Hi Everyone,
\nThis was the most legendary panel I had the privilege of hosting till date π I had hosted some of the best Kagglers on \"Best Practises for Training ML Models\", we discussed:
\nThanks to the absolute legends from H2O for doing the panel: Dott, Beluga, ilu, Psi, ybabkhin (b.e.s) π
\n\nAudio version (will be published on Sunday
", "messages": [ { "id": 2109024, "postDate": "2023-01-21T04:39:10.663Z", "content": "Hi Everyone,
\nThis was the most legendary panel I had the privilege of hosting till date π I had hosted some of the best Kagglers on \"Best Practises for Training ML Models\", we discussed:
\nThanks to the absolute legends from H2O for doing the panel: Dott, Beluga, ilu, Psi, ybabkhin (b.e.s) π
\n\nAudio version (will be published on Sunday
", "rawMarkdown": "Hi Everyone, \n\nThis was the most legendary panel I had the privilege of hosting till date π I had hosted some of the best Kagglers on \"Best Practises for Training ML Models\", we discussed:\n\n- Early Stopping\n- LR Schedulers, Warmup, Optimisers\n- Differential LR\n- Data Augmentation\n- CNNs Vs Transformers\n- Best Practises for NLP\n- Stacking \n\nThanks to the absolute legends from H2O for doing the panel: [Dott](https://www.kaggle.com/dott1718), [Beluga](https://www.kaggle.com/gaborfodor), [ilu](https://www.kaggle.com/ilu000), [Psi](https://www.kaggle.com/philippsinger), [ybabkhin (b.e.s)](https://www.kaggle.com/ybabakhin) π\n\n[Link to the panel video](https://www.youtube.com/watch?v=_mzrfMA8Qx4)\n\n[Audio version (will be published on Sunday](https://anchor.fm/chaitimedatascience)", "votes": 28 }, { "id": 2109231, "postDate": "2023-01-21T08:11:36.293Z", "content": "Very nice panel video. I watched twice. Lots of interesting insights and experiences. I recommend.
\nUnfortunately no (direct) answer how to improve score in rsna π
Thank you Remek, I'm happy you enjoyed the discussion! π
\nYeah, the audience tried really hard to extract Team Hydrogen's insights but they didn't spill any chai π
", "rawMarkdown": "Thank you Remek, I'm happy you enjoyed the discussion! π\n\nYeah, the audience tried really hard to extract Team Hydrogen's insights but they didn't spill any chai π", "votes": 3, "replies": [ { "id": 2109327, "postDate": "2023-01-21T10:21:40.410Z", "content": "Yes, I saw β¦ but β¦ one tip \"look into past comeptition\" π
", "rawMarkdown": "Yes, I saw ... but ... one tip \"look into past comeptition\" π", "votes": 1 }, { "id": 2109335, "postDate": "2023-01-21T10:32:28.220Z", "content": "Yes, unfortunately :D
", "rawMarkdown": "Yes, unfortunately :D " } ] }, { "id": 2109365, "postDate": "2023-01-21T11:16:57.277Z", "content": "lol. They hold secrets close to their chest. :P
", "rawMarkdown": "lol. They hold secrets close to their chest. :P ", "votes": 2, "replies": [ { "id": 2109383, "postDate": "2023-01-21T11:26:52.220Z", "content": "Very good. This is competition. They push us to search for solutions and learn.
\n@init27 it would be great to talk with winners (as a competition summary) how they organized work during this competition.
Thank you π
\nHonestly, the reason I stopped interviewing so frequently was because I have interviewed many Kagglers multiple times. To the point, it was my 9th Interview with Psi yesterday π±
\nI treat high ranked competitors with my sincerest regards so I want to avoid wasting their time. I know people still enjoy agreeing to my requests but I personally feel conflicted if the interview doesn't have some unique element or is repetitive.
\nJust because of that, I have live-streamed by myself now for previous 3-4 months where I just explain the solutions to the audience.
\nNot sure if that's the best approach, I might revert to interviewing again but I'm a bit conflicted as you can tell
", "rawMarkdown": "Thank you π\n\nHonestly, the reason I stopped interviewing so frequently was because I have interviewed many Kagglers multiple times. To the point, it was my 9th Interview with Psi yesterday π±\n\nI treat high ranked competitors with my sincerest regards so I want to avoid wasting their time. I know people still enjoy agreeing to my requests but I personally feel conflicted if the interview doesn't have some unique element or is repetitive. \n\nJust because of that, I have live-streamed by myself now for previous 3-4 months where I just explain the solutions to the audience. \n\nNot sure if that's the best approach, I might revert to interviewing again but I'm a bit conflicted as you can tell", "votes": 4 }, { "id": 2111155, "postDate": "2023-01-22T16:30:16.050Z", "content": "Thank you @init27 for hosting these.
