![\"Imgur\"](\"https://i.imgur.com/P45YHNn.png\")
With 1.110 score our team currently holds the first place on public LB. I don't think it is possible with \"real\" models, not even close. In our case we developed a specific strategy for smart LB probing which is working.
\n\nI don't want to share any more details before the competition ends, in order not to disrupt the fair game during the last days, but I will describe our strategy afterwards, it is quite interesting! This strategy doesn't tell anything about the private test, it looks like it doesn't help us at all with the final standing. I estimate we will get somewhere between place 100 and 200 on the private, based on our last real models.
\n\nGood luck to all the participants, don't be misled by overfitters at the top!
\n\nUPDATED: Find the description here.
", "messages": [ { "id": "540888", "postDate": "06/01/2019 09:55:31", "content": "With 1.110 score our team currently holds the first place on public LB. I don't think it is possible with \"real\" models, not even close. In our case we developed a specific strategy for smart LB probing which is working.
\n\nI don't want to share any more details before the competition ends, in order not to disrupt the fair game during the last days, but I will describe our strategy afterwards, it is quite interesting! This strategy doesn't tell anything about the private test, it looks like it doesn't help us at all with the final standing. I estimate we will get somewhere between place 100 and 200 on the private, based on our last real models.
\n\nGood luck to all the participants, don't be misled by overfitters at the top!
\n\nUPDATED: Find the description here.
", "rawMarkdown": "With 1.110 score our team currently holds the first place on public LB. I don't think it is possible with \"real\" models, not even close. In our case we developed a specific strategy for smart LB probing which is working. \n\nI don't want to share any more details before the competition ends, in order not to disrupt the fair game during the last days, but I will describe our strategy afterwards, it is quite interesting! This strategy doesn't tell anything about the private test, it looks like it doesn't help us at all with the final standing. I estimate we will get somewhere between place 100 and 200 on the private, based on our last real models. \n\nGood luck to all the participants, don't be misled by overfitters at the top!\n\nUPDATED: Find the [description here](https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/94570).", "votes": null }, { "id": "540890", "postDate": "06/01/2019 10:01:58", "content": "Unbelievable! I just intend to post a new topic asking about your unstoppable progress on public LB. That's pretty amazing.
", "rawMarkdown": "Unbelievable! I just intend to post a new topic asking about your unstoppable progress on public LB. That's pretty amazing.", "votes": null }, { "id": "540892", "postDate": "06/01/2019 10:04:33", "content": "I want to venture a guess. Did you guys train a specific group of EQ or specific time frame? If that's the case, then that is why. If you trained a specific EQ or cross validated on specific EQ for submission, then my theory that the public test is just 1 EQ may be valid.
", "rawMarkdown": "I want to venture a guess. Did you guys train a specific group of EQ or specific time frame? If that's the case, then that is why. If you trained a specific EQ or cross validated on specific EQ for submission, then my theory that the public test is just 1 EQ may be valid.", "votes": null }, { "id": "540893", "postDate": "06/01/2019 10:04:40", "content": "Is it about some sort of post processing like changing mean/median of the predictions (Just blink once if you can't talk ;) )?
", "rawMarkdown": "Is it about some sort of post processing like changing mean/median of the predictions (Just blink once if you can't talk ;) )?", "votes": null }, { "id": "540894", "postDate": "06/01/2019 10:06:44", "content": "Thanks for sharing and confirming what I claimed; you're way better than us at overfitting public LB.
\n\nOn our side we have done no LB probing, but, as I wrote before, we have done something that we know will be detrimental to private LB, hence we are overfiting as well.
", "rawMarkdown": "Thanks for sharing and confirming what I claimed; you're way better than us at overfitting public LB.\n\nOn our side we have done no LB probing, but, as I wrote before, we have done something that we know will be detrimental to private LB, hence we are overfiting as well.", "votes": null }, { "id": "540895", "postDate": "06/01/2019 10:08:42", "content": "@zaharch It feels like a sudden jolt. Hope you'll be a strong survivor!
\n\nLooking forward to seeing your probing technique.
", "rawMarkdown": "zaharch It feels like a sudden jolt. Hope you'll be a strong survivor!\n\nLooking forward to seeing your probing technique.", "votes": null }, { "id": "540904", "postDate": "06/01/2019 10:25:12", "content": "with enough blinks I can transmit to you the whole submission file!
