{ "id": 118777, "title": "Leaving the competition", "url": "/competitions/ashrae-energy-prediction/discussion/118777", "author_name": "", "post_date": "2019-11-24T13:11:21.307181700Z", "votes": 35, "comment_count": 19, "views": 0, "content": "

With all the data scraping this competition has gotten extremely weird.

", "messages": [ { "id": "680323", "postDate": "11/24/2019 13:11:21", "content": "

With all the data scraping this competition has gotten extremely weird.

", "rawMarkdown": "With all the data scraping this competition has gotten extremely weird.", "votes": null }, { "id": "680357", "postDate": "11/24/2019 14:15:50", "content": "

When I initially started this competition, the best was around 1.15. Now its around 0.96. It has become more of who is able to identify the leaks /from which site and share the data to the competitors in this competition.

", "rawMarkdown": "When I initially started this competition, the best was around 1.15. Now its around 0.96. It has become more of who is able to identify the leaks /from which site and share the data to the competitors in this competition.", "votes": null }, { "id": "680508", "postDate": "11/24/2019 19:35:09", "content": "

Possibly the worst Kaggle competition ever?

", "rawMarkdown": "Possibly the worst Kaggle competition ever?", "votes": null }, { "id": "680704", "postDate": "11/25/2019 05:12:50", "content": "

I started this competition in order to learn about time series but with the time here, the competition is more about finding leakage and exploiting it. All the recent kernels talk about leaks. I have already lost interest and just come here to check new techniques.

", "rawMarkdown": "I started this competition in order to learn about time series but with the time here, the competition is more about finding leakage and exploiting it. All the recent kernels talk about leaks. I have already lost interest and just come here to check new techniques.", "votes": null }, { "id": "681226", "postDate": "11/25/2019 21:30:30", "content": "

Good decision Scirpus !

", "rawMarkdown": "Good decision Scirpus !", "votes": null }, { "id": "681227", "postDate": "11/25/2019 21:31:18", "content": "

Might be !!! Might be !

", "rawMarkdown": "Might be !!! Might be !", "votes": null }, { "id": "681228", "postDate": "11/25/2019 21:33:16", "content": "

\"more of who is able to identify the leaks \"

\n\n

Yesss, this is what about leaks ! Not a Competition, just leaks.

\n\n

A pity !

", "rawMarkdown": "\"more of who is able to identify the leaks \" \n\nYesss, this is what about leaks ! Not a Competition, just leaks.\n\nA pity !", "votes": null }, { "id": "681302", "postDate": "11/26/2019 00:35:39", "content": "

1.08 with no leaks is pretty easily achievable. I'm not sure how much better folks have achieved, but I'd guess 1.05 would be a reasonable target.

", "rawMarkdown": "1.08 with no leaks is pretty easily achievable. I'm not sure how much better folks have achieved, but I'd guess 1.05 would be a reasonable target.", "votes": null }, { "id": "681413", "postDate": "11/26/2019 04:46:02", "content": "

It is not the case for me though. ㅠ_ㅠ

", "rawMarkdown": "It is not the case for me though. ㅠ_ㅠ", "votes": null }, { "id": "681438", "postDate": "11/26/2019 05:34:55", "content": "

You did some nice stuff if you achieved 0.99 without leaks.

", "rawMarkdown": "You did some nice stuff if you achieved 0.99 without leaks.", "votes": null }, { "id": "681552", "postDate": "11/26/2019 08:27:07", "content": "

Google Analytics Customer Revenue Prediction is the worst game

", "rawMarkdown": "Google Analytics Customer Revenue Prediction is the worst game", "votes": null }, { "id": "682163", "postDate": "11/27/2019 02:57:50", "content": "

This competition should be renamed as \"Leaks only\" competition!😂

", "rawMarkdown": "This competition should be renamed as \"Leaks only\" competition!😂", "votes": null }, { "id": "683662", "postDate": "11/28/2019 16:37:53", "content": "

Leaving the competition

", "rawMarkdown": "Leaving the competition", "votes": null }, { "id": "685850", "postDate": "12/02/2019 12:06:35", "content": "

Same with you. Just for finding skills dealing with time series. However, leaks everywhere😂

", "rawMarkdown": "Same with you. Just for finding skills dealing with time series. However, leaks everywhere😂", "votes": null }, { "id": "688003", "postDate": "12/05/2019 05:49:06", "content": "

That is why people can not improve their score by using the existing data anymore.

", "rawMarkdown": "That is why people can not improve their score by using the existing data anymore.", "votes": null }, { "id": "688030", "postDate": "12/05/2019 06:43:45", "content": "

@alisherianvar yeh,I know.I am facing the same situation.

