question, that you might be willing to help me with.\n\nYour LB-CV difference 0.032, but mine is is 0.44 (3.696 -3.652). I have appr. 160 features so that may be part of the problem. My lgb - parameters is:\n\n param = {'num_leaves': 32, \n 'min_data_in_leaf': 150, \n 'objective':'regression', \n 'max_depth': 5, \n 'learning_rate': 0.02, \n \"boosting\": \"gbdt\", \n \"feature_fraction\": 0.7522, \n \"bagging_freq\": 1, \n \"bagging_fraction\": 0.7083 , \n \"bagging_seed\": 11, \n \"metric\": 'rmse', \n \"lambda_l1\": 0.2634, \n \"random_state\": 133, \n \"nthread\": 6, \n \"verbosity\": -1}\n\nWould you suggest me to change any of these?", "replies": [ { "id": 463151, "postDate": "2019-01-29T14:28:20.830Z", "content": "I think the value you've alloted for 'mindataleaf' is too regularizing, try using values in between 30-50.\nI set 'maxdepth' to -1; and 'baggingfreq' to 4. And one thing as a heads up is the model tends to change on a significant margin by changing the random state, seed and date time features, so better keep them on some fixed value or take average of outputs given by different seed values to get more stable results.
", "rawMarkdown": "I think the value you've alloted for 'mindataleaf' is too regularizing, try using values in between 30-50.\nI set 'maxdepth' to -1; and 'baggingfreq' to 4. And one thing as a heads up is the model tends to change on a significant margin by changing the random state, seed and date time features, so better keep them on some fixed value or take average of outputs given by different seed values to get more stable results.", "votes": 2 }, { "id": 463251, "postDate": "2019-01-29T17:17:33.913Z", "content": "Thank you very much. I will try it out.
", "rawMarkdown": "Thank you very much. I will try it out." } ] }, { "id": 463072, "postDate": "2019-01-29T11:06:54.823Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 463088, "postDate": "2019-01-29T11:37:09.140Z", "rawMarkdown": "", "isDeleted": true }, { "id": 463439, "postDate": "2019-01-30T01:35:23.027Z", "rawMarkdown": "", "isDeleted": true }, { "id": 463523, "postDate": "2019-01-30T06:00:23.163Z", "rawMarkdown": "", "isDeleted": true }, { "id": 464066, "postDate": "2019-01-31T06:24:18.983Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 466895, "author_name": "Rahul Agarwal", "author_url": "", "post_date": "2019-02-06T06:51:32.020000", "content": "I have rank 170, with local CV as 3.655 and LB as 3.683, We are 2 people. Let me know if you are interested. Also, we are aiming for gold
", "votes": 1, "replies": [ { "id": 466906, "author_name": "PiGram", "author_url": "", "post_date": "2019-02-06T07:23:12.367000", "content": "@rahulin Can you send me a merge request? I can't track you on team search. What's your team name?
", "votes": 0, "replies": [] }, { "id": 467091, "author_name": "Rahul Agarwal", "author_url": "", "post_date": "2019-02-06T12:44:19.807000", "content": "@manu845 @finetuned\nMy team name is RR
", "votes": 1, "replies": [] }, { "id": 467113, "author_name": "AmirH", "author_url": "", "post_date": "2019-02-06T13:27:46.517000", "content": "@rahulin \nis your 3.683 score a single model? or blend of outlier\\non outlier\\other models?\n with\\without post processing?
\n\nI have a pipeline that blends my single 3.695 model to 3.679 LB without post processing.\nI worked almost only on blending pipeline, and I'm looking for strong single modelers.
\n\nThanks
", "votes": 0, "replies": [] }, { "id": 467162, "author_name": "Rahul Agarwal", "author_url": "", "post_date": "2019-02-06T15:10:53.457000", "content": "@AmirH @tpthegreat
\n\n3.683 is with post processing. Single model score was 3.688 (with 5 fold CV)
", "votes": 1, "replies": [] }, { "id": 468673, "author_name": "", "author_url": "", "post_date": "2019-02-09T12:29:46.253000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 470493, "author_name": "Link", "author_url": "", "post_date": "2019-02-13T04:24:29.647000", "content": "cv 3.63680\nlb 3.684 \na single model, 89 features after feature selection\ncould we have a talk
", "votes": 0, "replies": [ { "id": 470536, "author_name": "", "author_url": "", "post_date": "2019-02-13T06:21:01.863000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 463496, "author_name": "Aditya Vyas", "author_url": "", "post_date": "2019-01-30T04:51:02.687000", "content": "@manu845 How did you choose the features?