", "rawMarkdown": "with enough blinks I can transmit to you the whole submission file!", "votes": null }, { "id": "540930", "postDate": "06/01/2019 11:35:15", "content": "\n\n\nthen my theory that the public test is just 1 EQ may be valid.
\n
How can you have an EQ with max ttf between 9 and 10, and average 4.017 ?
", "rawMarkdown": "> then my theory that the public test is just 1 EQ may be valid. \n\nHow can you have an EQ with max ttf between 9 and 10, and average 4.017 ?", "votes": null }, { "id": "540950", "postDate": "06/01/2019 12:20:44", "content": "At least you were lb probing deliberately - I think that most of us have been doing it unconsciously - I have a strong feeling that my best score will come from my first 20 submissions.
", "rawMarkdown": "At least you were lb probing deliberately - I think that most of us have been doing it unconsciously - I have a strong feeling that my best score will come from my first 20 submissions.", "votes": null }, { "id": "540955", "postDate": "06/01/2019 12:28:26", "content": "
I give up. your right.
", "rawMarkdown": "\n\nI give up. your right.", "votes": null }, { "id": "540967", "postDate": "06/01/2019 12:46:08", "content": "My calculus is not what it used to be, but 2.2s ttf, then Eq, then ttf=10, then another 1.2s? So, a total span of 3.4s test sample with one Eq?
", "rawMarkdown": "My calculus is not what it used to be, but 2.2s ttf, then Eq, then ttf=10, then another 1.2s? So, a total span of 3.4s test sample with one Eq?", "votes": null }, { "id": "541002", "postDate": "06/01/2019 14:24:13", "content": "Thank you for the post. Our team are wondering what happened on public LB! I really look forward to your solution after the competition ends.
", "rawMarkdown": "Thank you for the post. Our team are wondering what happened on public LB! I really look forward to your solution after the competition ends.", "votes": null }, { "id": "541014", "postDate": "06/01/2019 14:54:34", "content": "why do you think you will only be between place 100 and 200?
", "rawMarkdown": "why do you think you will only be between place 100 and 200?", "votes": null }, { "id": "541019", "postDate": "06/01/2019 15:02:02", "content": "To make us think they just overfit when they actually have great models ;)
", "rawMarkdown": "To make us think they just overfit when they actually have great models ;)", "votes": null }, { "id": "541028", "postDate": "06/01/2019 15:30:11", "content": "Well, we have a regular model that scores in range of places 100-200, and our two submissions will be based on it... or what @cpmpml said :). Don't know why we would do it though.
", "rawMarkdown": "Well, we have a regular model that scores in range of places 100-200, and our two submissions will be based on it... or what @cpmpml said :). Don't know why we would do it though.", "votes": null }, { "id": "541038", "postDate": "06/01/2019 15:42:21", "content": "\n\n\nDon't know why we would do it though.
\n
Just teasing you ;)
", "rawMarkdown": "> Don't know why we would do it though.\n\nJust teasing you ;)", "votes": null }, { "id": "541039", "postDate": "06/01/2019 15:44:39", "content": "![\"\"](\"https://i.imgur.com/daoOHWW.jpg\")
Careful now...
", "rawMarkdown": "\n\nCareful now...", "votes": null }, { "id": "541040", "postDate": "06/01/2019 15:47:48", "content": "My Lord. You’re a king that awaits to be dethroned. Your majestic kingdom will last for only 2 more days. Your heritage will be devoured. Your mystique will be nakedly revealed. But, you’ll be always remembered...
", "rawMarkdown": "My Lord. You’re a king that awaits to be dethroned. Your majestic kingdom will last for only 2 more days. Your heritage will be devoured. Your mystique will be nakedly revealed. But, you’ll be always remembered...", "votes": null }, { "id": "541051", "postDate": "06/01/2019 16:07:41", "content": "Looks more like Public LB / Private LB
", "rawMarkdown": "Looks more like Public LB / Private LB", "votes": null }, { "id": "541120", "postDate": "06/01/2019 19:19:50", "content": "Isn't it pity to sacrifice submissions for probing instead of improving models during the final stage of the competition, or your view on that is different?
", "rawMarkdown": "Isn't it pity to sacrifice submissions for probing instead of improving models during the final stage of the competition, or your view on that is different?", "votes": null }, { "id": "541177", "postDate": "06/02/2019 01:03:52", "content": "The winning strategy of \"Don't Overfit II\" was well-designed LB probing: https://www.kaggle.com/cdeotte/lb-probing-strategies-0-890-2nd-place\nI believe it requires in-depth understanding to probe LB well. Good job regardless of the final standing!
