As far as I can see, some are clearer, with best definition than the others. What can we do to improve that?
", "messages": [ { "id": 895753, "postDate": "2020-06-21T15:28:42.130Z", "content": "As far as I can see, some are clearer, with best definition than the others. What can we do to improve that?
", "rawMarkdown": "As far as I can see, some are clearer, with best definition than the others. What can we do to improve that? ", "votes": 3 }, { "id": 901953, "postDate": "2020-06-25T20:11:04.960Z", "content": "I read Chris Notebook Yesterday and watched the workshop.
", "rawMarkdown": "I read Chris Notebook Yesterday and watched the workshop. ", "votes": 1 }, { "id": 901919, "postDate": "2020-06-25T19:40:32.057Z", "content": "In the Accelerator Power Hour for data science professionals with Kaggle Grandmasters, @cdeotte shared a very useful tip about how to increase accuracy for rare classes in post-processing, see Step 5 in his notebook How To Compete with GPUs Workshop.
", "rawMarkdown": "In the [Accelerator Power Hour for data science professionals with Kaggle Grandmasters](https://www.youtube.com/watch?v=DEuvGh4ZwaY), @cdeotte shared a very useful tip about how to increase accuracy for rare classes in post-processing, see Step 5 in his notebook [How To Compete with GPUs Workshop](https://www.kaggle.com/cdeotte/how-to-compete-with-gpus-workshop).\n", "votes": 1, "replies": [ { "id": 901987, "postDate": "2020-06-25T20:55:16.937Z", "content": "Thanks. I explain the method in detail in the discussion post here. This technique works very well when the competition metric is macro recall. You can try it with this competition's F1 metric, but it may not help.
", "rawMarkdown": "Thanks. I explain the method in detail in the discussion post [here][1]. This technique works very well when the competition metric is macro recall. You can try it with this competition's F1 metric, but it may not help.\n\n[1]: https://www.kaggle.com/c/bengaliai-cv19/discussion/136021", "votes": 2 }, { "id": 901996, "postDate": "2020-06-25T21:05:25.467Z", "content": "Thank you for explaining, Chris!
", "rawMarkdown": "Thank you for explaining, Chris!" }, { "id": 902084, "postDate": "2020-06-25T23:18:51.087Z", "content": "Thank you for your answer Chris. Since Martin Görner mentioned the definition of the Confusion Matrix in the previous Competition (flowers with TPU) I was curious about what can make it better (many of the Confusion Matrix weren't SO clear.) Dimitre Oliveira made excellent ones. But I couldn't identify which part of the codes could make that improvement. In fact, for someone that's beginning everything is difficult . Thanks for the post above (and all the others that helps this community).
", "rawMarkdown": "Thank you for your answer Chris. Since Martin Görner mentioned the definition of the Confusion Matrix in the previous Competition (flowers with TPU) I was curious about what can make it better (many of the Confusion Matrix weren't SO clear.) Dimitre Oliveira made excellent ones. But I couldn't identify which part of the codes could make that improvement. In fact, for someone that's beginning everything is difficult . Thanks for the post above (and all the others that helps this community)." }, { "id": 930566, "postDate": "2020-07-15T14:51:28.973Z", "content": "HI Chris, this is nice technique, by the way what is this approach called? is it some kind of calibration? is there any post processing steps for F1 score improvement in case if u know any..thanks
", "rawMarkdown": "HI Chris, this is nice technique, by the way what is this approach called? is it some kind of calibration? is there any post processing steps for F1 score improvement in case if u know any..thanks" } ] }, { "id": 897644, "postDate": "2020-06-23T02:26:34.960Z", "content": "I suppose that the best way would be to add extra training data for the categories that are most often misclassified.
", "rawMarkdown": "I suppose that the best way would be to add extra training data for the categories that are most often misclassified.", "votes": 1 }, { "id": 898500, "postDate": "2020-06-23T14:58:10.940Z", "content": "Thank you for answering it Lenka.
", "rawMarkdown": "Thank you for answering it Lenka." } ], "comments": [ { "id": 901953, "author_name": "Marília Prata", "author_url": "", "post_date": "2020-06-25T20:11:04.960000", "content": "I read Chris Notebook Yesterday and watched the workshop.
", "votes": 1, "replies": [] }, { "id": 901919, "author_name": "Lenka Čížková", "author_url": "", "post_date": "2020-06-25T19:40:32.057000", "content": "In the Accelerator Power Hour for data science professionals with Kaggle Grandmasters, @cdeotte shared a very useful tip about how to increase accuracy for rare classes in post-processing, see Step 5 in his notebook How To Compete with GPUs Workshop.
", "votes": 1, "replies": [ { "id": 901987, "author_name": "Chris Deotte", "author_url": "", "post_date": "2020-06-25T20:55:16.937000", "content": "Thanks. I explain the method in detail in the discussion post here. This technique works very well when the competition metric is macro recall. You can try it with this competition's F1 metric, but it may not help.
", "votes": 2, "replies": [] }, { "id": 901996, "author_name": "Lenka Čížková", "author_url": "", "post_date": "2020-06-25T21:05:25.467000", "content": "Thank you for explaining, Chris!
", "votes": 0, "replies": [] }, { "id": 902084, "author_name": "Marília Prata", "author_url": "", "post_date": "2020-06-25T23:18:51.087000", "content": "Thank you for your answer Chris. Since Martin Görner mentioned the definition of the Confusion Matrix in the previous Competition (flowers with TPU) I was curious about what can make it better (many of the Confusion Matrix weren't SO clear.) Dimitre Oliveira made excellent ones. But I couldn't identify which part of the codes could make that improvement. In fact, for someone that's beginning everything is difficult . Thanks for the post above (and all the others that helps this community).
", "votes": 0, "replies": [] }, { "id": 930566, "author_name": "Uday Kumar Gurugubelli", "author_url": "", "post_date": "2020-07-15T14:51:28.973000", "content": "HI Chris, this is nice technique, by the way what is this approach called? is it some kind of calibration? is there any post processing steps for F1 score improvement in case if u know any..thanks
", "votes": 0, "replies": [] } ] }, { "id": 897644, "author_name": "Lenka Čížková", "author_url": "", "post_date": "2020-06-23T02:26:34.960000", "content": "I suppose that the best way would be to add extra training data for the categories that are most often misclassified.
", "votes": 1, "replies": [] }, { "id": 898500, "author_name": "Marília Prata", "author_url": "", "post_date": "2020-06-23T14:58:10.940000", "content": "Thank you for answering it Lenka.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "895753": "As far as I can see, some are clearer, with best definition than the others. What can we do to improve that? ", "901953": "I read Chris Notebook Yesterday and watched the workshop. ", "901919": "In the [Accelerator Power Hour for data science professionals with Kaggle Grandmasters](https://www.youtube.com/watch?v=DEuvGh4ZwaY), @cdeotte shared a very useful tip about how to increase accuracy for rare classes in post-processing, see Step 5 in his notebook [How To Compete with GPUs Workshop](https://www.kaggle.com/cdeotte/how-to-compete-with-gpus-workshop).\n", "897644": "I suppose that the best way would be to add extra training data for the categories that are most often misclassified.", "898500": "Thank you for answering it Lenka." } }