{ "id": 340597, "title": "15-th place solution", "url": "/competitions/smartphone-decimeter-2022/discussion/340597", "author_name": "Egor Borisov", "post_date": "2022-07-30T00:44:58.641000", "votes": 13, "comment_count": 2, "views": 0, "content": "

We join the competition 10 days before the deadline, so the decision is simple and based on public notebooks:
\n1) Get positions following instructions from this notebook: https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py (rtk)
\n2) Get positions, speeds, and covariances from this notebook: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother (rwls)
\n3) Apply Kalman filter to rtk and rwls positions separately: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother
\n4) Apply kalman filter to raw rtk positions again: https://www.kaggle.com/code/tqa236/kalman-filter-hyperparameter-search-with-bo
\n5) Calculate and correct biases per phone following this notebook: https://www.kaggle.com/code/saitodevel01/gsdc-bias-correction, https://www.kaggle.com/code/saitodevel01/gsdc-bias-eda
\n6) Shift all positions to median difference with ground truth calculated on the train
\n7) Merge results with median and averaging with weights

\n

Some of the submissions hit the gold area but unfortunately, I choose other ones.
\nThanks for the competition and for the great notebooks!

", "messages": [ { "id": 1876646, "postDate": "2022-07-30T00:44:58.643Z", "content": "

We join the competition 10 days before the deadline, so the decision is simple and based on public notebooks:
\n1) Get positions following instructions from this notebook: https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py (rtk)
\n2) Get positions, speeds, and covariances from this notebook: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother (rwls)
\n3) Apply Kalman filter to rtk and rwls positions separately: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother
\n4) Apply kalman filter to raw rtk positions again: https://www.kaggle.com/code/tqa236/kalman-filter-hyperparameter-search-with-bo
\n5) Calculate and correct biases per phone following this notebook: https://www.kaggle.com/code/saitodevel01/gsdc-bias-correction, https://www.kaggle.com/code/saitodevel01/gsdc-bias-eda
\n6) Shift all positions to median difference with ground truth calculated on the train
\n7) Merge results with median and averaging with weights

\n

Some of the submissions hit the gold area but unfortunately, I choose other ones.
\nThanks for the competition and for the great notebooks!

", "rawMarkdown": "We join the competition 10 days before the deadline, so the decision is simple and based on public notebooks:\n1) Get positions following instructions from this notebook: https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py (rtk)\n2) Get positions, speeds, and covariances from this notebook: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother (rwls)\n3) Apply Kalman filter to rtk and rwls positions separately: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother\n4) Apply kalman filter to raw rtk positions again: https://www.kaggle.com/code/tqa236/kalman-filter-hyperparameter-search-with-bo\n5) Calculate and correct biases per phone following this notebook: https://www.kaggle.com/code/saitodevel01/gsdc-bias-correction, https://www.kaggle.com/code/saitodevel01/gsdc-bias-eda\n6) Shift all positions to median difference with ground truth calculated on the train\n7) Merge results with median and averaging with weights\n\nSome of the submissions hit the gold area but unfortunately, I choose other ones.\nThanks for the competition and for the great notebooks!\n\n\n", "votes": 13 }, { "id": 1877848, "postDate": "2022-07-31T03:35:06.583Z", "content": "

Thanks for sharing your process! Great job on getting 15th place!

", "rawMarkdown": "Thanks for sharing your process! Great job on getting 15th place!\n\n" }, { "id": 1876881, "postDate": "2022-07-30T06:21:12.640Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 1877848, "author_name": "The Devastator", "author_url": "", "post_date": "2022-07-31T03:35:06.583000", "content": "

Thanks for sharing your process! Great job on getting 15th place!

", "votes": 0, "replies": [] }, { "id": 1876881, "author_name": "", "author_url": "", "post_date": "2022-07-30T06:21:12.640000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "1876646": "We join the competition 10 days before the deadline, so the decision is simple and based on public notebooks:\n1) Get positions following instructions from this notebook: https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py (rtk)\n2) Get positions, speeds, and covariances from this notebook: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother (rwls)\n3) Apply Kalman filter to rtk and rwls positions separately: https://www.kaggle.com/code/taroz1461/carrier-smoothing-robust-wls-kalman-smoother\n4) Apply kalman filter to raw rtk positions again: https://www.kaggle.com/code/tqa236/kalman-filter-hyperparameter-search-with-bo\n5) Calculate and correct biases per phone following this notebook: https://www.kaggle.com/code/saitodevel01/gsdc-bias-correction, https://www.kaggle.com/code/saitodevel01/gsdc-bias-eda\n6) Shift all positions to median difference with ground truth calculated on the train\n7) Merge results with median and averaging with weights\n\nSome of the submissions hit the gold area but unfortunately, I choose other ones.\nThanks for the competition and for the great notebooks!\n\n\n", "1877848": "Thanks for sharing your process! Great job on getting 15th place!\n\n", "1876881": "" } }