", "rawMarkdown": "The winning strategy of \"Don't Overfit II\" was well-designed LB probing: https://www.kaggle.com/cdeotte/lb-probing-strategies-0-890-2nd-place\nI believe it requires in-depth understanding to probe LB well. Good job regardless of the final standing!", "votes": null }, { "id": "541188", "postDate": "06/02/2019 01:44:08", "content": "But my instinct tells me you'll still end up top 100 :)
", "rawMarkdown": "But my instinct tells me you'll still end up top 100 :)", "votes": null }, { "id": "541392", "postDate": "06/02/2019 11:15:17", "content": "This is helping their final submission.
", "rawMarkdown": "This is helping their final submission.", "votes": null }, { "id": "541583", "postDate": "06/02/2019 17:32:54", "content": "Wow! you played well with the words @khahuras
", "rawMarkdown": "Wow! you played well with the words @khahuras", "votes": null }, { "id": "541648", "postDate": "06/02/2019 20:19:30", "content": "Well said! Could you imagine that out of your 225 total posts this one will give you your third discussion golden medal? Kaggle likes dynamics deserves its own kaggle competition.
", "rawMarkdown": "Well said! Could you imagine that out of your 225 total posts this one will give you your third discussion golden medal? Kaggle likes dynamics deserves its own kaggle competition.", "votes": null }, { "id": "541721", "postDate": "06/03/2019 00:32:13", "content": "I think 90% of the teams are overfitting.
", "rawMarkdown": "I think 90% of the teams are overfitting.", "votes": null }, { "id": "542125", "postDate": "06/03/2019 13:39:32", "content": "Just wait for some hours to get LB drastically changing..... 😊
", "rawMarkdown": "Just wait for some hours to get LB drastically changing..... 😊", "votes": null }, { "id": "542322", "postDate": "06/03/2019 17:48:27", "content": "I'm super new and my strategy currently sucks.... LOL I'm nowhere near the top so I'll just say what I did I took the acoustic data and turned it into images and used a resnet classification it did not work so well hahaha Though I'm sure someone could probably take my idea and make it better because I'm only 2 weeks old in the machine learning world.
", "rawMarkdown": "I'm super new and my strategy currently sucks.... LOL I'm nowhere near the top so I'll just say what I did I took the acoustic data and turned it into images and used a resnet classification it did not work so well hahaha Though I'm sure someone could probably take my idea and make it better because I'm only 2 weeks old in the machine learning world.", "votes": null }, { "id": "543045", "postDate": "06/04/2019 09:47:36", "content": "Congrats on you medal! :)
", "rawMarkdown": "Congrats on you medal! :)", "votes": null }, { "id": "543080", "postDate": "06/04/2019 10:15:43", "content": "So, how did you do it? And congrats, you are the only 3rd drop from gold zone, after us and @khahuras unless mistaken. You ended better than you predicted!
", "rawMarkdown": "So, how did you do it? And congrats, you are the only 3rd drop from gold zone, after us and @khahuras unless mistaken. You ended better than you predicted!", "votes": null }, { "id": "543108", "postDate": "06/04/2019 10:32:39", "content": "Thanks, I will share it later today or tomorrow! So much info to process :)
", "rawMarkdown": "Thanks, I will share it later today or tomorrow! So much info to process :)", "votes": null }, { "id": "543209", "postDate": "06/04/2019 11:29:59", "content": "\n\n\nThanks, I will share it later today or tomorrow! So much info to process :)
\n
We'll wait for it! :)\nBy the way was it helpful in your models?
", "rawMarkdown": "> Thanks, I will share it later today or tomorrow! So much info to process :)\n\nWe'll wait for it! :)\nBy the way was it helpful in your models?", "votes": null }, { "id": "544219", "postDate": "06/05/2019 09:26:34", "content": "It took me some time to write it down, but the description of our strategy is ready now, please follow to this post. The link to the article on medium is embedded there.
", "rawMarkdown": "It took me some time to write it down, but the description of our strategy is ready now, please follow [to this post](https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/94570). The link to the article on medium is embedded there.", "votes": null } ], "comments": [ { "id": 540890, "author_name": "khahuras", "author_url": "", "post_date": "06/01/2019 10:01:58", "content": "Unbelievable! I just intend to post a new topic asking about your unstoppable progress on public LB. That's pretty amazing.
