Simple average is giving more higher Public LB score rather than weighted average (Blending).[Probably]\nI blended 6 of my submission files and got a score of 0.9553. When I averaged those files, I got public LB score of 0.9578 😊 \nHope that, it will work for you also.
", "messages": [ { "id": 940575, "postDate": "2020-07-23T03:12:06.527Z", "content": "Simple average is giving more higher Public LB score rather than weighted average (Blending).[Probably]\nI blended 6 of my submission files and got a score of 0.9553. When I averaged those files, I got public LB score of 0.9578 😊 \nHope that, it will work for you also.
", "rawMarkdown": "Simple average is giving more higher Public LB score rather than weighted average (Blending).[Probably]\nI ***blended*** 6 of my submission files and got a score of **0.9553.** When I ***averaged*** those files, I got public LB score of **0.9578** 😊 \nHope that, it will work for you also. ", "votes": 6 }, { "id": 940585, "postDate": "2020-07-23T03:24:40.663Z", "content": "Thanks for sharing this. How about your CV score？
", "rawMarkdown": "Thanks for sharing this. How about your CV score？", "votes": 1, "replies": [ { "id": 940594, "postDate": "2020-07-23T03:33:50.527Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 940608, "postDate": "2020-07-23T03:47:52.023Z", "content": "Thanks for your sharing @mdfahimreshm! I am a beginner and would you mind telling me how to get the CV score of the averaged models? I am sorry if that question is silly.
", "rawMarkdown": "Thanks for your sharing @mdfahimreshm! I am a beginner and would you mind telling me how to get the CV score of the averaged models? I am sorry if that question is silly." }, { "id": 940638, "postDate": "2020-07-23T04:09:31.967Z", "content": "@vincentren1997 I averaged in the submission files.\n Just like\n sub_avg['target'] = (sub_1['target']+sub_2['target'])/2\nwhere sub_1 and sub_2 are from different single models with different CV score of its own. I only know doing CV in one model not in multiple model.
@mdfahimreshm Thanks!
", "rawMarkdown": "@mdfahimreshm Thanks!" } ] }, { "id": 940617, "postDate": "2020-07-23T03:52:45.370Z", "content": "Ohhh.... alright thanks!\nBut that might be somewhat of an overfitting, have you tried training a meta classifier over your different models? It seems like a better idea, rather than plainly relying on the averages.
", "rawMarkdown": "Ohhh.... alright thanks!\nBut that might be somewhat of an overfitting, have you tried training a meta classifier over your different models? It seems like a better idea, rather than plainly relying on the averages.", "replies": [ { "id": 940628, "postDate": "2020-07-23T04:01:22.093Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ] }, { "id": 940581, "postDate": "2020-07-23T03:19:28.237Z", "content": "hi, @mdfahimreshm thanks for share this.\nwhat is your best single score on lb in these subs?
", "rawMarkdown": "hi, @mdfahimreshm thanks for share this.\nwhat is your best single score on lb in these subs?", "replies": [ { "id": 940587, "postDate": "2020-07-23T03:25:28.613Z", "content": "Best single submission score was 0.9544
", "rawMarkdown": "Best single submission score was 0.9544", "votes": 2 } ] } ], "comments": [ { "id": 940585, "author_name": "Zhiwei Ren", "author_url": "", "post_date": "2020-07-23T03:24:40.663000", "content": "Thanks for sharing this. How about your CV score？
", "votes": 1, "replies": [ { "id": 940594, "author_name": "", "author_url": "", "post_date": "2020-07-23T03:33:50.527000", "content": "", "votes": 1, "replies": [] }, { "id": 940608, "author_name": "vctn", "author_url": "", "post_date": "2020-07-23T03:47:52.023000", "content": "Thanks for your sharing @mdfahimreshm! I am a beginner and would you mind telling me how to get the CV score of the averaged models? I am sorry if that question is silly.
", "votes": 0, "replies": [] }, { "id": 940638, "author_name": "Md Fahim", "author_url": "", "post_date": "2020-07-23T04:09:31.967000", "content": "@vincentren1997 I averaged in the submission files.\n Just like\n sub_avg['target'] = (sub_1['target']+sub_2['target'])/2\nwhere sub_1 and sub_2 are from different single models with different CV score of its own. I only know doing CV in one model not in multiple model.
@mdfahimreshm Thanks!
", "votes": 0, "replies": [] } ] }, { "id": 940617, "author_name": "Saransh Agarwal", "author_url": "", "post_date": "2020-07-23T03:52:45.370000", "content": "Ohhh.... alright thanks!\nBut that might be somewhat of an overfitting, have you tried training a meta classifier over your different models? It seems like a better idea, rather than plainly relying on the averages.
", "votes": 0, "replies": [ { "id": 940628, "author_name": "", "author_url": "", "post_date": "2020-07-23T04:01:22.093000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 940581, "author_name": "yunxi", "author_url": "", "post_date": "2020-07-23T03:19:28.237000", "content": "hi, @mdfahimreshm thanks for share this.\nwhat is your best single score on lb in these subs?
", "votes": 0, "replies": [ { "id": 940587, "author_name": "Md Fahim", "author_url": "", "post_date": "2020-07-23T03:25:28.613000", "content": "Best single submission score was 0.9544
", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "940575": "Simple average is giving more higher Public LB score rather than weighted average (Blending).[Probably]\nI ***blended*** 6 of my submission files and got a score of **0.9553.** When I ***averaged*** those files, I got public LB score of **0.9578** 😊 \nHope that, it will work for you also. ", "940585": "Thanks for sharing this. How about your CV score？", "940617": "Ohhh.... alright thanks!\nBut that might be somewhat of an overfitting, have you tried training a meta classifier over your different models? It seems like a better idea, rather than plainly relying on the averages.", "940581": "hi, @mdfahimreshm thanks for share this.\nwhat is your best single score on lb in these subs?" } }