{ "id": 357115, "title": "Congratulations to the winners(top scorers), and Appreciate the host, Digital Turbine!", "url": "/competitions/digital-turbine-auction-bid-price-prediction/writeups/kentak0928-congratulations-to-the-winners-top-scor", "author_name": "", "post_date": "2022-10-03T08:36:08.057186600Z", "votes": 7, "comment_count": 13, "views": 0, "content": "

I would like to congratulate the winners(top scorers)!
\nAlso, I highly appreciate the host, Digital Turbine, preparing such wonderful opportunities to learn data science.

\n

Actually this competition was a great challenge for me, because I was involved in huge shake down on another completion lately.
\nOn this competition, the data amount of train-data is over 7 millions, and that of test-data is only around 40 thousands, moreover public LB was calculated based on around 20 thousands, it means public LB indicates only 0.2-0.3%.
\nThe situation was pretty similar to the competition which I was involved in big shake down on.

\n

Hence, I firstly checked the data distributions of train-data and test-data, and grasped which variables should be careful with Adversarial-validation concept.
\n(I shared the coding yesterday.)

\n

My approach was, , ,

\n\n

The validation score is 7.309
\nThe Public LB score is 16.01302
\nThe Private LB score is 16.49135

\n

Honestly I had wanted to apply XGBoost and CatBoos for stacking ensemble, but because of lack of memory, I couldn't…..even though I prepared \"memory reduction function\", haha…

\n

Hope that I will have an opportunity to participate in your another completion again in the future.

\n

Cheers,
\nKenta from Japan

", "messages": [ { "id": "1968863", "postDate": "10/03/2022 08:36:08", "content": "

I would like to congratulate the winners(top scorers)!
\nAlso, I highly appreciate the host, Digital Turbine, preparing such wonderful opportunities to learn data science.

\n

Actually this competition was a great challenge for me, because I was involved in huge shake down on another completion lately.
\nOn this competition, the data amount of train-data is over 7 millions, and that of test-data is only around 40 thousands, moreover public LB was calculated based on around 20 thousands, it means public LB indicates only 0.2-0.3%.
\nThe situation was pretty similar to the competition which I was involved in big shake down on.

\n

Hence, I firstly checked the data distributions of train-data and test-data, and grasped which variables should be careful with Adversarial-validation concept.
\n(I shared the coding yesterday.)

\n

My approach was, , ,

\n\n

The validation score is 7.309
\nThe Public LB score is 16.01302
\nThe Private LB score is 16.49135

\n

Honestly I had wanted to apply XGBoost and CatBoos for stacking ensemble, but because of lack of memory, I couldn't…..even though I prepared \"memory reduction function\", haha…

\n

Hope that I will have an opportunity to participate in your another completion again in the future.

\n

Cheers,
\nKenta from Japan

", "rawMarkdown": "I would like to congratulate the winners(top scorers)!\nAlso, I highly appreciate the host, Digital Turbine, preparing such wonderful opportunities to learn data science.\n\nActually this competition was a great challenge for me, because I was involved in huge shake down on another completion lately.\nOn this competition, the data amount of train-data is over 7 millions, and that of test-data is only around 40 thousands, moreover public LB was calculated based on around 20 thousands, it means public LB indicates only 0.2-0.3%.\nThe situation was pretty similar to the competition which I was involved in big shake down on.\n\nHence, I firstly checked the data distributions of train-data and test-data, and grasped which variables should be careful with Adversarial-validation concept.\n(I shared the coding yesterday.)\n\n\nMy approach was, , , \n\n- convert \"eventTimestamp\" into datetime\n\n- make new variables with converted eventTimestamps, such as month, day, date, hour\n (Through adversarial-validation, I understood there is a big difference between train and test on the variable.)\n\n- Grouping \"deviceId\".\n\n- LightGBM with the hyper parameters tuned manually.\n (I wanted to use Optuna, but data amount was huge, so I gave it up...)\n\n- StratifiedKFold and fold=15\n\n\nThe validation score is 7.309\nThe Public LB score is 16.01302\nThe Private LB score is 16.49135\n\n\nHonestly I had wanted to apply XGBoost and CatBoos for stacking ensemble, but because of lack of memory, I couldn't.....even though I prepared \"memory reduction function\", haha...\n\n\nHope that I will have an opportunity to participate in your another completion again in the future.\n\nCheers, \nKenta from Japan", "votes": null }, { "id": "1968921", "postDate": "10/03/2022 09:15:47", "content": "

Thanks @kentak0928 for your participation and feedback. Congratulations on your high score!

