The current public LB is meaningless and filled with memorizer GANs and imitation game kernels. Both of which are now disallowed. \nSo how do I track my progress? \nWhat are the current score cutoffs for gold, silver and bronze medals?\nI would guess - 25, 45 and 65.
", "messages": [ { "id": 591709, "postDate": "2019-08-04T06:23:32.023Z", "content": "The current public LB is meaningless and filled with memorizer GANs and imitation game kernels. Both of which are now disallowed. \nSo how do I track my progress? \nWhat are the current score cutoffs for gold, silver and bronze medals?\nI would guess - 25, 45 and 65.
", "rawMarkdown": "The current public LB is meaningless and filled with memorizer GANs and imitation game kernels. Both of which are now disallowed. \nSo how do I track my progress? \nWhat are the current score cutoffs for gold, silver and bronze medals?\nI would guess - 25, 45 and 65. ", "votes": 8 }, { "id": 591740, "postDate": "2019-08-04T07:20:46.103Z", "content": "I like your guess, although I expect bronze to be slightly lower. On this matter, I'll add two things:\n- It is likely that final LB scores are higher than public LB ones (for legit GANs), because the metric on public is evaluated on the same data models are trained on.\n- There are a lot of people with great models hiding their progress (see the Gendered Pronoun Resolution comp), I can name 3 people that are below 25 at the moment but I'm pretty sure there are more.
", "rawMarkdown": "I like your guess, although I expect bronze to be slightly lower. On this matter, I'll add two things:\n- It is likely that final LB scores are higher than public LB ones (for legit GANs), because the metric on public is evaluated on the same data models are trained on.\n- There are a lot of people with great models hiding their progress (see the Gendered Pronoun Resolution comp), I can name 3 people that are below 25 at the moment but I'm pretty sure there are more.", "votes": 4, "replies": [ { "id": 591757, "postDate": "2019-08-04T08:01:03.487Z", "content": "I guess you're right\nI just saw bestfitting with a score of 17.3 on the LB\nI need to work very hard
", "rawMarkdown": "I guess you're right\nI just saw bestfitting with a score of 17.3 on the LB\nI need to work very hard", "votes": 3 } ] }, { "id": 591710, "postDate": "2019-08-04T06:26:42.307Z", "content": "No way ... 65 is too normal score . I would think 45 is the cutoff for bronze
", "rawMarkdown": "No way ... 65 is too normal score . I would think 45 is the cutoff for bronze", "votes": 1, "replies": [ { "id": 591718, "postDate": "2019-08-04T06:31:46.923Z", "content": "That might be true in a few days but for now, a 45 scoring kernel stands at ~400.\nI think we can safely assume top ~300 are memorizer GANs. \nI am sure many kernels around rank 300(16)-400(45) are imitation games (a lot of 1-2 submissions made many days ago) so if not 65, 55-60 is definitely the bronze cutoff right now.
", "rawMarkdown": "That might be true in a few days but for now, a 45 scoring kernel stands at ~400.\nI think we can safely assume top ~300 are memorizer GANs. \nI am sure many kernels around rank 300(16)-400(45) are imitation games (a lot of 1-2 submissions made many days ago) so if not 65, 55-60 is definitely the bronze cutoff right now.\n ", "votes": 1 } ] }, { "id": 597899, "postDate": "2019-08-12T21:53:10.657Z", "content": "We can expect this:
\n40 winners + golds \n60 silvers cutoff\n70 bronzes cutoff
Just let's wait 2 days and it's done.
\n\nIt would be really good to have some button to re-submit some kernels as sort of -pre-final removing previous ones with memorization just to see approximately who is who and where we are.
\n\nGood luck to everyone in this mess of positions.
", "rawMarkdown": "We can expect this: \n40 winners + golds \n60 silvers cutoff\n70 bronzes cutoff\n\nJust let's wait 2 days and it's done. \n\nIt would be really good to have some button to re-submit some kernels as sort of -pre-final removing previous ones with memorization just to see approximately who is who and where we are. \n\nGood luck to everyone in this mess of positions. ", "votes": 2, "replies": [ { "id": 598012, "postDate": "2019-08-13T02:28:25.857Z", "content": "There is a public kernel that scores LB 60. So you'll need better than that to get a bronze or silver
", "rawMarkdown": "There is a public kernel that scores LB 60. So you'll need better than that to get a bronze or silver", "votes": 4 } ] }, { "id": 591906, "postDate": "2019-08-04T13:02:50.580Z", "content": "bronze I guess about 55 because public best is 60
", "rawMarkdown": "bronze I guess about 55 because public best is 60", "replies": [ { "id": 591925, "postDate": "2019-08-04T13:39:41.840Z", "content": "LoL ... Did I make a mistake then ..
", "rawMarkdown": "LoL ... Did I make a mistake then ..", "votes": 3 }, { "id": 598015, "postDate": "2019-08-13T02:32:07.293Z", "content": "You made it harder for us :) but competition is fun. My best was only 64 when Kaggle changed the rule disqualifying Memorizer GANs. So I need to improve it to beat your public kernel. I've gotten my GAN to mid 50s. Hoping to improve it more in the next few days.
