![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14984949%2Fda50e27e9c2aac44f6cba5afd73444e5%2FScreenshot%202025-07-08%20194214.png?generation=1751996542864798&alt=media\")
You might think it's Ariel2024, only a little bit harder. This is not the case.
\nIt'll probably be Ariel2024+Yale/UNC-CH on steroids.
\nSomething very important is common to Ariel2025 and Yale/UNC-CH: we have access to the code to generate data. It means all the tricks we have seen in Yale are in the game here. Although doing things with the test set (e.g., overfitting each sample or predict-simulate-train loop) will be harder since we can only do them during the prediction, as this is a code competition. But at the very least, we can generate more data and train on it- it was enough to get 3rd and 4th place in Yale.
\nI'm not sure it's smart for me to write this secret, but I'm sure that top competitors here are already aware of this fact, since they participated, like me, both in Ariel2024 and Yale, and they are too smart to miss something so obvious. So, competing for the price range will probably remain the same difficulty. So I'm like, you know, I'm just feeling like letting chaos rain a little bit in this competition, and let's see what will happen :) Since now I have made sure that everyone is aware of this fact, I hope to see interesting data generation and clever tricks notebooks. It's a great opportunity to earn some notebooks and datasets gold medals, go go go :)
Here is a link to an example from the data generation repo
", "messages": [ { "id": 3240134, "postDate": "2025-07-03T13:45:40.137Z", "content": "You might think it's Ariel2024, only a little bit harder. This is not the case.
\nIt'll probably be Ariel2024+Yale/UNC-CH on steroids.
\nSomething very important is common to Ariel2025 and Yale/UNC-CH: we have access to the code to generate data. It means all the tricks we have seen in Yale are in the game here. Although doing things with the test set (e.g., overfitting each sample or predict-simulate-train loop) will be harder since we can only do them during the prediction, as this is a code competition. But at the very least, we can generate more data and train on it- it was enough to get 3rd and 4th place in Yale.
\nI'm not sure it's smart for me to write this secret, but I'm sure that top competitors here are already aware of this fact, since they participated, like me, both in Ariel2024 and Yale, and they are too smart to miss something so obvious. So, competing for the price range will probably remain the same difficulty. So I'm like, you know, I'm just feeling like letting chaos rain a little bit in this competition, and let's see what will happen :) Since now I have made sure that everyone is aware of this fact, I hope to see interesting data generation and clever tricks notebooks. It's a great opportunity to earn some notebooks and datasets gold medals, go go go :)
Here is a link to an example from the data generation repo
", "rawMarkdown": "You might think it's Ariel2024, only a little bit harder. This is not the case.\nIt'll probably be Ariel2024+Yale/UNC-CH on steroids. \nSomething very important is common to Ariel2025 and Yale/UNC-CH: we have access to the code to generate data. It means all the tricks we have seen in Yale are in the game here. Although doing things with the test set (e.g., overfitting each sample or predict-simulate-train loop) will be harder since we can only do them during the prediction, as this is a code competition. But at the very least, we can generate more data and train on it- it was enough to get 3rd and 4th place in Yale. \nI'm not sure it's **smart** for me to write this secret, but I'm sure that top competitors here are already aware of this fact, since they participated, like me, both in Ariel2024 and Yale, and they are too smart to miss something so obvious. So, competing for the price range will probably remain the same difficulty. So I'm like, you know, I'm just feeling like letting chaos rain a little bit in this competition, and let's see what will happen :) Since now I have made sure that everyone is aware of this fact, I hope to see interesting data generation and clever tricks notebooks. It's a great opportunity to earn some notebooks and datasets gold medals, go go go :)\n\n[Here is a link to an example from the data generation repo](https://github.com/arielmission-space/ExoSim2-public/blob/main/examples/example_pipeline.ipynb)", "votes": 35 }, { "id": 3244945, "postDate": "2025-07-08T17:42:26.550Z", "content": "I must remind everyone of @shlomoron's wise words last year, which it seems everyone is going to have to relearn this year.