\nHave you tried inviting multiple top teams after a competition ends (with RSVP - that way you know how much time you will be live or questions to plan etc)
\nIt can be lively and quite informative discussion - different teams sharing and asking questions about their solutions. I believe top teams will also like this because there's lot to learn from different approaches to solving a problem.
Nice video, thanks, enjoyed a lot.
\nLot's of insights indeed on the way to train NN: the tips concerning the early stopping not to overfit the validation set are pretty insightfull !!!
", "rawMarkdown": "Nice video, thanks, enjoyed a lot. \n\nLot's of insights indeed on the way to train NN: the tips concerning the early stopping not to overfit the validation set are pretty insightfull !!! ", "votes": 1 }, { "id": 2110385, "postDate": "2023-01-22T07:01:03.963Z", "content": "Nice Video @init27 .Very mUch Informative . Sanyam ,you tried hard to create a fighting environment but Team Hydrogen is superb . All Of Them Often Same ViewPoint On A Topic.
\nIt is great to see that every time sanyam thought Now The Original Debate will Start ,
\n But Few Minutes Later Sanyam Be Like:
\nMay Be I have To Go For Another Oneπ
Very nice panel video. I watched twice. Lots of interesting insights and experiences. I recommend.
\nUnfortunately no (direct) answer how to improve score in rsna π
Thank you Remek, I'm happy you enjoyed the discussion! π
\nYeah, the audience tried really hard to extract Team Hydrogen's insights but they didn't spill any chai π
", "votes": 3, "replies": [ { "id": 2109327, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-01-21T10:21:40.410000", "content": "Yes, I saw β¦ but β¦ one tip \"look into past comeptition\" π
", "votes": 1, "replies": [] }, { "id": 2109335, "author_name": "slime", "author_url": "", "post_date": "2023-01-21T10:32:28.220000", "content": "Yes, unfortunately :D
", "votes": 0, "replies": [] } ] }, { "id": 2109365, "author_name": "Phaedrus", "author_url": "", "post_date": "2023-01-21T11:16:57.277000", "content": "lol. They hold secrets close to their chest. :P
", "votes": 2, "replies": [ { "id": 2109383, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-01-21T11:26:52.220000", "content": "Very good. This is competition. They push us to search for solutions and learn.
\n@init27 it would be great to talk with winners (as a competition summary) how they organized work during this competition.
Thank you π
\nHonestly, the reason I stopped interviewing so frequently was because I have interviewed many Kagglers multiple times. To the point, it was my 9th Interview with Psi yesterday π±
\nI treat high ranked competitors with my sincerest regards so I want to avoid wasting their time. I know people still enjoy agreeing to my requests but I personally feel conflicted if the interview doesn't have some unique element or is repetitive.
\nJust because of that, I have live-streamed by myself now for previous 3-4 months where I just explain the solutions to the audience.
\nNot sure if that's the best approach, I might revert to interviewing again but I'm a bit conflicted as you can tell
", "votes": 4, "replies": [] }, { "id": 2111155, "author_name": "RB", "author_url": "", "post_date": "2023-01-22T16:30:16.050000", "content": "Thank you @init27 for hosting these.
\nHave you tried inviting multiple top teams after a competition ends (with RSVP - that way you know how much time you will be live or questions to plan etc)
\nIt can be lively and quite informative discussion - different teams sharing and asking questions about their solutions. I believe top teams will also like this because there's lot to learn from different approaches to solving a problem.
Nice video, thanks, enjoyed a lot.
\nLot's of insights indeed on the way to train NN: the tips concerning the early stopping not to overfit the validation set are pretty insightfull !!!
", "votes": 1, "replies": [] }, { "id": 2110385, "author_name": "SOUMENDRA PRASAD MOHANTY", "author_url": "", "post_date": "2023-01-22T07:01:03.963000", "content": "Nice Video @init27 .Very mUch Informative . Sanyam ,you tried hard to create a fighting environment but Team Hydrogen is superb . All Of Them Often Same ViewPoint On A Topic.
\nIt is great to see that every time sanyam thought Now The Original Debate will Start ,
\n But Few Minutes Later Sanyam Be Like:
\nMay Be I have To Go For Another Oneπ