", "votes": null, "replies": [] }, { "id": 540892, "author_name": "teeyee314", "author_url": "", "post_date": "06/01/2019 10:04:33", "content": "I want to venture a guess. Did you guys train a specific group of EQ or specific time frame? If that's the case, then that is why. If you trained a specific EQ or cross validated on specific EQ for submission, then my theory that the public test is just 1 EQ may be valid.
", "votes": null, "replies": [ { "id": 540930, "author_name": "cpmpml", "author_url": "", "post_date": "06/01/2019 11:35:15", "content": "\n\n\nthen my theory that the public test is just 1 EQ may be valid.
\n
How can you have an EQ with max ttf between 9 and 10, and average 4.017 ?
", "votes": null, "replies": [] }, { "id": 540955, "author_name": "teeyee314", "author_url": "", "post_date": "06/01/2019 12:28:26", "content": "
I give up. your right.
", "votes": null, "replies": [] }, { "id": 540967, "author_name": "glimmung", "author_url": "", "post_date": "06/01/2019 12:46:08", "content": "My calculus is not what it used to be, but 2.2s ttf, then Eq, then ttf=10, then another 1.2s? So, a total span of 3.4s test sample with one Eq?
", "votes": null, "replies": [] } ] }, { "id": 540893, "author_name": "kainsama", "author_url": "", "post_date": "06/01/2019 10:04:40", "content": "Is it about some sort of post processing like changing mean/median of the predictions (Just blink once if you can't talk ;) )?
", "votes": null, "replies": [ { "id": 540904, "author_name": "zaharch", "author_url": "", "post_date": "06/01/2019 10:25:12", "content": "with enough blinks I can transmit to you the whole submission file!
", "votes": null, "replies": [] } ] }, { "id": 540894, "author_name": "cpmpml", "author_url": "", "post_date": "06/01/2019 10:06:44", "content": "Thanks for sharing and confirming what I claimed; you're way better than us at overfitting public LB.
\n\nOn our side we have done no LB probing, but, as I wrote before, we have done something that we know will be detrimental to private LB, hence we are overfiting as well.
", "votes": null, "replies": [] }, { "id": 540895, "author_name": "amitkumarjaiswal", "author_url": "", "post_date": "06/01/2019 10:08:42", "content": "@zaharch It feels like a sudden jolt. Hope you'll be a strong survivor!
\n\nLooking forward to seeing your probing technique.
", "votes": null, "replies": [] }, { "id": 540950, "author_name": "scirpus", "author_url": "", "post_date": "06/01/2019 12:20:44", "content": "At least you were lb probing deliberately - I think that most of us have been doing it unconsciously - I have a strong feeling that my best score will come from my first 20 submissions.
", "votes": null, "replies": [] }, { "id": 541002, "author_name": "sishihara", "author_url": "", "post_date": "06/01/2019 14:24:13", "content": "Thank you for the post. Our team are wondering what happened on public LB! I really look forward to your solution after the competition ends.
", "votes": null, "replies": [] }, { "id": 541014, "author_name": "takeiy", "author_url": "", "post_date": "06/01/2019 14:54:34", "content": "why do you think you will only be between place 100 and 200?
", "votes": null, "replies": [ { "id": 541019, "author_name": "cpmpml", "author_url": "", "post_date": "06/01/2019 15:02:02", "content": "To make us think they just overfit when they actually have great models ;)
", "votes": null, "replies": [] }, { "id": 541028, "author_name": "zaharch", "author_url": "", "post_date": "06/01/2019 15:30:11", "content": "Well, we have a regular model that scores in range of places 100-200, and our two submissions will be based on it... or what @cpmpml said :). Don't know why we would do it though.
", "votes": null, "replies": [] }, { "id": 541038, "author_name": "cpmpml", "author_url": "", "post_date": "06/01/2019 15:42:21", "content": "\n\n\nDon't know why we would do it though.
\n
Just teasing you ;)
", "votes": null, "replies": [] } ] }, { "id": 541039, "author_name": "bigironsphere", "author_url": "", "post_date": "06/01/2019 15:44:39", "content": "
Careful now...