\n
Unless it's going to be signal processing again, and I just sent half the community on the wrong track! (Which is the correct answer? I don't know myself yet :) )
", "rawMarkdown": "Unless it's going to be signal processing again, and I just sent half the community on the wrong track! (Which is the correct answer? I don't know myself yet :) )", "votes": 2 } ] }, { "id": 3241496, "postDate": "2025-07-04T20:40:20.813Z", "content": "This topic this discussion I really amazed by this @shlomoron
", "rawMarkdown": "This topic this discussion I really amazed by this @shlomoron", "votes": 3 }, { "id": 3241494, "postDate": "2025-07-04T20:39:01.100Z", "content": "Good to know! Time to generate some data then. Game on! thanks for this topic discussion @shlomoron
", "rawMarkdown": "Good to know! Time to generate some data then. Game on! thanks for this topic discussion @shlomoron ", "votes": 3 }, { "id": 3240843, "postDate": "2025-07-04T09:19:19.470Z", "content": "I spent a lot of time on this last year. Still, there are a lot of things unknown when setting up the generation pipeline regarding the sensor setup and parameters. I had a very hard time getting close to the competition data, but indeed may be something for this year to try.
\nThank you for bringing it up here again.
This was actually also known last year - it's just that only one team actually realized itβ¦
", "rawMarkdown": "This was actually also known last year - it's just that only one team actually realized it...", "votes": 4, "replies": [ { "id": 3240728, "postDate": "2025-07-04T07:11:54.660Z", "content": "And they only used it for understanding how to eliminate some noise, yeah.
\nBut now we have the experience of Yale- we are not so naive anymore! This would probably no longer be enough to win! π
Looking forward to ML solutions this year!
\n(Well you can classify Gaussian Process Regression as ML, but let's forget about that for now)
Data crunching competition using many many GPU time again? π±π±π±
", "rawMarkdown": "Data crunching competition using many many GPU time again? π±π±π±", "votes": 1 }, { "id": 3240581, "postDate": "2025-07-04T02:37:40.880Z", "content": "2024's 5th place used Taurex3, but couldn't get clean matches, it might pay off this year
", "rawMarkdown": "2024's 5th place used Taurex3, but couldn't get clean matches, it might pay off this year", "votes": 1, "replies": [ { "id": 3240686, "postDate": "2025-07-04T05:47:36.740Z", "content": "We also tried taurex3 last year, but didn't work.
", "rawMarkdown": "We also tried taurex3 last year, but didn't work." }, { "id": 3240737, "postDate": "2025-07-04T07:21:31.677Z", "content": "Thats right, simulation for this competition was made afaik in ExoSim2 and not Taurex3, ao it makes sense you will have hard time get good data with Taurex3. But if someone manage to find good hyperparameters it'll be huge!
\n(but probably easier to just follow ExoSim2)
Last year, our team focused hard on Trauex3 and was able to boost cv, but it didn't work on LB because different types of molecules were used for LB.
\nI'm assuming it's the same for this year, and that using the spetrcum of each molecule won't help.
Unless you manage to probe LB for the new moleculesβ¦
", "rawMarkdown": "Unless you manage to probe LB for the new molecules..." }, { "id": 3241094, "postDate": "2025-07-04T14:33:51.373Z", "content": "ExoSim2 (if done right) can act as a bridge between Trauex3 and Ariel's deliberately unrealistic reality
", "rawMarkdown": "ExoSim2 (if done right) can act as a bridge between Trauex3 and Ariel's deliberately unrealistic reality" } ] }, { "id": 3240842, "postDate": "2025-07-04T09:17:09.633Z", "content": "yes, unknown molecules was the tricky part, though, as there are many degrees of freedom.
", "rawMarkdown": "yes, unknown molecules was the tricky part, though, as there are many degrees of freedom.", "votes": 1, "replies": [ { "id": 3241036, "postDate": "2025-07-04T13:07:40.980Z", "content": "Nice to see you joining this year too!
\nIt's going to be a tougher competition than last year.
100%.
", "rawMarkdown": "100%. ", "votes": 1 }, { "id": 3241097, "postDate": "2025-07-04T14:37:13.483Z", "content": "Newbie too!
", "rawMarkdown": "Newbie too!" } ] } ] }, { "id": 3240254, "postDate": "2025-07-03T16:16:15.600Z", "content": "Thanks for sharing @shlomoron, more data is the secret of Yale is what i missed in last 3 months. But sharing at initial stage of competition is really helpful.