", "votes": 0, "replies": [ { "id": 463521, "author_name": "PiGram", "author_url": "", "post_date": "2019-01-30T05:57:53.670000", "content": "I used few feature selection methods described in the 'Less is more' discussion thread like boruta and some other personal tricks and methods majorly based on split and gain.
", "votes": 0, "replies": [] } ] }, { "id": 463105, "author_name": "PeterSorensen", "author_url": "", "post_date": "2019-01-29T12:09:49.173000", "content": "I am behind, so you will not want to team up with me, but I have a question, that you might be willing to help me with.
\n\nYour LB-CV difference 0.032, but mine is is 0.44 (3.696 -3.652). I have appr. 160 features so that may be part of the problem. My lgb - parameters is:
\n\nparam = {'num_leaves': 32, \n 'min_data_in_leaf': 150, \n 'objective':'regression', \n 'max_depth': 5, \n 'learning_rate': 0.02, \n \"boosting\": \"gbdt\", \n \"feature_fraction\": 0.7522, \n \"bagging_freq\": 1, \n \"bagging_fraction\": 0.7083 , \n \"bagging_seed\": 11, \n \"metric\": 'rmse', \n \"lambda_l1\": 0.2634, \n \"random_state\": 133, \n \"nthread\": 6, \n \"verbosity\": -1}\nWould you suggest me to change any of these?
", "votes": 0, "replies": [ { "id": 463151, "author_name": "PiGram", "author_url": "", "post_date": "2019-01-29T14:28:20.830000", "content": "I think the value you've alloted for 'mindataleaf' is too regularizing, try using values in between 30-50.\nI set 'maxdepth' to -1; and 'baggingfreq' to 4. And one thing as a heads up is the model tends to change on a significant margin by changing the random state, seed and date time features, so better keep them on some fixed value or take average of outputs given by different seed values to get more stable results.
", "votes": 2, "replies": [] }, { "id": 463251, "author_name": "PeterSorensen", "author_url": "", "post_date": "2019-01-29T17:17:33.913000", "content": "Thank you very much. I will try it out.
", "votes": 0, "replies": [] } ] }, { "id": 463072, "author_name": "", "author_url": "", "post_date": "2019-01-29T11:06:54.823000", "content": "", "votes": 0, "replies": [ { "id": 463088, "author_name": "", "author_url": "", "post_date": "2019-01-29T11:37:09.140000", "content": "", "votes": 0, "replies": [] }, { "id": 463439, "author_name": "", "author_url": "", "post_date": "2019-01-30T01:35:23.027000", "content": "", "votes": 0, "replies": [] }, { "id": 463523, "author_name": "", "author_url": "", "post_date": "2019-01-30T06:00:23.163000", "content": "", "votes": 0, "replies": [] }, { "id": 464066, "author_name": "", "author_url": "", "post_date": "2019-01-31T06:24:18.983000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "462991": "Hey, We are currently having a best model's score of **3.676 on LB** and **local CV of 3.646** without post processing.\nOur current best model has **250+ features** after feature selection.\n\nLooking for one competitor with strong CV and LB score. Interested one's can ping me here.", "466895": "I have rank 170, with local CV as 3.655 and LB as 3.683, We are 2 people. Let me know if you are interested. Also, we are aiming for gold", "470493": "cv 3.63680\nlb 3.684 \na single model, 89 features after feature selection\ncould we have a talk", "463496": "@manu845 How did you choose the features?", "463105": "I am behind, so you will not want to team up with me, but I have a question, that you might be willing to help me with.\n\nYour LB-CV difference 0.032, but mine is is 0.44 (3.696 -3.652). I have appr. 160 features so that may be part of the problem. My lgb - parameters is:\n\n param = {'num_leaves': 32, \n 'min_data_in_leaf': 150, \n 'objective':'regression', \n 'max_depth': 5, \n 'learning_rate': 0.02, \n \"boosting\": \"gbdt\", \n \"feature_fraction\": 0.7522, \n \"bagging_freq\": 1, \n \"bagging_fraction\": 0.7083 , \n \"bagging_seed\": 11, \n \"metric\": 'rmse', \n \"lambda_l1\": 0.2634, \n \"random_state\": 133, \n \"nthread\": 6, \n \"verbosity\": -1}\n\nWould you suggest me to change any of these?", "463072": "" } }