Hi folks, I have tried to blend several submissions of different models but it failed. I am sure they can get a score when submitting them separately. Do you guys have some thoughts about it?
", "messages": [ { "id": "1002469", "postDate": "09/08/2020 06:43:55", "content": "Hi folks, I have tried to blend several submissions of different models but it failed. I am sure they can get a score when submitting them separately. Do you guys have some thoughts about it?
", "rawMarkdown": "Hi folks, I have tried to blend several submissions of different models but it failed. I am sure they can get a score when submitting them separately. Do you guys have some thoughts about it?", "votes": null }, { "id": "1002755", "postDate": "09/08/2020 12:03:14", "content": "Hi, are you talking about ensembling (bagging)? If yes, how did you do it?
\nI think bagging might help but it might also be too heavy because of the time limit.
No, just simple model averaging. And exactly the time limit is the reason why I choose to apply the ensemble process in a separate notebook.
", "rawMarkdown": "No, just simple model averaging. And exactly the time limit is the reason why I choose to apply the ensemble process in a separate notebook.", "votes": null }, { "id": "1004035", "postDate": "09/09/2020 12:15:44", "content": "\n\napply the ensemble process in a separate notebook.
\n
Are you talking about the training phase? Because you still have to predict for each model and then average and I think that the prediction for each model might be long.
\nIf I have the time to averaging several models, I'll put the results here.
It sounds like you have two notebooks that create submission files and you are trying to average them in a third notebook. That doesn't work in this contest. You do not have access to the true submission file from the first two notebooks. You can train your models in separate notebooks, but you have to copy the trained models into another notebook and rerun the predict step and average there.
", "rawMarkdown": "It sounds like you have two notebooks that create submission files and you are trying to average them in a third notebook. That doesn't work in this contest. You do not have access to the true submission file from the first two notebooks. You can train your models in separate notebooks, but you have to copy the trained models into another notebook and rerun the predict step and average there.", "votes": null }, { "id": "1004058", "postDate": "09/09/2020 12:40:00", "content": "I get it. Thanks for your comment.
", "rawMarkdown": "I get it. Thanks for your comment.", "votes": null }, { "id": "1004094", "postDate": "09/09/2020 13:00:03", "content": "It explaines why I didn't understand how you could reduce the time by using another notebook. If you succeed doing average modeling in one notebook, I would be please to have your results.
", "rawMarkdown": "It explaines why I didn't understand how you could reduce the time by using another notebook. If you succeed doing average modeling in one notebook, I would be please to have your results.", "votes": null }, { "id": "1004109", "postDate": "09/09/2020 13:17:33", "content": "I don’t manage it. I can only copy my models from another notebooks and then get my predictions and blend them. Directly blending submissions seems impossible.
", "rawMarkdown": "I don’t manage it. I can only copy my models from another notebooks and then get my predictions and blend them. Directly blending submissions seems impossible.", "votes": null }, { "id": "1004232", "postDate": "09/09/2020 14:51:36", "content": "Okay thanks anyway !
", "rawMarkdown": "Okay thanks anyway !", "votes": null } ], "comments": [ { "id": 1002755, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/08/2020 12:03:14", "content": "Hi, are you talking about ensembling (bagging)? If yes, how did you do it?
\nI think bagging might help but it might also be too heavy because of the time limit.
No, just simple model averaging. And exactly the time limit is the reason why I choose to apply the ensemble process in a separate notebook.
", "votes": null, "replies": [ { "id": 1004039, "author_name": "richardepstein", "author_url": "", "post_date": "09/09/2020 12:19:47", "content": "It sounds like you have two notebooks that create submission files and you are trying to average them in a third notebook. That doesn't work in this contest. You do not have access to the true submission file from the first two notebooks. You can train your models in separate notebooks, but you have to copy the trained models into another notebook and rerun the predict step and average there.
", "votes": null, "replies": [] }, { "id": 1004058, "author_name": "vincentrenzw", "author_url": "", "post_date": "09/09/2020 12:40:00", "content": "I get it. Thanks for your comment.
", "votes": null, "replies": [] }, { "id": 1004094, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/09/2020 13:00:03", "content": "It explaines why I didn't understand how you could reduce the time by using another notebook. If you succeed doing average modeling in one notebook, I would be please to have your results.
", "votes": null, "replies": [] }, { "id": 1004109, "author_name": "vincentrenzw", "author_url": "", "post_date": "09/09/2020 13:17:33", "content": "I don’t manage it. I can only copy my models from another notebooks and then get my predictions and blend them. Directly blending submissions seems impossible.
", "votes": null, "replies": [] }, { "id": 1004232, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/09/2020 14:51:36", "content": "Okay thanks anyway !
", "votes": null, "replies": [] } ] }, { "id": 1004035, "author_name": "yohannwattiez", "author_url": "", "post_date": "09/09/2020 12:15:44", "content": "\n\napply the ensemble process in a separate notebook.
\n
Are you talking about the training phase? Because you still have to predict for each model and then average and I think that the prediction for each model might be long.
\nIf I have the time to averaging several models, I'll put the results here.