", "rawMarkdown": "Thanks for sharing @shlomoron, more data is the secret of Yale is what i missed in last 3 months. But sharing at initial stage of competition is really helpful.", "votes": 1 }, { "id": 3241384, "postDate": "2025-07-04T18:46:46.893Z", "content": "Hi, I am a beginner at kaggle competitions and I often don't understand notebooks and discussions from kaggle geniuses. Can you tell me How did you and where did you learn all these things. PS. I am a self learner , so a bit of hints and roadmaps are appreciated! TIA
", "rawMarkdown": "Hi, I am a beginner at kaggle competitions and I often don't understand notebooks and discussions from kaggle geniuses. Can you tell me How did you and where did you learn all these things. PS. I am a self learner , so a bit of hints and roadmaps are appreciated! TIA", "replies": [ { "id": 3241401, "postDate": "2025-07-04T19:08:43.403Z", "content": "I tried to win Kaggle competitions π€·ββοΈ This is how you learn. There is no shortcut. You just have to seriously aim for the top and be good enough.
", "rawMarkdown": "I tried to win Kaggle competitions π€·ββοΈ This is how you learn. There is no shortcut. You just have to seriously aim for the top and be good enough.", "votes": 4, "replies": [ { "id": 3241426, "postDate": "2025-07-04T19:30:01.860Z", "content": "Ha ha, I like that energy \"This is how you learn.\", I am simply asking for a resources that might be help to be start good and keep going. Can you share any sources?
", "rawMarkdown": "Ha ha, I like that energy \"This is how you learn.\", I am simply asking for a resources that might be help to be start good and keep going. Can you share any sources?" }, { "id": 3241432, "postDate": "2025-07-04T19:34:32.210Z", "content": "The resources are Kaggle notebooks and past competitions top solutions.
", "rawMarkdown": "The resources are Kaggle notebooks and past competitions top solutions.", "votes": 2 } ] } ] }, { "id": 3243143, "postDate": "2025-07-06T19:15:41.913Z", "content": "Yeah its great info
", "rawMarkdown": "Yeah its great info " }, { "id": 3240632, "postDate": "2025-07-04T04:37:24.273Z", "content": "Nice sharing!
\nI see someone downvoting, maybe they're not happy with this :)
I probably just made this competition like two or three times more demanding for anything below price range, so I'm sure some people be angry! π€£
", "rawMarkdown": "I probably just made this competition like two or three times more demanding for anything below price range, so I'm sure some people be angry! π€£" } ] }, { "id": 3240202, "postDate": "2025-07-03T15:08:12.957Z", "content": "ItHard! True, my 90% of the submissions end with 0.0
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F23276411%2Fdc01cade481f30e810458ef1174a8cdd%2FScreenshot%20from%202025-07-03%2020-34-17.png?generation=1751555232465597&alt=media\")
Last year it took me a month to get a score above 0.000. Note that if your model goes haywire on just a single dataset it can already cause this. Having sanity checks and fallbacks in your code is critical.
", "rawMarkdown": "Last year it took me a month to get a score above 0.000. Note that if your model goes haywire on just a single dataset it can already cause this. Having sanity checks and fallbacks in your code is critical.", "votes": 1 } ] }, { "id": 3240179, "postDate": "2025-07-03T14:32:07.827Z", "content": "I missed Yale competition. May I ask for a link?
", "rawMarkdown": "I missed Yale competition. May I ask for a link?", "replies": [ { "id": 3240183, "postDate": "2025-07-03T14:35:37.310Z", "content": "https://www.kaggle.com/competitions/waveform-inversion
", "rawMarkdown": "https://www.kaggle.com/competitions/waveform-inversion", "votes": 1 } ] } ], "comments": [ { "id": 3244945, "author_name": "Jeroen Cottaar", "author_url": "", "post_date": "2025-07-08T17:42:26.550000", "content": "I must remind everyone of @shlomoron's wise words last year, which it seems everyone is going to have to relearn this year.
\n
Unless it's going to be signal processing again, and I just sent half the community on the wrong track! (Which is the correct answer? I don't know myself yet :) )
", "votes": 2, "replies": [] } ] }, { "id": 3241496, "author_name": "HijabZahra", "author_url": "", "post_date": "2025-07-04T20:40:20.813000", "content": "This topic this discussion I really amazed by this @shlomoron
", "votes": 3, "replies": [] }, { "id": 3241494, "author_name": "Sadia Shahid latif", "author_url": "", "post_date": "2025-07-04T20:39:01.100000", "content": "Good to know! Time to generate some data then. Game on! thanks for this topic discussion @shlomoron
", "votes": 3, "replies": [] }, { "id": 3240843, "author_name": "Pascal Pfeiffer", "author_url": "", "post_date": "2025-07-04T09:19:19.470000", "content": "I spent a lot of time on this last year. Still, there are a lot of things unknown when setting up the generation pipeline regarding the sensor setup and parameters. I had a very hard time getting close to the competition data, but indeed may be something for this year to try.
