Hi guys! Could you share what is your LB submitting your results without 'new _ whales' on the csv? We are getting 0.611 without 'new _ whales' and after choose a threshold 0.830, and you?
", "messages": [ { "id": 455310, "postDate": "2019-01-13T15:12:16.793Z", "content": "I would not share, but new_whales are at least 1/3, based on the score of the sample_submission. And identifying them correctly is one of the most important challenges of the current challenge :)
", "rawMarkdown": "I would not share, but new_whales are at least 1/3, based on the score of the sample_submission. And identifying them correctly is one of the most important challenges of the current challenge :)", "votes": 3 }, { "id": 455450, "postDate": "2019-01-13T23:40:25.447Z", "content": "I've not tried without new_whale, but my score has ranged from .73 to .86 based on the threshold.
", "rawMarkdown": "I've not tried without new_whale, but my score has ranged from .73 to .86 based on the threshold.", "votes": 2, "replies": [ { "id": 456310, "postDate": "2019-01-15T14:34:34.590Z", "content": "You had experienced that Siamese its too 'inestable'? Epoch 65 LB 0.586.... Epoch 75 LB 0.572
", "rawMarkdown": "You had experienced that Siamese its too 'inestable'? Epoch 65 LB 0.586.... Epoch 75 LB 0.572" }, { "id": 461247, "postDate": "2019-01-25T16:05:56.600Z", "content": "you tested finally? :D
", "rawMarkdown": "you tested finally? :D" }, { "id": 461328, "postDate": "2019-01-25T19:08:12.333Z", "content": "Still running the model, it is pretty complex and taking many days to train. It is only at .620 locally so I'm guessing it will be 4 or 5 more days.
", "rawMarkdown": "Still running the model, it is pretty complex and taking many days to train. It is only at .620 locally so I'm guessing it will be 4 or 5 more days." }, { "id": 465705, "postDate": "2019-02-03T20:18:07.110Z", "content": "I don’t know if I would call the Siamese unstable per se, mine can bounce between +/- 0.0075 lb score after retraining for one epoch.
\n\nI believe this variability is coming from marginal images and likely mislabeling. About half of the images that my model has the most difficulty with are poor enough that I would exclude them as bad data, though we don’t have that luxury here.
", "rawMarkdown": "I don’t know if I would call the Siamese unstable per se, mine can bounce between +/- 0.0075 lb score after retraining for one epoch. \n\nI believe this variability is coming from marginal images and likely mislabeling. About half of the images that my model has the most difficulty with are poor enough that I would exclude them as bad data, though we don’t have that luxury here. " } ] }, { "id": 455264, "postDate": "2019-01-13T12:33:14.837Z", "content": "Hi guys! Could you share what is your LB submitting your results without 'new _ whales' on the csv? We are getting 0.611 without 'new _ whales' and after choose a threshold 0.830, and you?
", "rawMarkdown": "Hi guys! Could you share what is your LB submitting your results without 'new _ whales' on the csv? We are getting 0.611 without 'new _ whales' and after choose a threshold 0.830, and you?", "votes": 2 }, { "id": 483178, "postDate": "2019-03-04T09:11:51.633Z", "content": "Have you tried checking your score without new_whales in a late submission? I ended up with 0.68112 / 0.68950 - private / public.
", "rawMarkdown": "Have you tried checking your score without new_whales in a late submission? I ended up with 0.68112 / 0.68950 - private / public.", "replies": [ { "id": 483296, "postDate": "2019-03-04T12:53:19.673Z", "content": "My score for a single model is : 0.67574 / 0.68264 - private / public.
", "rawMarkdown": "My score for a single model is : 0.67574 / 0.68264 - private / public. ", "votes": 1 } ] }, { "id": 456093, "postDate": "2019-01-15T06:01:05.863Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 456205, "postDate": "2019-01-15T10:53:24.570Z", "content": "For now I choose the threshold manually. You can see an example https://www.kaggle.com/seesee/siamese-pretrained-0-822/output
", "rawMarkdown": "For now I choose the threshold manually. You can see an example https://www.kaggle.com/seesee/siamese-pretrained-0-822/output" } ] } ], "comments": [ { "id": 455310, "author_name": "old-ufo", "author_url": "", "post_date": "2019-01-13T15:12:16.793000", "content": "I would not share, but new_whales are at least 1/3, based on the score of the sample_submission. And identifying them correctly is one of the most important challenges of the current challenge :)
", "votes": 3, "replies": [] }, { "id": 455450, "author_name": "Brian", "author_url": "", "post_date": "2019-01-13T23:40:25.447000", "content": "I've not tried without new_whale, but my score has ranged from .73 to .86 based on the threshold.
", "votes": 2, "replies": [ { "id": 456310, "author_name": "Mario Parreño Lara", "author_url": "", "post_date": "2019-01-15T14:34:34.590000", "content": "You had experienced that Siamese its too 'inestable'? Epoch 65 LB 0.586.... Epoch 75 LB 0.572
", "votes": 0, "replies": [] }, { "id": 461247, "author_name": "Mario Parreño Lara", "author_url": "", "post_date": "2019-01-25T16:05:56.600000", "content": "you tested finally? :D
", "votes": 0, "replies": [] }, { "id": 461328, "author_name": "Brian", "author_url": "", "post_date": "2019-01-25T19:08:12.333000", "content": "Still running the model, it is pretty complex and taking many days to train. It is only at .620 locally so I'm guessing it will be 4 or 5 more days.
", "votes": 0, "replies": [] }, { "id": 465705, "author_name": "interneuron", "author_url": "", "post_date": "2019-02-03T20:18:07.110000", "content": "I don’t know if I would call the Siamese unstable per se, mine can bounce between +/- 0.0075 lb score after retraining for one epoch.
\n\nI believe this variability is coming from marginal images and likely mislabeling. About half of the images that my model has the most difficulty with are poor enough that I would exclude them as bad data, though we don’t have that luxury here.
", "votes": 0, "replies": [] } ] }, { "id": 483178, "author_name": "Niclas Stoltenberg", "author_url": "", "post_date": "2019-03-04T09:11:51.633000", "content": "Have you tried checking your score without new_whales in a late submission? I ended up with 0.68112 / 0.68950 - private / public.
", "votes": 0, "replies": [ { "id": 483296, "author_name": "Bartek", "author_url": "", "post_date": "2019-03-04T12:53:19.673000", "content": "My score for a single model is : 0.67574 / 0.68264 - private / public.
", "votes": 1, "replies": [] } ] }, { "id": 456093, "author_name": "", "author_url": "", "post_date": "2019-01-15T06:01:05.863000", "content": "", "votes": 0, "replies": [ { "id": 456205, "author_name": "Mario Parreño Lara", "author_url": "", "post_date": "2019-01-15T10:53:24.570000", "content": "For now I choose the threshold manually. You can see an example https://www.kaggle.com/seesee/siamese-pretrained-0-822/output
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "455310": "I would not share, but new_whales are at least 1/3, based on the score of the sample_submission. And identifying them correctly is one of the most important challenges of the current challenge :)", "455450": "I've not tried without new_whale, but my score has ranged from .73 to .86 based on the threshold.", "455264": "Hi guys! Could you share what is your LB submitting your results without 'new _ whales' on the csv? We are getting 0.611 without 'new _ whales' and after choose a threshold 0.830, and you?", "483178": "Have you tried checking your score without new_whales in a late submission? I ended up with 0.68112 / 0.68950 - private / public.", "456093": "" } }