", "rawMarkdown": "alisherianvar yeh,I know.I am facing the same situation.", "votes": null }, { "id": "693807", "postDate": "12/12/2019 20:02:29", "content": "

any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data... is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another. I realize many of the competitions are won that way. probably most but I'm never gonna do that. So I quit a while back too.

\n\n

i did make a new change to my algorithm that lets me fit GBM sets of training/test data based on groups in the data that then expand and get larger (in a cascading sort of way once improvements cant be found) works pretty well, does tend to fit to that data which might be an overfit if the data is noisy (like this competition) but i'm happy with it. so there's that. :D (but to that point my local score was like .91 the LB score was 1.22 .. likely cause of data cleanup issues that i'm like \"yep, not gonna deal with it\")

", "rawMarkdown": "any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data... is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another. I realize many of the competitions are won that way. probably most but I'm never gonna do that. So I quit a while back too. \n\ni did make a new change to my algorithm that lets me fit GBM sets of training/test data based on groups in the data that then expand and get larger (in a cascading sort of way once improvements cant be found) works pretty well, does tend to fit to that data which might be an overfit if the data is noisy (like this competition) but i'm happy with it. so there's that. :D (but to that point my local score was like .91 the LB score was 1.22 .. likely cause of data cleanup issues that i'm like \"yep, not gonna deal with it\")", "votes": null }, { "id": "693969", "postDate": "12/13/2019 02:11:07", "content": "
\n

any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data… is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another.

\n
\n\n

Here's where we differ. From the standpoint of \"clean competition\", you may be right, but from the standpoint of \"real-world data science\", this is much more real than the average Kaggle dataset. In my time as an ML professional, the vast majority of my efforts were in obtaining clean data. Feature engineering was secondary, and algorithm selection was almost a non-issue. I've had to become a domain expert in diapers (as a confirmed bachelor), automobiles, and yams just in order to get to the ML part of my job.

\n\n

I'm not a fan of the leaks and the web-scraping, but I consider the data cleanup to be quite a legitimate aspect of this, and any other, competition. On the other hand, there's lots of competitions out there, and it's good that you can choose not to bother with this one.

", "rawMarkdown": "> any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data… is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another.\n\nHere's where we differ. From the standpoint of \"clean competition\", you may be right, but from the standpoint of \"real-world data science\", this is much *more* real than the average Kaggle dataset. In my time as an ML professional, the vast majority of my efforts were in obtaining clean data. Feature engineering was secondary, and algorithm selection was almost a non-issue. I've had to become a domain expert in diapers (as a confirmed bachelor), automobiles, and yams just in order to get to the ML part of my job.\n\nI'm not a fan of the leaks and the web-scraping, but I consider the data cleanup to be quite a legitimate aspect of this, and any other, competition. On the other hand, there's lots of competitions out there, and it's good that you can choose not to bother with this one.", "votes": null }, { "id": "695245", "postDate": "12/14/2019 20:43:30", "content": "

The only real world data science i do routinely is with stocks. And my data comes in as clean as it can be but of course it costs money too. So other than purchasing more data and trying to select potentially relevant data, data work begins and ends there. (Which is really how i like it.)

\n\n

The fun part for me is writing an improved algorithm for getting the best angle on the data you have. That might be auto feature creation (i dont generally make my own features) or algorithm to tell me weighting or something that actually does slicing and dicing of the data in a new way (some new method or improvement on existing expert systems). but you are right most alogirthims are 6 of 1 half dozen of the other. And if you have a perfect view of the data there is only so much you can do before you are in the weeds as t were.

", "rawMarkdown": "The only real world data science i do routinely is with stocks. And my data comes in as clean as it can be but of course it costs money too. So other than purchasing more data and trying to select potentially relevant data, data work begins and ends there. (Which is really how i like it.)\n\nThe fun part for me is writing an improved algorithm for getting the best angle on the data you have. That might be auto feature creation (i dont generally make my own features) or algorithm to tell me weighting or something that actually does slicing and dicing of the data in a new way (some new method or improvement on existing expert systems). but you are right most alogirthims are 6 of 1 half dozen of the other. And if you have a perfect view of the data there is only so much you can do before you are in the weeds as t were.", "votes": null }, { "id": "697191", "postDate": "12/17/2019 15:55:01", "content": "

You did some nice stuff if you achieved 0.99 without leaks.

", "rawMarkdown": "You did some nice stuff if you achieved 0.99 without leaks.", "votes": null } ], "comments": [ { "id": 680357, "author_name": "manojprabhaakr", "author_url": "", "post_date": "11/24/2019 14:15:50", "content": "

When I initially started this competition, the best was around 1.15. Now its around 0.96. It has become more of who is able to identify the leaks /from which site and share the data to the competitors in this competition.