\n

Together with @danielhen we spent a lot of time preparing this competition and especially the dataset, so I am happy you found it interesting.

", "rawMarkdown": "Thanks @kentak0928 for your participation and feedback. Congratulations on your high score! \n\nTogether with @danielhen we spent a lot of time preparing this competition and especially the dataset, so I am happy you found it interesting.", "votes": null }, { "id": "1969267", "postDate": "10/03/2022 12:04:03", "content": "

Cool @kentak0928

\n

Are you planning to apply for the DS position?

", "rawMarkdown": "Cool @kentak0928 \n\nAre you planning to apply for the DS position?", "votes": null }, { "id": "1969271", "postDate": "10/03/2022 12:08:06", "content": "

@kentak0928 this is a great feedback.
\nIndeed I spent a lot of time on it. I am glad you enjoyed it!

", "rawMarkdown": "kentak0928 this is a great feedback.\nIndeed I spent a lot of time on it. I am glad you enjoyed it!", "votes": null }, { "id": "1969365", "postDate": "10/03/2022 13:01:16", "content": "

Thanks for the comment and for the compliment too😄
\nWhen I have an opportunity to join your next competition, I will definitely invite friends of mine too.

", "rawMarkdown": "Thanks for the comment and for the compliment too😄\nWhen I have an opportunity to join your next competition, I will definitely invite friends of mine too.", "votes": null }, { "id": "1969368", "postDate": "10/03/2022 13:03:28", "content": "

Thanks for the comment😄
\nWhen I explore your data-set(EDA), I felt there were a lot of tips for actual situation of data science.
\nLook forward that you will hold a competition again soon.

", "rawMarkdown": "Thanks for the comment😄\nWhen I explore your data-set(EDA), I felt there were a lot of tips for actual situation of data science.\nLook forward that you will hold a competition again soon.", "votes": null }, { "id": "1969370", "postDate": "10/03/2022 13:05:47", "content": "

Thanks for the comment, ArchaeonSeq San😄
\nActually I'm already 45 years old, thus it could be difficult to change my career, haha.
\nHowever, I'm planning to make use of machine-learning skills and causal inference concepts for my profession in the future.

", "rawMarkdown": "Thanks for the comment, ArchaeonSeq San😄\nActually I'm already 45 years old, thus it could be difficult to change my career, haha.\nHowever, I'm planning to make use of machine-learning skills and causal inference concepts for my profession in the future.", "votes": null }, { "id": "1969401", "postDate": "10/03/2022 13:23:03", "content": "

Oh okay. That sounds cool 😄.
\nI see you also took part in the Predict Potential Spammers on Fiverr competition. I was the 4th(2nd in public Leaderboard) in that competition, even though I later ran the same notebook that gave me that position and was awarded a score that should have placed me in the 2nd position. That competition was one of the reasons I couldn't focus on this one(after submitting my initial baselines) and had to rush it in the last 2 days.

", "rawMarkdown": "Oh okay. That sounds cool 😄.\nI see you also took part in the Predict Potential Spammers on Fiverr competition. I was the 4th(2nd in public Leaderboard) in that competition, even though I later ran the same notebook that gave me that position and was awarded a score that should have placed me in the 2nd position. That competition was one of the reasons I couldn't focus on this one(after submitting my initial baselines) and had to rush it in the last 2 days.", "votes": null }, { "id": "1969440", "postDate": "10/03/2022 13:39:09", "content": "

@kentak0928 you are never too old, and with the score like that, you could be a Top Data Scientist if you wanted. Just saying :)

", "rawMarkdown": "kentak0928 you are never too old, and with the score like that, you could be a Top Data Scientist if you wanted. Just saying :)", "votes": null }, { "id": "1969470", "postDate": "10/03/2022 13:55:40", "content": "

Yes @kentak0928, I think you can definitely become a Top DS as well. Your Kaggle competition records looks awesome.

