I trained some models with a result of 0.97 and tried to combine the model predictions and although the score did not improve, the ranking went from 132->77
", "messages": [ { "id": 1323412, "postDate": "2021-05-26T08:04:48.237Z", "content": "I trained some models with a result of 0.97 and tried to combine the model predictions and although the score did not improve, the ranking went from 132->77
", "rawMarkdown": "I trained some models with a result of 0.97 and tried to combine the model predictions and although the score did not improve, the ranking went from 132->77", "votes": 8 }, { "id": 1323448, "postDate": "2021-05-26T08:31:26.160Z", "content": "Hey Can you please private the notebook. Releasing such a high scoring notebook is very problematic to the people who have a lower than rank than you. Many people can Just copy your code and can get in the medal range . You can use lower scoring models with the same code which does not have the same effect.Recently a person posted a public notebook which can get you into the medal range which got 35 copy and edit. This affects people who work honestly and their rank goes down because of public notebook.
", "rawMarkdown": "Hey Can you please private the notebook. Releasing such a high scoring notebook is very problematic to the people who have a lower than rank than you. Many people can Just copy your code and can get in the medal range . You can use lower scoring models with the same code which does not have the same effect.Recently a person posted a public notebook which can get you into the medal range which got 35 copy and edit. This affects people who work honestly and their rank goes down because of public notebook. ", "votes": 6, "replies": [ { "id": 1323461, "postDate": "2021-05-26T08:43:07.550Z", "content": "Thank you for the suggestion, I will make the data used private
", "rawMarkdown": "Thank you for the suggestion, I will make the data used private", "votes": 5 }, { "id": 1323467, "postDate": "2021-05-26T08:47:18.640Z", "content": "Thank You very Much. Also please the part where you create the submission.csv cause they just download that and submit it .
", "rawMarkdown": "Thank You very Much. Also please the part where you create the submission.csv cause they just download that and submit it .", "votes": 3 }, { "id": 1323688, "postDate": "2021-05-26T11:57:05.927Z", "content": "Thank you for the suggestion,
", "rawMarkdown": "Thank you for the suggestion,", "votes": 2 }, { "id": 1323787, "postDate": "2021-05-26T12:48:29.683Z", "content": "I personally was disappointed to see an interesting public notebook (that I had upvoted) suddenly disappear. From what I observed during its brief appearance, the overall design was rather similar to another notebook that has been there for some time with no problems. The essential idea was to form a linear blend of different original sources. In fact, I think @xuxu1234 took the interesting further step of including the output of the previous ensembling in the new notebook. In my view, it's rather stretching things to imagine that a notebook that ranks ~82nd two months out from the end of the competition would get anyone a medal in its present form, and there is already at least one public kernel that demonstrates a similar ensembling technique. The prohibition on sharing high-scoring code is normally limited to the last week of a competition. I say let @xuxu1234 keep the Notebook upvotes.
", "rawMarkdown": "I personally was disappointed to see an interesting public notebook (that I had upvoted) suddenly disappear. From what I observed during its brief appearance, the overall design was rather similar to another notebook that has been there for some time with no problems. The essential idea was to form a linear blend of different original sources. In fact, I think @xuxu1234 took the interesting further step of including the output of the previous ensembling in the new notebook. In my view, it's rather stretching things to imagine that a notebook that ranks ~82nd two months out from the end of the competition would get anyone a medal in its present form, and there is already at least one public kernel that demonstrates a similar ensembling technique. The prohibition on sharing high-scoring code is normally limited to the last week of a competition. I say let @xuxu1234 keep the Notebook upvotes.", "votes": 3 }, { "id": 1323807, "postDate": "2021-05-26T12:59:47.093Z", "content": "We already have notebook 90 percent similar to this one.
", "rawMarkdown": "We already have notebook 90 percent similar to this one." }, { "id": 1323822, "postDate": "2021-05-26T13:06:12.180Z", "content": "@mithilsalunkhe I absolutely agree with you on the statement of fact, but if all notebooks of >90% similarity to one another were weeded out there would be many fewer on display. That @xuxu1234 was able to get such a high score with a simple model was nonetheless of interest.