", "votes": null, "replies": [ { "id": 681228, "author_name": "pfichou", "author_url": "", "post_date": "11/25/2019 21:33:16", "content": "

\"more of who is able to identify the leaks \"

\n\n

Yesss, this is what about leaks ! Not a Competition, just leaks.

\n\n

A pity !

", "votes": null, "replies": [] }, { "id": 681302, "author_name": "purist1024", "author_url": "", "post_date": "11/26/2019 00:35:39", "content": "

1.08 with no leaks is pretty easily achievable. I'm not sure how much better folks have achieved, but I'd guess 1.05 would be a reasonable target.

", "votes": null, "replies": [] }, { "id": 681413, "author_name": "hanjoonchoe", "author_url": "", "post_date": "11/26/2019 04:46:02", "content": "

It is not the case for me though. ㅠ_ㅠ

", "votes": null, "replies": [] }, { "id": 681438, "author_name": "tejashshah", "author_url": "", "post_date": "11/26/2019 05:34:55", "content": "

You did some nice stuff if you achieved 0.99 without leaks.

", "votes": null, "replies": [] } ] }, { "id": 680508, "author_name": "stecasasso", "author_url": "", "post_date": "11/24/2019 19:35:09", "content": "

Possibly the worst Kaggle competition ever?

", "votes": null, "replies": [ { "id": 681227, "author_name": "pfichou", "author_url": "", "post_date": "11/25/2019 21:31:18", "content": "

Might be !!! Might be !

", "votes": null, "replies": [] }, { "id": 681552, "author_name": "laplaceplanet", "author_url": "", "post_date": "11/26/2019 08:27:07", "content": "

Google Analytics Customer Revenue Prediction is the worst game

", "votes": null, "replies": [] } ] }, { "id": 680704, "author_name": "tejashshah", "author_url": "", "post_date": "11/25/2019 05:12:50", "content": "

I started this competition in order to learn about time series but with the time here, the competition is more about finding leakage and exploiting it. All the recent kernels talk about leaks. I have already lost interest and just come here to check new techniques.

", "votes": null, "replies": [ { "id": 685850, "author_name": "kobehu", "author_url": "", "post_date": "12/02/2019 12:06:35", "content": "

Same with you. Just for finding skills dealing with time series. However, leaks everywhere😂

", "votes": null, "replies": [] } ] }, { "id": 681226, "author_name": "pfichou", "author_url": "", "post_date": "11/25/2019 21:30:30", "content": "

Good decision Scirpus !

", "votes": null, "replies": [] }, { "id": 682163, "author_name": "shahules", "author_url": "", "post_date": "11/27/2019 02:57:50", "content": "

This competition should be renamed as \"Leaks only\" competition!😂

", "votes": null, "replies": [ { "id": 688003, "author_name": "alisherianvar", "author_url": "", "post_date": "12/05/2019 05:49:06", "content": "

That is why people can not improve their score by using the existing data anymore.

", "votes": null, "replies": [] }, { "id": 688030, "author_name": "shahules", "author_url": "", "post_date": "12/05/2019 06:43:45", "content": "

@alisherianvar yeh,I know.I am facing the same situation.

", "votes": null, "replies": [] } ] }, { "id": 683662, "author_name": "triplelift", "author_url": "", "post_date": "11/28/2019 16:37:53", "content": "

Leaving the competition

", "votes": null, "replies": [] }, { "id": 693807, "author_name": "jscheibel", "author_url": "", "post_date": "12/12/2019 20:02:29", "content": "

any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data... is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another. I realize many of the competitions are won that way. probably most but I'm never gonna do that. So I quit a while back too.

\n\n

i did make a new change to my algorithm that lets me fit GBM sets of training/test data based on groups in the data that then expand and get larger (in a cascading sort of way once improvements cant be found) works pretty well, does tend to fit to that data which might be an overfit if the data is noisy (like this competition) but i'm happy with it. so there's that. :D (but to that point my local score was like .91 the LB score was 1.22 .. likely cause of data cleanup issues that i'm like \"yep, not gonna deal with it\")

", "votes": null, "replies": [ { "id": 693969, "author_name": "purist1024", "author_url": "", "post_date": "12/13/2019 02:11:07", "content": "
\n

any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data… is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another.

\n
\n\n

Here's where we differ. From the standpoint of \"clean competition\", you may be right, but from the standpoint of \"real-world data science\", this is much more real than the average Kaggle dataset. In my time as an ML professional, the vast majority of my efforts were in obtaining clean data. Feature engineering was secondary, and algorithm selection was almost a non-issue. I've had to become a domain expert in diapers (as a confirmed bachelor), automobiles, and yams just in order to get to the ML part of my job.