", "rawMarkdown": "You made it harder for us :) but competition is fun. My best was only 64 when Kaggle changed the rule disqualifying Memorizer GANs. So I need to improve it to beat your public kernel. I've gotten my GAN to mid 50s. Hoping to improve it more in the next few days." } ] }, { "id": 597897, "postDate": "2019-08-12T21:48:58.467Z", "rawMarkdown": "", "isDeleted": true }, { "id": 597896, "postDate": "2019-08-12T21:46:24.643Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 591740, "author_name": "Theo Viel", "author_url": "", "post_date": "2019-08-04T07:20:46.103000", "content": "I like your guess, although I expect bronze to be slightly lower. On this matter, I'll add two things:\n- It is likely that final LB scores are higher than public LB ones (for legit GANs), because the metric on public is evaluated on the same data models are trained on.\n- There are a lot of people with great models hiding their progress (see the Gendered Pronoun Resolution comp), I can name 3 people that are below 25 at the moment but I'm pretty sure there are more.
", "votes": 4, "replies": [ { "id": 591757, "author_name": "timetraveller", "author_url": "", "post_date": "2019-08-04T08:01:03.487000", "content": "I guess you're right\nI just saw bestfitting with a score of 17.3 on the LB\nI need to work very hard
", "votes": 3, "replies": [] } ] }, { "id": 591710, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2019-08-04T06:26:42.307000", "content": "No way ... 65 is too normal score . I would think 45 is the cutoff for bronze
", "votes": 1, "replies": [ { "id": 591718, "author_name": "timetraveller", "author_url": "", "post_date": "2019-08-04T06:31:46.923000", "content": "That might be true in a few days but for now, a 45 scoring kernel stands at ~400.\nI think we can safely assume top ~300 are memorizer GANs. \nI am sure many kernels around rank 300(16)-400(45) are imitation games (a lot of 1-2 submissions made many days ago) so if not 65, 55-60 is definitely the bronze cutoff right now.
", "votes": 1, "replies": [] } ] }, { "id": 597899, "author_name": "Mukharbek Organokov", "author_url": "", "post_date": "2019-08-12T21:53:10.657000", "content": "We can expect this:
\n40 winners + golds \n60 silvers cutoff\n70 bronzes cutoff
Just let's wait 2 days and it's done.
\n\nIt would be really good to have some button to re-submit some kernels as sort of -pre-final removing previous ones with memorization just to see approximately who is who and where we are.
\n\nGood luck to everyone in this mess of positions.
", "votes": 2, "replies": [ { "id": 598012, "author_name": "Chris Deotte", "author_url": "", "post_date": "2019-08-13T02:28:25.857000", "content": "There is a public kernel that scores LB 60. So you'll need better than that to get a bronze or silver
", "votes": 4, "replies": [] } ] }, { "id": 591906, "author_name": "saewonYang", "author_url": "", "post_date": "2019-08-04T13:02:50.580000", "content": "bronze I guess about 55 because public best is 60
", "votes": 0, "replies": [ { "id": 591925, "author_name": "Nirjhar Roy", "author_url": "", "post_date": "2019-08-04T13:39:41.840000", "content": "LoL ... Did I make a mistake then ..
", "votes": 3, "replies": [] }, { "id": 598015, "author_name": "Chris Deotte", "author_url": "", "post_date": "2019-08-13T02:32:07.293000", "content": "You made it harder for us :) but competition is fun. My best was only 64 when Kaggle changed the rule disqualifying Memorizer GANs. So I need to improve it to beat your public kernel. I've gotten my GAN to mid 50s. Hoping to improve it more in the next few days.
", "votes": 0, "replies": [] } ] }, { "id": 597897, "author_name": "", "author_url": "", "post_date": "2019-08-12T21:48:58.467000", "content": "", "votes": 0, "replies": [] }, { "id": 597896, "author_name": "", "author_url": "", "post_date": "2019-08-12T21:46:24.643000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "591709": "The current public LB is meaningless and filled with memorizer GANs and imitation game kernels. Both of which are now disallowed. \nSo how do I track my progress? \nWhat are the current score cutoffs for gold, silver and bronze medals?\nI would guess - 25, 45 and 65. ", "591740": "I like your guess, although I expect bronze to be slightly lower. On this matter, I'll add two things:\n- It is likely that final LB scores are higher than public LB ones (for legit GANs), because the metric on public is evaluated on the same data models are trained on.\n- There are a lot of people with great models hiding their progress (see the Gendered Pronoun Resolution comp), I can name 3 people that are below 25 at the moment but I'm pretty sure there are more.", "591710": "No way ... 65 is too normal score . I would think 45 is the cutoff for bronze", "597899": "We can expect this: \n40 winners + golds \n60 silvers cutoff\n70 bronzes cutoff\n\nJust let's wait 2 days and it's done. \n\nIt would be really good to have some button to re-submit some kernels as sort of -pre-final removing previous ones with memorization just to see approximately who is who and where we are. \n\nGood luck to everyone in this mess of positions. ", "591906": "bronze I guess about 55 because public best is 60", "597897": "", "597896": "" } }