\nThank you for bringing it up here again.
This was actually also known last year - it's just that only one team actually realized itβ¦
", "votes": 4, "replies": [ { "id": 3240728, "author_name": "greySnow", "author_url": "", "post_date": "2025-07-04T07:11:54.660000", "content": "And they only used it for understanding how to eliminate some noise, yeah.
\nBut now we have the experience of Yale- we are not so naive anymore! This would probably no longer be enough to win! π
Looking forward to ML solutions this year!
\n(Well you can classify Gaussian Process Regression as ML, but let's forget about that for now)
Data crunching competition using many many GPU time again? π±π±π±
", "votes": 1, "replies": [] }, { "id": 3240581, "author_name": "g john rao", "author_url": "", "post_date": "2025-07-04T02:37:40.880000", "content": "2024's 5th place used Taurex3, but couldn't get clean matches, it might pay off this year
", "votes": 1, "replies": [ { "id": 3240686, "author_name": "c-number", "author_url": "", "post_date": "2025-07-04T05:47:36.740000", "content": "We also tried taurex3 last year, but didn't work.
", "votes": 0, "replies": [] }, { "id": 3240737, "author_name": "greySnow", "author_url": "", "post_date": "2025-07-04T07:21:31.677000", "content": "Thats right, simulation for this competition was made afaik in ExoSim2 and not Taurex3, ao it makes sense you will have hard time get good data with Taurex3. But if someone manage to find good hyperparameters it'll be huge!
\n(but probably easier to just follow ExoSim2)
Last year, our team focused hard on Trauex3 and was able to boost cv, but it didn't work on LB because different types of molecules were used for LB.
\nI'm assuming it's the same for this year, and that using the spetrcum of each molecule won't help.
Unless you manage to probe LB for the new moleculesβ¦
", "votes": 0, "replies": [] }, { "id": 3241094, "author_name": "g john rao", "author_url": "", "post_date": "2025-07-04T14:33:51.373000", "content": "ExoSim2 (if done right) can act as a bridge between Trauex3 and Ariel's deliberately unrealistic reality
", "votes": 0, "replies": [] } ] }, { "id": 3240842, "author_name": "Pascal Pfeiffer", "author_url": "", "post_date": "2025-07-04T09:17:09.633000", "content": "yes, unknown molecules was the tricky part, though, as there are many degrees of freedom.
", "votes": 1, "replies": [ { "id": 3241036, "author_name": "c-number", "author_url": "", "post_date": "2025-07-04T13:07:40.980000", "content": "Nice to see you joining this year too!
\nIt's going to be a tougher competition than last year.
100%.
", "votes": 1, "replies": [] }, { "id": 3241097, "author_name": "Sangram Patil", "author_url": "", "post_date": "2025-07-04T14:37:13.483000", "content": "Newbie too!
", "votes": 0, "replies": [] } ] } ] }, { "id": 3240254, "author_name": "SeshuRaju π§ββοΈ", "author_url": "", "post_date": "2025-07-03T16:16:15.600000", "content": "Thanks for sharing @shlomoron, more data is the secret of Yale is what i missed in last 3 months. But sharing at initial stage of competition is really helpful.
", "votes": 1, "replies": [] }, { "id": 3241384, "author_name": "Repello mugglentum!", "author_url": "", "post_date": "2025-07-04T18:46:46.893000", "content": "Hi, I am a beginner at kaggle competitions and I often don't understand notebooks and discussions from kaggle geniuses. Can you tell me How did you and where did you learn all these things. PS. I am a self learner , so a bit of hints and roadmaps are appreciated! TIA
", "votes": 0, "replies": [ { "id": 3241401, "author_name": "greySnow", "author_url": "", "post_date": "2025-07-04T19:08:43.403000", "content": "I tried to win Kaggle competitions π€·ββοΈ This is how you learn. There is no shortcut. You just have to seriously aim for the top and be good enough.
", "votes": 4, "replies": [ { "id": 3241426, "author_name": "Repello mugglentum!", "author_url": "", "post_date": "2025-07-04T19:30:01.860000", "content": "Ha ha, I like that energy \"This is how you learn.\", I am simply asking for a resources that might be help to be start good and keep going. Can you share any sources?
