I have just taken a look at the leaderboard for this competition and I have noticed an 4989282 isoscore string has formed, which has been created by well over thirty people submitting the output file from the currently highest scoring public notebook \"ingv_catboost_baseline+tsfresh\" by Alexander Lyubchenko.
\nI think it is worth mentioning that
\nThis is not a code competition, so submitting somebody else's output file is perfectly fine (and sometimes even results in a medal!) But in this case there is absolutely no upside at all. There is also very little to be gained by ensembling high-scoring output files near the end of the competition either (unless, that is, you wish to hone your blending skills, which is a machine learning sub-genre in itself).
\nThe nice thing about this competition is that it is all about learning and exploring, so if you are stuck, or new, my advice is to fork his notebook (and upvote it at the same time to acknowledge his hard work), tweak a few parameters, and then run his script for yourself. The score probably won't change much, and may even be quite a bit lower, but you will be learning, and that is by far the best prize!
\nAll the best,
\ncarl
I have just taken a look at the leaderboard for this competition and I have noticed an 4989282 isoscore string has formed, which has been created by well over thirty people submitting the output file from the currently highest scoring public notebook \"ingv_catboost_baseline+tsfresh\" by Alexander Lyubchenko.
\nI think it is worth mentioning that
\nThis is not a code competition, so submitting somebody else's output file is perfectly fine (and sometimes even results in a medal!) But in this case there is absolutely no upside at all. There is also very little to be gained by ensembling high-scoring output files near the end of the competition either (unless, that is, you wish to hone your blending skills, which is a machine learning sub-genre in itself).
\nThe nice thing about this competition is that it is all about learning and exploring, so if you are stuck, or new, my advice is to fork his notebook (and upvote it at the same time to acknowledge his hard work), tweak a few parameters, and then run his script for yourself. The score probably won't change much, and may even be quite a bit lower, but you will be learning, and that is by far the best prize!
\nAll the best,
\ncarl
Kaggle is gonna Kaggle…
\nThis competition is in playground and might have a lot of newer people. I think it is understandable that those newer to Kaggle might see a high scoring notebook and run it through and submit, but it is discouraging to see those of Expert rank or higher doing the same and then abandoning the competition.
\nThe point of Kaggle is to learn, and to your point the aforementioned notebook has a lot of features to explore, you could do a lot with feature selection and optimization. You could probably get a similar score using XGBoost and paring down the feature list. The dimension of the feature vector (~7700) is larger than the number of observations. There's a lot of room for improvement, would like to see people take up the challenge.
", "rawMarkdown": "Kaggle is gonna Kaggle...\n\nThis competition is in playground and might have a lot of newer people. I think it is understandable that those newer to Kaggle might see a high scoring notebook and run it through and submit, but it is discouraging to see those of Expert rank or higher doing the same and then abandoning the competition.\n\nThe point of Kaggle is to learn, and to your point the aforementioned notebook has a lot of features to explore, you could do a lot with feature selection and optimization. You could probably get a similar score using XGBoost and paring down the feature list. The dimension of the feature vector (~7700) is larger than the number of observations. There's a lot of room for improvement, would like to see people take up the challenge.", "votes": null } ], "comments": [ { "id": 1085637, "author_name": "ajcostarino", "author_url": "", "post_date": "11/21/2020 03:35:20", "content": "Kaggle is gonna Kaggle…
\nThis competition is in playground and might have a lot of newer people. I think it is understandable that those newer to Kaggle might see a high scoring notebook and run it through and submit, but it is discouraging to see those of Expert rank or higher doing the same and then abandoning the competition.
\nThe point of Kaggle is to learn, and to your point the aforementioned notebook has a lot of features to explore, you could do a lot with feature selection and optimization. You could probably get a similar score using XGBoost and paring down the feature list. The dimension of the feature vector (~7700) is larger than the number of observations. There's a lot of room for improvement, would like to see people take up the challenge.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1084539": "I have just taken a look at the [leaderboard](https://www.kaggle.com/c/predict-volcanic-eruptions-ingv-oe/leaderboard) for this competition and I have noticed an 4989282 [isoscore string](https://www.kaggle.com/carlmcbrideellis/shakeup-scatterplots-boxes-strings-and-things) has formed, which has been created by well over thirty people submitting the output file from the currently highest scoring public notebook [*\"ingv_catboost_baseline+tsfresh\"*](https://www.kaggle.com/carpediemamigo/ingv-catboost-baseline-tsfresh) by [Alexander Lyubchenko](https://www.kaggle.com/carpediemamigo). \n\nI think it is worth mentioning that\n\n* This competition does not award ranking points\n* This competition does not count towards tiers\n\nThis is not a code competition, so submitting somebody else's output file is perfectly fine (and sometimes even results in a medal!) But in this case there is absolutely no *upside* at all. There is also very little to be gained by ensembling high-scoring output files near the end of the competition either (unless, that is, you wish to hone your [blending skills](https://mlwave.com/kaggle-ensembling-guide/), which is a machine learning sub-genre in itself). \n\nThe nice thing about this competition is that it is all about learning and exploring, so if you are stuck, or new, my advice is to fork his notebook (and upvote it at the same time to acknowledge his hard work), tweak a few parameters, and then run his script for yourself. The score probably won't change much, and may even be quite a bit lower, but you will be learning, and that is *by far* the best prize!\n\nAll the best,\ncarl", "1085637": "Kaggle is gonna Kaggle...\n\nThis competition is in playground and might have a lot of newer people. I think it is understandable that those newer to Kaggle might see a high scoring notebook and run it through and submit, but it is discouraging to see those of Expert rank or higher doing the same and then abandoning the competition.\n\nThe point of Kaggle is to learn, and to your point the aforementioned notebook has a lot of features to explore, you could do a lot with feature selection and optimization. You could probably get a similar score using XGBoost and paring down the feature list. The dimension of the feature vector (~7700) is larger than the number of observations. There's a lot of room for improvement, would like to see people take up the challenge." }, "source": "meta" }