Followings things didn't work for me in this competition.
\nFollowings things didn't work for me in this competition.
\nI am really hoping at comp end we get a better picture of the problem on the LB because I have tried some things I didn’t expect to work at all and some that I thought were guaranteed to work and my results have occasionally been opposite lol
", "rawMarkdown": "I am really hoping at comp end we get a better picture of the problem on the LB because I have tried some things I didn’t expect to work at all and some that I thought were guaranteed to work and my results have occasionally been opposite lol ", "votes": 4 }, { "id": 2817512, "postDate": "2024-05-17T00:57:35.190Z", "content": "I had a pretty big list, but then found a bug in my submission script where I wasn't using all of the audio. so i'm back to square one.
", "rawMarkdown": "I had a pretty big list, but then found a bug in my submission script where I wasn't using all of the audio. so i'm back to square one.", "votes": 1 }, { "id": 2811896, "postDate": "2024-05-14T01:44:56.407Z", "content": "Interesting! Thanks for that list I will keep in mind not to waste time on those strategies!
", "rawMarkdown": "Interesting! Thanks for that list I will keep in mind not to waste time on those strategies!", "votes": 1 }, { "id": 2817379, "postDate": "2024-05-16T21:23:50.973Z", "content": "For me for now did not get good results with effb1. But so much unstability that it is difficult to tell what is not working as soon as it does not drop too much the model
", "rawMarkdown": "For me for now did not get good results with effb1. But so much unstability that it is difficult to tell what is not working as soon as it does not drop too much the model ", "votes": 2 }, { "id": 2812559, "postDate": "2024-05-14T09:34:10.247Z", "content": "I can confirm that oversampling doesn't work well here.
", "rawMarkdown": "I can confirm that oversampling doesn't work well here.", "votes": 2 }, { "id": 2811617, "postDate": "2024-05-13T19:44:44.817Z", "content": "Have you tried using torch.utils.data.WeightedRandomSampler?
that is oversampling mentioned in No. 6
", "rawMarkdown": "that is oversampling mentioned in No. 6", "votes": 1 } ] }, { "id": 2823056, "postDate": "2024-05-19T02:29:39.923Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 2824485, "postDate": "2024-05-19T19:33:12.750Z", "content": "Not so far for me
", "rawMarkdown": "Not so far for me" } ] } ], "comments": [ { "id": 2811980, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-05-14T03:45:59.177000", "content": "I am really hoping at comp end we get a better picture of the problem on the LB because I have tried some things I didn’t expect to work at all and some that I thought were guaranteed to work and my results have occasionally been opposite lol
", "votes": 4, "replies": [] }, { "id": 2817512, "author_name": "Will Rice", "author_url": "", "post_date": "2024-05-17T00:57:35.190000", "content": "I had a pretty big list, but then found a bug in my submission script where I wasn't using all of the audio. so i'm back to square one.
", "votes": 1, "replies": [] }, { "id": 2811896, "author_name": "MaxG6", "author_url": "", "post_date": "2024-05-14T01:44:56.407000", "content": "Interesting! Thanks for that list I will keep in mind not to waste time on those strategies!
", "votes": 1, "replies": [] }, { "id": 2817379, "author_name": "Shiro", "author_url": "", "post_date": "2024-05-16T21:23:50.973000", "content": "For me for now did not get good results with effb1. But so much unstability that it is difficult to tell what is not working as soon as it does not drop too much the model
", "votes": 2, "replies": [] }, { "id": 2812559, "author_name": "Araik Tamazian", "author_url": "", "post_date": "2024-05-14T09:34:10.247000", "content": "I can confirm that oversampling doesn't work well here.
", "votes": 2, "replies": [] }, { "id": 2811617, "author_name": "Araik Tamazian", "author_url": "", "post_date": "2024-05-13T19:44:44.817000", "content": "Have you tried using torch.utils.data.WeightedRandomSampler?
that is oversampling mentioned in No. 6
", "votes": 1, "replies": [] } ] }, { "id": 2823056, "author_name": "", "author_url": "", "post_date": "2024-05-19T02:29:39.923000", "content": "", "votes": 0, "replies": [ { "id": 2824485, "author_name": "Arindam Roy", "author_url": "", "post_date": "2024-05-19T19:33:12.750000", "content": "Not so far for me
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2810562": "Followings things didn't work for me in this competition.\n1. Arcface\n2. Lovasz Loss\n3. SED (5 sec training)\n4. Bestfitting's FCAN\n5. Puzzle CAM\n6. Oversampling / Undersampling\n7. Iterative Sampling in each epoch\n8. Google Bird Model based Distillation.\n\n\n", "2811980": "I am really hoping at comp end we get a better picture of the problem on the LB because I have tried some things I didn’t expect to work at all and some that I thought were guaranteed to work and my results have occasionally been opposite lol ", "2817512": "I had a pretty big list, but then found a bug in my submission script where I wasn't using all of the audio. so i'm back to square one.", "2811896": "Interesting! Thanks for that list I will keep in mind not to waste time on those strategies!", "2817379": "For me for now did not get good results with effb1. But so much unstability that it is difficult to tell what is not working as soon as it does not drop too much the model ", "2812559": "I can confirm that oversampling doesn't work well here.", "2811617": "Have you tried using `torch.utils.data.WeightedRandomSampler`?", "2823056": "" } }