Does anyone can explain how the time series API of this competition will work? I just joined this competition, but after some readings I am still not clear how does this work. I really appreciate if any of you can help me clear this?
\nDoes anyone can explain how the time series API of this competition will work? I just joined this competition, but after some readings I am still not clear how does this work. I really appreciate if any of you can help me clear this?
\nYou would have to create a template data frame to save pervious iterations. I would use a numpy array though because it should be faster. Just append to the end of the array the new data and remove the oldest timestep from the array that was needed for your greatest lagged feature in the previous timestep.
\nI think train will stay the same and the supplemental train will bridge the gap between train and test, but it would be nice to get confirmation.
", "rawMarkdown": "You would have to create a template data frame to save pervious iterations. I would use a numpy array though because it should be faster. Just append to the end of the array the new data and remove the oldest timestep from the array that was needed for your greatest lagged feature in the previous timestep.\n\nI think train will stay the same and the supplemental train will bridge the gap between train and test, but it would be nice to get confirmation.", "votes": 4, "replies": [ { "id": 1647545, "postDate": "2022-01-12T17:25:01.340Z", "content": "Yes, I also think \"Just append to the end of the array the new data and remove the oldest timestep from the array that was needed for your greatest lagged feature in the previous timestep.\" is workable.
\n\"I think train will stay the same and the supplemental train will bridge the gap between train and test, but it would be nice to get confirmation.\" I agree, it's better to get a confirmation from the competition host.
That's correct, the train.csv will stay the same and the supplemental will bridge the gap between train and test.
", "rawMarkdown": "That's correct, the train.csv will stay the same and the supplemental will bridge the gap between train and test.", "votes": 4 }, { "id": 1648426, "postDate": "2022-01-13T12:22:31.027Z", "content": "@cstein06 Thanks for confirming, very helpful. Does target label will also be provided in supplemental?
", "rawMarkdown": "@cstein06 Thanks for confirming, very helpful. Does target label will also be provided in supplemental?" }, { "id": 1648436, "postDate": "2022-01-13T12:28:06.110Z", "content": "Yes. We will give more info on the supplemental later this week.
", "rawMarkdown": "Yes. We will give more info on the supplemental later this week.", "votes": 2 }, { "id": 1648440, "postDate": "2022-01-13T12:34:34.240Z", "content": "Fabulous!!!
", "rawMarkdown": "Fabulous!!!" } ] }, { "id": 1647111, "postDate": "2022-01-12T10:16:35.810Z", "content": "@andrewscherer can you help to clarify this?
", "rawMarkdown": "@andrewscherer can you help to clarify this?", "votes": 1 }, { "id": 1647964, "postDate": "2022-01-13T03:00:30.407Z", "content": "Nice question! Have the same question before, thanks!
", "rawMarkdown": "Nice question! Have the same question before, thanks!" }, { "id": 1647466, "postDate": "2022-01-12T16:19:54.533Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 1647442, "author_name": "Kyle Peters", "author_url": "", "post_date": "2022-01-12T16:05:09.310000", "content": "You would have to create a template data frame to save pervious iterations. I would use a numpy array though because it should be faster. Just append to the end of the array the new data and remove the oldest timestep from the array that was needed for your greatest lagged feature in the previous timestep.
\nI think train will stay the same and the supplemental train will bridge the gap between train and test, but it would be nice to get confirmation.
", "votes": 4, "replies": [ { "id": 1647545, "author_name": "Jie Wu", "author_url": "", "post_date": "2022-01-12T17:25:01.340000", "content": "Yes, I also think \"Just append to the end of the array the new data and remove the oldest timestep from the array that was needed for your greatest lagged feature in the previous timestep.\" is workable.
\n\"I think train will stay the same and the supplemental train will bridge the gap between train and test, but it would be nice to get confirmation.\" I agree, it's better to get a confirmation from the competition host.
That's correct, the train.csv will stay the same and the supplemental will bridge the gap between train and test.
", "votes": 4, "replies": [] }, { "id": 1648426, "author_name": "Jie Wu", "author_url": "", "post_date": "2022-01-13T12:22:31.027000", "content": "@cstein06 Thanks for confirming, very helpful. Does target label will also be provided in supplemental?
", "votes": 0, "replies": [] }, { "id": 1648436, "author_name": "Carlos Stein N Brito", "author_url": "", "post_date": "2022-01-13T12:28:06.110000", "content": "Yes. We will give more info on the supplemental later this week.
", "votes": 2, "replies": [] }, { "id": 1648440, "author_name": "Jie Wu", "author_url": "", "post_date": "2022-01-13T12:34:34.240000", "content": "Fabulous!!!
", "votes": 0, "replies": [] } ] }, { "id": 1647111, "author_name": "Jie Wu", "author_url": "", "post_date": "2022-01-12T10:16:35.810000", "content": "@andrewscherer can you help to clarify this?
", "votes": 1, "replies": [] }, { "id": 1647964, "author_name": "Susan Paige", "author_url": "", "post_date": "2022-01-13T03:00:30.407000", "content": "Nice question! Have the same question before, thanks!
", "votes": 0, "replies": [] }, { "id": 1647466, "author_name": "", "author_url": "", "post_date": "2022-01-12T16:19:54.533000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "1647077": "Does anyone can explain how the time series API of this competition will work? I just joined this competition, but after some readings I am still not clear how does this work. I really appreciate if any of you can help me clear this?\n\n1. In the [data description ](https://www.kaggle.com/c/g-research-crypto-forecasting/data)says\n\"supplemental_train.csv - After the submission period is over this file's data will be replaced with cryptoasset prices from the submission period. In the Evaluation phase, the train, train supplement, and test set will be contiguous in time, apart from any missing data. The current copy, which is just filled approximately the right amount of data from train.csv is provided as a placeholder.\" What does this mean?\nDoes this mean in the evaluation phase, train and train supplement dataset will be updated to the latest time? For instance, let's assume first evaluation will happen at the end of February, second evaluation will happen at the end of March. Does this mean train.csv will have 2018-01-01 to 2021-09-21 + 2021-09-22 to 2022-01-31 data, and train supplemental_train.csv will have 2021-09-22 to 2022-01-31 data when 1st evaluation happen? Then train.csv will be updated to have data 2018-01-01 to 2022-02-28, and supplemental_train.csv will have data 2022 February data when 2nd evaluation happen?\nOr this means something else?\n1. While we are calling the inference API, next(iter_test), if data of every previous iteration will be automatically appended to train.csv and supplemental_train.csv? Can we use previous iteration as current iteration input features? Or we must (can we) create a template dataframe to save previous iterations data, if we want to use them?", "1647442": "You would have to create a template data frame to save pervious iterations. I would use a numpy array though because it should be faster. Just append to the end of the array the new data and remove the oldest timestep from the array that was needed for your greatest lagged feature in the previous timestep.\n\nI think train will stay the same and the supplemental train will bridge the gap between train and test, but it would be nice to get confirmation.", "1647111": "@andrewscherer can you help to clarify this?", "1647964": "Nice question! Have the same question before, thanks!", "1647466": "" } }