", "votes": null, "replies": [ { "id": 541051, "author_name": "ivanpan", "author_url": "", "post_date": "06/01/2019 16:07:41", "content": "Looks more like Public LB / Private LB
", "votes": null, "replies": [] } ] }, { "id": 541040, "author_name": "khahuras", "author_url": "", "post_date": "06/01/2019 15:47:48", "content": "My Lord. You’re a king that awaits to be dethroned. Your majestic kingdom will last for only 2 more days. Your heritage will be devoured. Your mystique will be nakedly revealed. But, you’ll be always remembered...
", "votes": null, "replies": [ { "id": 541583, "author_name": "karanjakhar", "author_url": "", "post_date": "06/02/2019 17:32:54", "content": "Wow! you played well with the words @khahuras
", "votes": null, "replies": [] }, { "id": 541648, "author_name": "zaharch", "author_url": "", "post_date": "06/02/2019 20:19:30", "content": "Well said! Could you imagine that out of your 225 total posts this one will give you your third discussion golden medal? Kaggle likes dynamics deserves its own kaggle competition.
", "votes": null, "replies": [] } ] }, { "id": 541120, "author_name": "davids1992", "author_url": "", "post_date": "06/01/2019 19:19:50", "content": "Isn't it pity to sacrifice submissions for probing instead of improving models during the final stage of the competition, or your view on that is different?
", "votes": null, "replies": [ { "id": 541392, "author_name": "cpmpml", "author_url": "", "post_date": "06/02/2019 11:15:17", "content": "This is helping their final submission.
", "votes": null, "replies": [] } ] }, { "id": 541177, "author_name": "vivaroma", "author_url": "", "post_date": "06/02/2019 01:03:52", "content": "The winning strategy of \"Don't Overfit II\" was well-designed LB probing: https://www.kaggle.com/cdeotte/lb-probing-strategies-0-890-2nd-place\nI believe it requires in-depth understanding to probe LB well. Good job regardless of the final standing!
", "votes": null, "replies": [] }, { "id": 541188, "author_name": "vivaroma", "author_url": "", "post_date": "06/02/2019 01:44:08", "content": "But my instinct tells me you'll still end up top 100 :)
", "votes": null, "replies": [] }, { "id": 541721, "author_name": "felipefonte99", "author_url": "", "post_date": "06/03/2019 00:32:13", "content": "I think 90% of the teams are overfitting.
", "votes": null, "replies": [] }, { "id": 542125, "author_name": "himusoni", "author_url": "", "post_date": "06/03/2019 13:39:32", "content": "Just wait for some hours to get LB drastically changing..... 😊
", "votes": null, "replies": [] }, { "id": 542322, "author_name": "jidoja", "author_url": "", "post_date": "06/03/2019 17:48:27", "content": "I'm super new and my strategy currently sucks.... LOL I'm nowhere near the top so I'll just say what I did I took the acoustic data and turned it into images and used a resnet classification it did not work so well hahaha Though I'm sure someone could probably take my idea and make it better because I'm only 2 weeks old in the machine learning world.
", "votes": null, "replies": [] }, { "id": 543045, "author_name": "prashanththangavel", "author_url": "", "post_date": "06/04/2019 09:47:36", "content": "Congrats on you medal! :)
", "votes": null, "replies": [] }, { "id": 543080, "author_name": "cpmpml", "author_url": "", "post_date": "06/04/2019 10:15:43", "content": "So, how did you do it? And congrats, you are the only 3rd drop from gold zone, after us and @khahuras unless mistaken. You ended better than you predicted!
", "votes": null, "replies": [ { "id": 543108, "author_name": "zaharch", "author_url": "", "post_date": "06/04/2019 10:32:39", "content": "Thanks, I will share it later today or tomorrow! So much info to process :)
", "votes": null, "replies": [] }, { "id": 543209, "author_name": "sergeyzlobin", "author_url": "", "post_date": "06/04/2019 11:29:59", "content": "\n\n\nThanks, I will share it later today or tomorrow! So much info to process :)
\n
We'll wait for it! :)\nBy the way was it helpful in your models?
