This competition is pretty much done and dusted for a lot of folks interested in learning and practising Machine Learning. The web scraping and external data has ruined the leaderboard and probably most of the top models will be exploiting this in some way, rendering them useless in production.
\n\nI decided to do something more useful and implemented Facebook's Prophet model to see how well a newer time series model performs compared to boosting methods. Time series models cannot seem to match up to the performance of boosting models in most Kaggle competitions and this one is no different.
\n\nLink to Notebook: https://www.kaggle.com/rohanrao/ashrae-prophet-s-prophecy
\nNote that this is a pure time series model without exploiting anything nor using any leak or external data. The notebook is for learning purposes (so do not expect an exceptional performance) and provides a basic end-to-end flow for Prophet without any optimizations.
There's more to Kaggle than just leaks and leaderboard.
", "messages": [ { "id": 681432, "postDate": "2019-11-26T05:24:29.593Z", "content": "This competition is pretty much done and dusted for a lot of folks interested in learning and practising Machine Learning. The web scraping and external data has ruined the leaderboard and probably most of the top models will be exploiting this in some way, rendering them useless in production.
\n\nI decided to do something more useful and implemented Facebook's Prophet model to see how well a newer time series model performs compared to boosting methods. Time series models cannot seem to match up to the performance of boosting models in most Kaggle competitions and this one is no different.
\n\nLink to Notebook: https://www.kaggle.com/rohanrao/ashrae-prophet-s-prophecy
\nNote that this is a pure time series model without exploiting anything nor using any leak or external data. The notebook is for learning purposes (so do not expect an exceptional performance) and provides a basic end-to-end flow for Prophet without any optimizations.
There's more to Kaggle than just leaks and leaderboard.
", "rawMarkdown": "This competition is pretty much done and dusted for a lot of folks interested in learning and practising Machine Learning. The web scraping and external data has ruined the leaderboard and probably most of the top models will be exploiting this in some way, rendering them useless in production.\n\nI decided to do something more useful and implemented Facebook's Prophet model to see how well a newer time series model performs compared to boosting methods. Time series models cannot seem to match up to the performance of boosting models in most Kaggle competitions and this one is no different.\n\nLink to Notebook: https://www.kaggle.com/rohanrao/ashrae-prophet-s-prophecy \nNote that this is a pure time series model without exploiting anything nor using any leak or external data. The notebook is for learning purposes (so do not expect an exceptional performance) and provides a basic end-to-end flow for Prophet without any optimizations.\n\nThere's more to Kaggle than just leaks and leaderboard.", "votes": 22 }, { "id": 681922, "postDate": "2019-11-26T17:26:54.853Z", "content": "I am here to learn,so I just keep away from leaked data.π @rohanrao
", "rawMarkdown": "I am here to learn,so I just keep away from leaked data.π @rohanrao " }, { "id": 681770, "postDate": "2019-11-26T14:30:16.817Z", "content": "I appreciated your last words \"There's more to Kaggle than just leaks and leaderboard.\"
", "rawMarkdown": "I appreciated your last words \"There's more to Kaggle than just leaks and leaderboard.\"" }, { "id": 681571, "postDate": "2019-11-26T09:08:07.303Z", "content": "I tried using Prophet to do building-wise regression just like you but just looking at the graph of Prophet's prediction it was completely useless so I didn't even bother submitting. I am a complete newbie so this may be due to my poor abilities. Do you find the predictions you get are somewhat plausible? Are you planning to submit them?
", "rawMarkdown": "I tried using Prophet to do building-wise regression just like you but just looking at the graph of Prophet's prediction it was completely useless so I didn't even bother submitting. I am a complete newbie so this may be due to my poor abilities. Do you find the predictions you get are somewhat plausible? Are you planning to submit them?" }, { "id": 681524, "postDate": "2019-11-26T07:42:36.883Z", "content": "I am not going to give up. My only purpose is to get better understanding of time series modelling. Therefore i am still working on my model without glancing for external data.\nThanks for sharing your approach which is a completely new idea for me, and best regards
", "rawMarkdown": "I am not going to give up. My only purpose is to get better understanding of time series modelling. Therefore i am still working on my model without glancing for external data.\nThanks for sharing your approach which is a completely new idea for me, and best regards" }, { "id": 681448, "postDate": "2019-11-26T05:55:51.720Z", "content": "I am trying to use (GP)genetic programming, and it is working but not improved generation by generation Since I am not sufficiently understanding this yet. still have to learn what's wrong in the code ;)
", "rawMarkdown": "I am trying to use (GP)genetic programming, and it is working but not improved generation by generation Since I am not sufficiently understanding this yet. still have to learn what's wrong in the code ;)" }, { "id": 681433, "postDate": "2019-11-26T05:28:34.870Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 681439, "postDate": "2019-11-26T05:36:15.360Z", "rawMarkdown": "" } ] }, { "id": 683252, "postDate": "2019-11-28T08:44:53.350Z", "content": "Thanks for sharing.