", "votes": 0, "replies": [] }, { "id": 3241432, "author_name": "greySnow", "author_url": "", "post_date": "2025-07-04T19:34:32.210000", "content": "The resources are Kaggle notebooks and past competitions top solutions.
", "votes": 2, "replies": [] } ] } ] }, { "id": 3243143, "author_name": "Zunairaa Manshad", "author_url": "", "post_date": "2025-07-06T19:15:41.913000", "content": "Yeah its great info
", "votes": 0, "replies": [] }, { "id": 3240632, "author_name": "c-number", "author_url": "", "post_date": "2025-07-04T04:37:24.273000", "content": "Nice sharing!
\nI see someone downvoting, maybe they're not happy with this :)
I probably just made this competition like two or three times more demanding for anything below price range, so I'm sure some people be angry! π€£
", "votes": 0, "replies": [] } ] }, { "id": 3240202, "author_name": "Sangram Patil", "author_url": "", "post_date": "2025-07-03T15:08:12.957000", "content": "ItHard! True, my 90% of the submissions end with 0.0
\n
Last year it took me a month to get a score above 0.000. Note that if your model goes haywire on just a single dataset it can already cause this. Having sanity checks and fallbacks in your code is critical.
", "votes": 1, "replies": [] } ] }, { "id": 3240179, "author_name": "DennisSakva", "author_url": "", "post_date": "2025-07-03T14:32:07.827000", "content": "I missed Yale competition. May I ask for a link?
", "votes": 0, "replies": [ { "id": 3240183, "author_name": "greySnow", "author_url": "", "post_date": "2025-07-03T14:35:37.310000", "content": "", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "3240134": "You might think it's Ariel2024, only a little bit harder. This is not the case.\nIt'll probably be Ariel2024+Yale/UNC-CH on steroids. \nSomething very important is common to Ariel2025 and Yale/UNC-CH: we have access to the code to generate data. It means all the tricks we have seen in Yale are in the game here. Although doing things with the test set (e.g., overfitting each sample or predict-simulate-train loop) will be harder since we can only do them during the prediction, as this is a code competition. But at the very least, we can generate more data and train on it- it was enough to get 3rd and 4th place in Yale. \nI'm not sure it's **smart** for me to write this secret, but I'm sure that top competitors here are already aware of this fact, since they participated, like me, both in Ariel2024 and Yale, and they are too smart to miss something so obvious. So, competing for the price range will probably remain the same difficulty. So I'm like, you know, I'm just feeling like letting chaos rain a little bit in this competition, and let's see what will happen :) Since now I have made sure that everyone is aware of this fact, I hope to see interesting data generation and clever tricks notebooks. It's a great opportunity to earn some notebooks and datasets gold medals, go go go :)\n\n[Here is a link to an example from the data generation repo](https://github.com/arielmission-space/ExoSim2-public/blob/main/examples/example_pipeline.ipynb)", "3244945": "I must remind everyone of @shlomoron's wise words last year, which it seems everyone is going to have to relearn this year.\n\n", "3241496": "This topic this discussion I really amazed by this @shlomoron", "3241494": "Good to know! Time to generate some data then. Game on! thanks for this topic discussion @shlomoron ", "3240843": "I spent a lot of time on this last year. Still, there are a lot of things unknown when setting up the generation pipeline regarding the sensor setup and parameters. I had a very hard time getting close to the competition data, but indeed may be something for this year to try.\nThank you for bringing it up here again.", "3240644": "This was actually also known last year - it's just that only one team actually realized it...", "3241573": "Data crunching competition using many many GPU time again? π±π±π±", "3240581": "2024's 5th place used Taurex3, but couldn't get clean matches, it might pay off this year", "3240254": "Thanks for sharing @shlomoron, more data is the secret of Yale is what i missed in last 3 months. But sharing at initial stage of competition is really helpful.", "3241384": "Hi, I am a beginner at kaggle competitions and I often don't understand notebooks and discussions from kaggle geniuses. Can you tell me How did you and where did you learn all these things. PS. I am a self learner , so a bit of hints and roadmaps are appreciated! TIA", "3243143": "Yeah its great info ", "3240632": "Nice sharing!\nI see someone downvoting, maybe they're not happy with this :)", "3240202": "```python\nIt'll probably be Ariel2024+Yale/UNC-CH on steroids. \n```\nHard! True, my 90% of the submissions end with 0.0\n\n", "3240179": "I missed Yale competition. May I ask for a link?" } }