", "votes": null, "replies": [] } ] }, { "id": 544219, "author_name": "zaharch", "author_url": "", "post_date": "06/05/2019 09:26:34", "content": "It took me some time to write it down, but the description of our strategy is ready now, please follow to this post. The link to the article on medium is embedded there.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "540888": "With 1.110 score our team currently holds the first place on public LB. I don't think it is possible with \"real\" models, not even close. In our case we developed a specific strategy for smart LB probing which is working. \n\nI don't want to share any more details before the competition ends, in order not to disrupt the fair game during the last days, but I will describe our strategy afterwards, it is quite interesting! This strategy doesn't tell anything about the private test, it looks like it doesn't help us at all with the final standing. I estimate we will get somewhere between place 100 and 200 on the private, based on our last real models. \n\nGood luck to all the participants, don't be misled by overfitters at the top!\n\nUPDATED: Find the [description here](https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/94570).", "540890": "Unbelievable! I just intend to post a new topic asking about your unstoppable progress on public LB. That's pretty amazing.", "540892": "I want to venture a guess. Did you guys train a specific group of EQ or specific time frame? If that's the case, then that is why. If you trained a specific EQ or cross validated on specific EQ for submission, then my theory that the public test is just 1 EQ may be valid.", "540893": "Is it about some sort of post processing like changing mean/median of the predictions (Just blink once if you can't talk ;) )?", "540894": "Thanks for sharing and confirming what I claimed; you're way better than us at overfitting public LB.\n\nOn our side we have done no LB probing, but, as I wrote before, we have done something that we know will be detrimental to private LB, hence we are overfiting as well.", "540895": "zaharch It feels like a sudden jolt. Hope you'll be a strong survivor!\n\nLooking forward to seeing your probing technique.", "540904": "with enough blinks I can transmit to you the whole submission file!", "540930": "> then my theory that the public test is just 1 EQ may be valid. \n\nHow can you have an EQ with max ttf between 9 and 10, and average 4.017 ?", "540950": "At least you were lb probing deliberately - I think that most of us have been doing it unconsciously - I have a strong feeling that my best score will come from my first 20 submissions.", "540955": "\n\nI give up. your right.", "540967": "My calculus is not what it used to be, but 2.2s ttf, then Eq, then ttf=10, then another 1.2s? So, a total span of 3.4s test sample with one Eq?", "541002": "Thank you for the post. Our team are wondering what happened on public LB! I really look forward to your solution after the competition ends.", "541014": "why do you think you will only be between place 100 and 200?", "541019": "To make us think they just overfit when they actually have great models ;)", "541028": "Well, we have a regular model that scores in range of places 100-200, and our two submissions will be based on it... or what @cpmpml said :). Don't know why we would do it though.", "541038": "> Don't know why we would do it though.\n\nJust teasing you ;)", "541039": "\n\nCareful now...", "541040": "My Lord. You’re a king that awaits to be dethroned. Your majestic kingdom will last for only 2 more days. Your heritage will be devoured. Your mystique will be nakedly revealed. But, you’ll be always remembered...", "541051": "Looks more like Public LB / Private LB", "541120": "Isn't it pity to sacrifice submissions for probing instead of improving models during the final stage of the competition, or your view on that is different?", "541177": "The winning strategy of \"Don't Overfit II\" was well-designed LB probing: https://www.kaggle.com/cdeotte/lb-probing-strategies-0-890-2nd-place\nI believe it requires in-depth understanding to probe LB well. Good job regardless of the final standing!", "541188": "But my instinct tells me you'll still end up top 100 :)", "541392": "This is helping their final submission.", "541583": "Wow! you played well with the words @khahuras", "541648": "Well said! Could you imagine that out of your 225 total posts this one will give you your third discussion golden medal? Kaggle likes dynamics deserves its own kaggle competition.", "541721": "I think 90% of the teams are overfitting.", "542125": "Just wait for some hours to get LB drastically changing..... 😊", "542322": "I'm super new and my strategy currently sucks.... LOL I'm nowhere near the top so I'll just say what I did I took the acoustic data and turned it into images and used a resnet classification it did not work so well hahaha Though I'm sure someone could probably take my idea and make it better because I'm only 2 weeks old in the machine learning world.", "543045": "Congrats on you medal! :)", "543080": "So, how did you do it? And congrats, you are the only 3rd drop from gold zone, after us and @khahuras unless mistaken. You ended better than you predicted!", "543108": "Thanks, I will share it later today or tomorrow! So much info to process :)", "543209": "> Thanks, I will share it later today or tomorrow! So much info to process :)\n\nWe'll wait for it! :)\nBy the way was it helpful in your models?", "544219": "It took me some time to write it down, but the description of our strategy is ready now, please follow [to this post](https://www.kaggle.com/c/LANL-Earthquake-Prediction/discussion/94570). The link to the article on medium is embedded there." }, "source": "meta" }