\n\n

I'm not a fan of the leaks and the web-scraping, but I consider the data cleanup to be quite a legitimate aspect of this, and any other, competition. On the other hand, there's lots of competitions out there, and it's good that you can choose not to bother with this one.

", "votes": null, "replies": [] }, { "id": 695245, "author_name": "jscheibel", "author_url": "", "post_date": "12/14/2019 20:43:30", "content": "

The only real world data science i do routinely is with stocks. And my data comes in as clean as it can be but of course it costs money too. So other than purchasing more data and trying to select potentially relevant data, data work begins and ends there. (Which is really how i like it.)

\n\n

The fun part for me is writing an improved algorithm for getting the best angle on the data you have. That might be auto feature creation (i dont generally make my own features) or algorithm to tell me weighting or something that actually does slicing and dicing of the data in a new way (some new method or improvement on existing expert systems). but you are right most alogirthims are 6 of 1 half dozen of the other. And if you have a perfect view of the data there is only so much you can do before you are in the weeds as t were.

", "votes": null, "replies": [] } ] }, { "id": 697191, "author_name": "tzozeze", "author_url": "", "post_date": "12/17/2019 15:55:01", "content": "

You did some nice stuff if you achieved 0.99 without leaks.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "680323": "With all the data scraping this competition has gotten extremely weird.", "680357": "When I initially started this competition, the best was around 1.15. Now its around 0.96. It has become more of who is able to identify the leaks /from which site and share the data to the competitors in this competition.", "680508": "Possibly the worst Kaggle competition ever?", "680704": "I started this competition in order to learn about time series but with the time here, the competition is more about finding leakage and exploiting it. All the recent kernels talk about leaks. I have already lost interest and just come here to check new techniques.", "681226": "Good decision Scirpus !", "681227": "Might be !!! Might be !", "681228": "\"more of who is able to identify the leaks \" \n\nYesss, this is what about leaks ! Not a Competition, just leaks.\n\nA pity !", "681302": "1.08 with no leaks is pretty easily achievable. I'm not sure how much better folks have achieved, but I'd guess 1.05 would be a reasonable target.", "681413": "It is not the case for me though. ㅠ_ㅠ", "681438": "You did some nice stuff if you achieved 0.99 without leaks.", "681552": "Google Analytics Customer Revenue Prediction is the worst game", "682163": "This competition should be renamed as \"Leaks only\" competition!😂", "683662": "Leaving the competition", "685850": "Same with you. Just for finding skills dealing with time series. However, leaks everywhere😂", "688003": "That is why people can not improve their score by using the existing data anymore.", "688030": "alisherianvar yeh,I know.I am facing the same situation.", "693807": "any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data... is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another. I realize many of the competitions are won that way. probably most but I'm never gonna do that. So I quit a while back too. \n\ni did make a new change to my algorithm that lets me fit GBM sets of training/test data based on groups in the data that then expand and get larger (in a cascading sort of way once improvements cant be found) works pretty well, does tend to fit to that data which might be an overfit if the data is noisy (like this competition) but i'm happy with it. so there's that. :D (but to that point my local score was like .91 the LB score was 1.22 .. likely cause of data cleanup issues that i'm like \"yep, not gonna deal with it\")", "693969": "> any competition where to win you have to spend lots (or really any) time cleaning or hand identifying bits of data… is a competition where you are doing their work for them to win. training and picking algorithms or trying to create features is one thing. becoming an analyst / expert in their data (to find bad/missing data) is another.\n\nHere's where we differ. From the standpoint of \"clean competition\", you may be right, but from the standpoint of \"real-world data science\", this is much *more* real than the average Kaggle dataset. In my time as an ML professional, the vast majority of my efforts were in obtaining clean data. Feature engineering was secondary, and algorithm selection was almost a non-issue. I've had to become a domain expert in diapers (as a confirmed bachelor), automobiles, and yams just in order to get to the ML part of my job.\n\nI'm not a fan of the leaks and the web-scraping, but I consider the data cleanup to be quite a legitimate aspect of this, and any other, competition. On the other hand, there's lots of competitions out there, and it's good that you can choose not to bother with this one.", "695245": "The only real world data science i do routinely is with stocks. And my data comes in as clean as it can be but of course it costs money too. So other than purchasing more data and trying to select potentially relevant data, data work begins and ends there. (Which is really how i like it.)\n\nThe fun part for me is writing an improved algorithm for getting the best angle on the data you have. That might be auto feature creation (i dont generally make my own features) or algorithm to tell me weighting or something that actually does slicing and dicing of the data in a new way (some new method or improvement on existing expert systems). but you are right most alogirthims are 6 of 1 half dozen of the other. And if you have a perfect view of the data there is only so much you can do before you are in the weeds as t were.", "697191": "You did some nice stuff if you achieved 0.99 without leaks." }, "source": "meta" }