", "rawMarkdown": "Yes @kentak0928, I think you can definitely become a Top DS as well. Your Kaggle competition records looks awesome.", "votes": null }, { "id": "1970173", "postDate": "10/03/2022 22:39:03", "content": "

Thanks for the warm message, Narsil San😄 I will keep making the best effort to improve my data science skills😄💪

", "rawMarkdown": "Thanks for the warm message, Narsil San😄 I will keep making the best effort to improve my data science skills😄💪", "votes": null }, { "id": "1970174", "postDate": "10/03/2022 22:41:58", "content": "

Thanks for the kind message, ArchaeonSeq San too😄 I really enjoy learning this field, hope that we(you and I) will be able to keep learning it with a great fun♩Have a keep trying the best each other😆

", "rawMarkdown": "Thanks for the kind message, ArchaeonSeq San too😄 I really enjoy learning this field, hope that we(you and I) will be able to keep learning it with a great fun♩Have a keep trying the best each other😆", "votes": null }, { "id": "1972720", "postDate": "10/05/2022 09:46:11", "content": "

Your age is almost same with my brother, you're not old so much.

", "rawMarkdown": "Your age is almost same with my brother, you're not old so much.", "votes": null }, { "id": "1972841", "postDate": "10/05/2022 11:08:58", "content": "

Thanks for the warm message, Hiroh San as well😄 I will keep doing my best with feeling young….😅

", "rawMarkdown": "Thanks for the warm message, Hiroh San as well😄 I will keep doing my best with feeling young....😅", "votes": null } ], "comments": [ { "id": 1968921, "author_name": "narsil", "author_url": "", "post_date": "10/03/2022 09:15:47", "content": "

Thanks @kentak0928 for your participation and feedback. Congratulations on your high score!

\n

Together with @danielhen we spent a lot of time preparing this competition and especially the dataset, so I am happy you found it interesting.

", "votes": null, "replies": [ { "id": 1969365, "author_name": "kentak0928", "author_url": "", "post_date": "10/03/2022 13:01:16", "content": "

Thanks for the comment and for the compliment too😄
\nWhen I have an opportunity to join your next competition, I will definitely invite friends of mine too.

", "votes": null, "replies": [] } ] }, { "id": 1969267, "author_name": "pearsejim01", "author_url": "", "post_date": "10/03/2022 12:04:03", "content": "

Cool @kentak0928

\n

Are you planning to apply for the DS position?

", "votes": null, "replies": [ { "id": 1969370, "author_name": "kentak0928", "author_url": "", "post_date": "10/03/2022 13:05:47", "content": "

Thanks for the comment, ArchaeonSeq San😄
\nActually I'm already 45 years old, thus it could be difficult to change my career, haha.
\nHowever, I'm planning to make use of machine-learning skills and causal inference concepts for my profession in the future.

", "votes": null, "replies": [] }, { "id": 1969401, "author_name": "pearsejim01", "author_url": "", "post_date": "10/03/2022 13:23:03", "content": "

Oh okay. That sounds cool 😄.
\nI see you also took part in the Predict Potential Spammers on Fiverr competition. I was the 4th(2nd in public Leaderboard) in that competition, even though I later ran the same notebook that gave me that position and was awarded a score that should have placed me in the 2nd position. That competition was one of the reasons I couldn't focus on this one(after submitting my initial baselines) and had to rush it in the last 2 days.

", "votes": null, "replies": [] }, { "id": 1969440, "author_name": "narsil", "author_url": "", "post_date": "10/03/2022 13:39:09", "content": "

@kentak0928 you are never too old, and with the score like that, you could be a Top Data Scientist if you wanted. Just saying :)

", "votes": null, "replies": [] }, { "id": 1969470, "author_name": "pearsejim01", "author_url": "", "post_date": "10/03/2022 13:55:40", "content": "

Yes @kentak0928, I think you can definitely become a Top DS as well. Your Kaggle competition records looks awesome.

", "votes": null, "replies": [] }, { "id": 1970173, "author_name": "kentak0928", "author_url": "", "post_date": "10/03/2022 22:39:03", "content": "

Thanks for the warm message, Narsil San😄 I will keep making the best effort to improve my data science skills😄💪

", "votes": null, "replies": [] }, { "id": 1970174, "author_name": "kentak0928", "author_url": "", "post_date": "10/03/2022 22:41:58", "content": "

Thanks for the kind message, ArchaeonSeq San too😄 I really enjoy learning this field, hope that we(you and I) will be able to keep learning it with a great fun♩Have a keep trying the best each other😆

", "votes": null, "replies": [] }, { "id": 1972720, "author_name": "hiroh03", "author_url": "", "post_date": "10/05/2022 09:46:11", "content": "

Your age is almost same with my brother, you're not old so much.

