@ CompetitionHost: can you provide sample code that shows how to use the gresearch_crypto module to predict with time-lagged features.
\nI wanted to use \"Open\", \"High\", \"Low\" and \"Close\" as time lagged features. (To predict target at time t use feature Close at time t-1,t-2 etc.)
\nI have not yet been able to identify a clear method in the description that is 100% reliable even in phase 2
As a second point, it would be important for me to know which information train and supplemental_train contain in phase 2.
\nIs the content of train in phase 1 and 2 the same and does supplemental_train always contain the additional new data?
@ CompetitionHost: can you provide sample code that shows how to use the gresearch_crypto module to predict with time-lagged features.
\nI wanted to use \"Open\", \"High\", \"Low\" and \"Close\" as time lagged features. (To predict target at time t use feature Close at time t-1,t-2 etc.)
\nI have not yet been able to identify a clear method in the description that is 100% reliable even in phase 2
As a second point, it would be important for me to know which information train and supplemental_train contain in phase 2.
\nIs the content of train in phase 1 and 2 the same and does supplemental_train always contain the additional new data?
Not sure it is the host role to provide this code. But I've shared some On-line feature engineering notebook here (also see the complete discussion here). Regarding your second point, as I understood it train contain data up to the beginning of the competition and supplemental_train will contain new data and will be updated.
", "rawMarkdown": "Not sure it is the host role to provide this code. But I've shared some On-line feature engineering notebook [here](https://www.kaggle.com/lucasmorin/on-line-feature-engineering) (also see the complete discussion [here](https://www.kaggle.com/c/g-research-crypto-forecasting/discussion/293994)). Regarding your second point, as I understood it train contain data up to the beginning of the competition and supplemental_train will contain new data and will be updated.", "votes": null }, { "id": "1612177", "postDate": "12/08/2021 16:27:18", "content": "Since the submissions for this competition have to be made via notebooks, I think it is the role of the host to show how we can excess past records in the gresearch_crypto module to calculate and submit lagged features.
\nThere are great notebooks like this with lagged features but I couldn't find any of them with the lagged features calculated in the gresearch_crypto module for submit
On the second point, I also think that the supplemental_train will contain new data.
\nBut I'm not sure. We put a lot of time into developing the model, but at the end we have to guess what information is then available in phase 2
https://www.kaggle.com/sohier/detailed-api-introduction - They describe how to use the API for submissions here. I haven't been focusing on submissions yet, so I can't answer your question directly.
", "rawMarkdown": "https://www.kaggle.com/sohier/detailed-api-introduction - They describe how to use the API for submissions here. I haven't been focusing on submissions yet, so I can't answer your question directly.", "votes": null } ], "comments": [ { "id": 1611379, "author_name": "lucasmorin", "author_url": "", "post_date": "12/07/2021 23:18:13", "content": "Not sure it is the host role to provide this code. But I've shared some On-line feature engineering notebook here (also see the complete discussion here). Regarding your second point, as I understood it train contain data up to the beginning of the competition and supplemental_train will contain new data and will be updated.
", "votes": null, "replies": [ { "id": 1612177, "author_name": "allmendinger", "author_url": "", "post_date": "12/08/2021 16:27:18", "content": "Since the submissions for this competition have to be made via notebooks, I think it is the role of the host to show how we can excess past records in the gresearch_crypto module to calculate and submit lagged features.
\nThere are great notebooks like this with lagged features but I couldn't find any of them with the lagged features calculated in the gresearch_crypto module for submit
On the second point, I also think that the supplemental_train will contain new data.
\nBut I'm not sure. We put a lot of time into developing the model, but at the end we have to guess what information is then available in phase 2
https://www.kaggle.com/sohier/detailed-api-introduction - They describe how to use the API for submissions here. I haven't been focusing on submissions yet, so I can't answer your question directly.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1610483": "CompetitionHost: can you provide sample code that shows how to use the gresearch_crypto module to predict with time-lagged features.\nI wanted to use \"Open\", \"High\", \"Low\" and \"Close\" as time lagged features. (To predict target at time t use feature Close at time t-1,t-2 etc.) \nI have not yet been able to identify a clear method in the description that is 100% reliable even in phase 2\n\nAs a second point, it would be important for me to know which information train and supplemental_train contain in phase 2.\nIs the content of train in phase 1 and 2 the same and does supplemental_train always contain the additional new data?", "1611379": "Not sure it is the host role to provide this code. But I've shared some On-line feature engineering notebook [here](https://www.kaggle.com/lucasmorin/on-line-feature-engineering) (also see the complete discussion [here](https://www.kaggle.com/c/g-research-crypto-forecasting/discussion/293994)). Regarding your second point, as I understood it train contain data up to the beginning of the competition and supplemental_train will contain new data and will be updated.", "1612177": "Since the submissions for this competition have to be made via notebooks, I think it is the role of the host to show how we can excess past records in the gresearch_crypto module to calculate and submit lagged features. \nThere are great notebooks [like this](https://www.kaggle.com/cstein06/tutorial-to-the-g-research-crypto-competition) with lagged features but I couldn't find any of them with the lagged features calculated in the gresearch_crypto module for submit\n\n\nOn the second point, I also think that the supplemental_train will contain new data. \nBut I'm not sure. We put a lot of time into developing the model, but at the end we have to guess what information is then available in phase 2", "1612400": "https://www.kaggle.com/sohier/detailed-api-introduction - They describe how to use the API for submissions here. I haven't been focusing on submissions yet, so I can't answer your question directly." }, "source": "meta" }