", "rawMarkdown": "@mithilsalunkhe I absolutely agree with you on the statement of fact, but if all notebooks of >90% similarity to one another were weeded out there would be many fewer on display. That @xuxu1234 was able to get such a high score with a simple model was nonetheless of interest. ", "votes": 1 }, { "id": 1323939, "postDate": "2021-05-26T14:11:58.683Z", "content": "Yes, I think now that the model is fitting the raw data to a certain extent, can we do some effective data augmentation to make the results better, and I have tried Focal loss in effb1, which improves the score by 0.1, but deepening the network does not improve the results
", "rawMarkdown": "Yes, I think now that the model is fitting the raw data to a certain extent, can we do some effective data augmentation to make the results better, and I have tried Focal loss in effb1, which improves the score by 0.1, but deepening the network does not improve the results", "votes": 2 } ] } ], "comments": [ { "id": 1323448, "author_name": "Mithil Salunkhe", "author_url": "", "post_date": "2021-05-26T08:31:26.160000", "content": "Hey Can you please private the notebook. Releasing such a high scoring notebook is very problematic to the people who have a lower than rank than you. Many people can Just copy your code and can get in the medal range . You can use lower scoring models with the same code which does not have the same effect.Recently a person posted a public notebook which can get you into the medal range which got 35 copy and edit. This affects people who work honestly and their rank goes down because of public notebook.
", "votes": 6, "replies": [ { "id": 1323461, "author_name": "xuxu_sky", "author_url": "", "post_date": "2021-05-26T08:43:07.550000", "content": "Thank you for the suggestion, I will make the data used private
", "votes": 5, "replies": [] }, { "id": 1323467, "author_name": "Mithil Salunkhe", "author_url": "", "post_date": "2021-05-26T08:47:18.640000", "content": "Thank You very Much. Also please the part where you create the submission.csv cause they just download that and submit it .
", "votes": 3, "replies": [] }, { "id": 1323688, "author_name": "xuxu_sky", "author_url": "", "post_date": "2021-05-26T11:57:05.927000", "content": "Thank you for the suggestion,
", "votes": 2, "replies": [] }, { "id": 1323787, "author_name": "John Mitchell", "author_url": "", "post_date": "2021-05-26T12:48:29.683000", "content": "I personally was disappointed to see an interesting public notebook (that I had upvoted) suddenly disappear. From what I observed during its brief appearance, the overall design was rather similar to another notebook that has been there for some time with no problems. The essential idea was to form a linear blend of different original sources. In fact, I think @xuxu1234 took the interesting further step of including the output of the previous ensembling in the new notebook. In my view, it's rather stretching things to imagine that a notebook that ranks ~82nd two months out from the end of the competition would get anyone a medal in its present form, and there is already at least one public kernel that demonstrates a similar ensembling technique. The prohibition on sharing high-scoring code is normally limited to the last week of a competition. I say let @xuxu1234 keep the Notebook upvotes.
", "votes": 3, "replies": [] }, { "id": 1323807, "author_name": "Mithil Salunkhe", "author_url": "", "post_date": "2021-05-26T12:59:47.093000", "content": "We already have notebook 90 percent similar to this one.
", "votes": 0, "replies": [] }, { "id": 1323822, "author_name": "John Mitchell", "author_url": "", "post_date": "2021-05-26T13:06:12.180000", "content": "@mithilsalunkhe I absolutely agree with you on the statement of fact, but if all notebooks of >90% similarity to one another were weeded out there would be many fewer on display. That @xuxu1234 was able to get such a high score with a simple model was nonetheless of interest.
", "votes": 1, "replies": [] }, { "id": 1323939, "author_name": "xuxu_sky", "author_url": "", "post_date": "2021-05-26T14:11:58.683000", "content": "Yes, I think now that the model is fitting the raw data to a certain extent, can we do some effective data augmentation to make the results better, and I have tried Focal loss in effb1, which improves the score by 0.1, but deepening the network does not improve the results
", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "1323412": "I trained some models with a result of 0.97 and tried to combine the model predictions and although the score did not improve, the ranking went from 132->77", "1323448": "Hey Can you please private the notebook. Releasing such a high scoring notebook is very problematic to the people who have a lower than rank than you. Many people can Just copy your code and can get in the medal range . You can use lower scoring models with the same code which does not have the same effect.Recently a person posted a public notebook which can get you into the medal range which got 35 copy and edit. This affects people who work honestly and their rank goes down because of public notebook. " } }