I measured my model's G(gini coefficient) and D(Default rate at 4%). Then there are remarkable result that G score is significantly higher than D score.
\nMine is
\nG: 0.9309\nD: 0.6873\nCV: 0.8091\nHow about yours? Do your model have low D and high G score same as mine? Can we handle this result with some techniques? I want to talk about this issue continuously :)
", "messages": [ { "id": 1898433, "postDate": "2022-08-14T15:28:26.123Z", "content": "If you look at the diagram in the Graphical explanation of the competition metric, it becomes obvious that D has to be lower than G.
", "rawMarkdown": "If you look at the diagram in the [Graphical explanation of the competition metric](https://www.kaggle.com/competitions/amex-default-prediction/discussion/327464), it becomes obvious that D has to be lower than G.", "votes": 4 }, { "id": 1899754, "postDate": "2022-08-15T13:35:52.560Z", "content": "Oh, there is already nice discussion and notebook. Thanks for sharing! @ambrosm
", "rawMarkdown": "Oh, there is already nice discussion and notebook. Thanks for sharing! @ambrosm ", "votes": 1 }, { "id": 1898338, "postDate": "2022-08-14T14:21:33.097Z", "content": "The relationship between my scores is similar, but the scores are lower for me. It seems that you have some kind of leakage in your validation process.
", "rawMarkdown": "The relationship between my scores is similar, but the scores are lower for me. It seems that you have some kind of leakage in your validation process.", "votes": 1, "replies": [ { "id": 1899761, "postDate": "2022-08-15T13:39:29.117Z", "content": "Oh, you saved me. I checked my process again and I knew that I preprocessed the train and valid data together.
", "rawMarkdown": "Oh, you saved me. I checked my process again and I knew that I preprocessed the train and valid data together. ", "votes": 1 } ] }, { "id": 1898224, "postDate": "2022-08-14T12:15:24.600Z", "content": "I measured my model's G(gini coefficient) and D(Default rate at 4%). Then there are remarkable result that G score is significantly higher than D score.
\nMine is
\nG: 0.9309\nD: 0.6873\nCV: 0.8091\nHow about yours? Do your model have low D and high G score same as mine? Can we handle this result with some techniques? I want to talk about this issue continuously :)
", "rawMarkdown": "I measured my model's G(gini coefficient) and D(Default rate at 4%). Then there are remarkable result that G score is significantly higher than D score.\n\nMine is\n```\nG: 0.9309\nD: 0.6873\nCV: 0.8091\n```\n\nHow about yours? Do your model have low D and high G score same as mine? Can we handle this result with some techniques? I want to talk about this issue continuously :)", "votes": 1 }, { "id": 1900936, "postDate": "2022-08-16T11:13:48.827Z", "content": "It is fine, the D is always lower than G by definition.
", "rawMarkdown": "It is fine, the D is always lower than G by definition.\n" }, { "id": 1898842, "postDate": "2022-08-14T21:47:47.503Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1898433, "author_name": "AmbrosM", "author_url": "", "post_date": "2022-08-14T15:28:26.123000", "content": "If you look at the diagram in the Graphical explanation of the competition metric, it becomes obvious that D has to be lower than G.
", "votes": 4, "replies": [] }, { "id": 1899754, "author_name": "L0Z1K", "author_url": "", "post_date": "2022-08-15T13:35:52.560000", "content": "Oh, there is already nice discussion and notebook. Thanks for sharing! @ambrosm
", "votes": 1, "replies": [] }, { "id": 1898338, "author_name": "Fritz Cremer", "author_url": "", "post_date": "2022-08-14T14:21:33.097000", "content": "The relationship between my scores is similar, but the scores are lower for me. It seems that you have some kind of leakage in your validation process.
", "votes": 1, "replies": [ { "id": 1899761, "author_name": "L0Z1K", "author_url": "", "post_date": "2022-08-15T13:39:29.117000", "content": "Oh, you saved me. I checked my process again and I knew that I preprocessed the train and valid data together.
", "votes": 1, "replies": [] } ] }, { "id": 1900936, "author_name": "The Devastator", "author_url": "", "post_date": "2022-08-16T11:13:48.827000", "content": "It is fine, the D is always lower than G by definition.
", "votes": 0, "replies": [] }, { "id": 1898842, "author_name": "", "author_url": "", "post_date": "2022-08-14T21:47:47.503000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1898433": "If you look at the diagram in the [Graphical explanation of the competition metric](https://www.kaggle.com/competitions/amex-default-prediction/discussion/327464), it becomes obvious that D has to be lower than G.", "1899754": "Oh, there is already nice discussion and notebook. Thanks for sharing! @ambrosm ", "1898338": "The relationship between my scores is similar, but the scores are lower for me. It seems that you have some kind of leakage in your validation process.", "1898224": "I measured my model's G(gini coefficient) and D(Default rate at 4%). Then there are remarkable result that G score is significantly higher than D score.\n\nMine is\n```\nG: 0.9309\nD: 0.6873\nCV: 0.8091\n```\n\nHow about yours? Do your model have low D and high G score same as mine? Can we handle this result with some techniques? I want to talk about this issue continuously :)", "1900936": "It is fine, the D is always lower than G by definition.\n", "1898842": "" } }