", "rawMarkdown": "Thanks for sharing." } ], "comments": [ { "id": 681922, "author_name": "Shahules", "author_url": "", "post_date": "2019-11-26T17:26:54.853000", "content": "I am here to learn,so I just keep away from leaked data.π @rohanrao
", "votes": 0, "replies": [] }, { "id": 681770, "author_name": "MarΓlia Prata", "author_url": "", "post_date": "2019-11-26T14:30:16.817000", "content": "I appreciated your last words \"There's more to Kaggle than just leaks and leaderboard.\"
", "votes": 0, "replies": [] }, { "id": 681571, "author_name": "Larry D.", "author_url": "", "post_date": "2019-11-26T09:08:07.303000", "content": "I tried using Prophet to do building-wise regression just like you but just looking at the graph of Prophet's prediction it was completely useless so I didn't even bother submitting. I am a complete newbie so this may be due to my poor abilities. Do you find the predictions you get are somewhat plausible? Are you planning to submit them?
", "votes": 0, "replies": [] }, { "id": 681524, "author_name": "Marek Swieton", "author_url": "", "post_date": "2019-11-26T07:42:36.883000", "content": "I am not going to give up. My only purpose is to get better understanding of time series modelling. Therefore i am still working on my model without glancing for external data.\nThanks for sharing your approach which is a completely new idea for me, and best regards
", "votes": 0, "replies": [] }, { "id": 681448, "author_name": "Hanjoon Choe", "author_url": "", "post_date": "2019-11-26T05:55:51.720000", "content": "I am trying to use (GP)genetic programming, and it is working but not improved generation by generation Since I am not sufficiently understanding this yet. still have to learn what's wrong in the code ;)
", "votes": 0, "replies": [] }, { "id": 681433, "author_name": "", "author_url": "", "post_date": "2019-11-26T05:28:34.870000", "content": "", "votes": 0, "replies": [ { "id": 681439, "author_name": "Vopani", "author_url": "", "post_date": "2019-11-26T05:36:15.360000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 683252, "author_name": "Prachi", "author_url": "", "post_date": "2019-11-28T08:44:53.350000", "content": "Thanks for sharing.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "681432": "This competition is pretty much done and dusted for a lot of folks interested in learning and practising Machine Learning. The web scraping and external data has ruined the leaderboard and probably most of the top models will be exploiting this in some way, rendering them useless in production.\n\nI decided to do something more useful and implemented Facebook's Prophet model to see how well a newer time series model performs compared to boosting methods. Time series models cannot seem to match up to the performance of boosting models in most Kaggle competitions and this one is no different.\n\nLink to Notebook: https://www.kaggle.com/rohanrao/ashrae-prophet-s-prophecy \nNote that this is a pure time series model without exploiting anything nor using any leak or external data. The notebook is for learning purposes (so do not expect an exceptional performance) and provides a basic end-to-end flow for Prophet without any optimizations.\n\nThere's more to Kaggle than just leaks and leaderboard.", "681922": "I am here to learn,so I just keep away from leaked data.π @rohanrao ", "681770": "I appreciated your last words \"There's more to Kaggle than just leaks and leaderboard.\"", "681571": "I tried using Prophet to do building-wise regression just like you but just looking at the graph of Prophet's prediction it was completely useless so I didn't even bother submitting. I am a complete newbie so this may be due to my poor abilities. Do you find the predictions you get are somewhat plausible? Are you planning to submit them?", "681524": "I am not going to give up. My only purpose is to get better understanding of time series modelling. Therefore i am still working on my model without glancing for external data.\nThanks for sharing your approach which is a completely new idea for me, and best regards", "681448": "I am trying to use (GP)genetic programming, and it is working but not improved generation by generation Since I am not sufficiently understanding this yet. still have to learn what's wrong in the code ;)", "681433": "", "683252": "Thanks for sharing." } }