", "votes": null, "replies": [] }, { "id": 1972841, "author_name": "kentak0928", "author_url": "", "post_date": "10/05/2022 11:08:58", "content": "

Thanks for the warm message, Hiroh San as well😄 I will keep doing my best with feeling young….😅

", "votes": null, "replies": [] } ] }, { "id": 1969271, "author_name": "danielhen", "author_url": "", "post_date": "10/03/2022 12:08:06", "content": "

@kentak0928 this is a great feedback.
\nIndeed I spent a lot of time on it. I am glad you enjoyed it!

", "votes": null, "replies": [ { "id": 1969368, "author_name": "kentak0928", "author_url": "", "post_date": "10/03/2022 13:03:28", "content": "

Thanks for the comment😄
\nWhen I explore your data-set(EDA), I felt there were a lot of tips for actual situation of data science.
\nLook forward that you will hold a competition again soon.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1968863": "I would like to congratulate the winners(top scorers)!\nAlso, I highly appreciate the host, Digital Turbine, preparing such wonderful opportunities to learn data science.\n\nActually this competition was a great challenge for me, because I was involved in huge shake down on another completion lately.\nOn this competition, the data amount of train-data is over 7 millions, and that of test-data is only around 40 thousands, moreover public LB was calculated based on around 20 thousands, it means public LB indicates only 0.2-0.3%.\nThe situation was pretty similar to the competition which I was involved in big shake down on.\n\nHence, I firstly checked the data distributions of train-data and test-data, and grasped which variables should be careful with Adversarial-validation concept.\n(I shared the coding yesterday.)\n\n\nMy approach was, , , \n\n- convert \"eventTimestamp\" into datetime\n\n- make new variables with converted eventTimestamps, such as month, day, date, hour\n (Through adversarial-validation, I understood there is a big difference between train and test on the variable.)\n\n- Grouping \"deviceId\".\n\n- LightGBM with the hyper parameters tuned manually.\n (I wanted to use Optuna, but data amount was huge, so I gave it up...)\n\n- StratifiedKFold and fold=15\n\n\nThe validation score is 7.309\nThe Public LB score is 16.01302\nThe Private LB score is 16.49135\n\n\nHonestly I had wanted to apply XGBoost and CatBoos for stacking ensemble, but because of lack of memory, I couldn't.....even though I prepared \"memory reduction function\", haha...\n\n\nHope that I will have an opportunity to participate in your another completion again in the future.\n\nCheers, \nKenta from Japan", "1968921": "Thanks @kentak0928 for your participation and feedback. Congratulations on your high score! \n\nTogether with @danielhen we spent a lot of time preparing this competition and especially the dataset, so I am happy you found it interesting.", "1969267": "Cool @kentak0928 \n\nAre you planning to apply for the DS position?", "1969271": "kentak0928 this is a great feedback.\nIndeed I spent a lot of time on it. I am glad you enjoyed it!", "1969365": "Thanks for the comment and for the compliment too😄\nWhen I have an opportunity to join your next competition, I will definitely invite friends of mine too.", "1969368": "Thanks for the comment😄\nWhen I explore your data-set(EDA), I felt there were a lot of tips for actual situation of data science.\nLook forward that you will hold a competition again soon.", "1969370": "Thanks for the comment, ArchaeonSeq San😄\nActually I'm already 45 years old, thus it could be difficult to change my career, haha.\nHowever, I'm planning to make use of machine-learning skills and causal inference concepts for my profession in the future.", "1969401": "Oh okay. That sounds cool 😄.\nI see you also took part in the Predict Potential Spammers on Fiverr competition. I was the 4th(2nd in public Leaderboard) in that competition, even though I later ran the same notebook that gave me that position and was awarded a score that should have placed me in the 2nd position. That competition was one of the reasons I couldn't focus on this one(after submitting my initial baselines) and had to rush it in the last 2 days.", "1969440": "kentak0928 you are never too old, and with the score like that, you could be a Top Data Scientist if you wanted. Just saying :)", "1969470": "Yes @kentak0928, I think you can definitely become a Top DS as well. Your Kaggle competition records looks awesome.", "1970173": "Thanks for the warm message, Narsil San😄 I will keep making the best effort to improve my data science skills😄💪", "1970174": "Thanks for the kind message, ArchaeonSeq San too😄 I really enjoy learning this field, hope that we(you and I) will be able to keep learning it with a great fun♩Have a keep trying the best each other😆", "1972720": "Your age is almost same with my brother, you're not old so much.", "1972841": "Thanks for the warm message, Hiroh San as well😄 I will keep doing my best with feeling young....😅" }, "source": "meta" }