For predicting a value( of responder_6) , can we use past timestamp values ? can we use fututre time stamp values ? or we need to predict from that particular row data? In test data will it be provided the some future time stamp values( as it is given in sample test file here). I mean to say on a particular date at 10:30 AM can I use feature data of 10:31 AM
", "messages": [ { "id": "3032571", "postDate": "10/31/2024 05:20:59", "content": "For predicting a value( of responder_6) , can we use past timestamp values ? can we use fututre time stamp values ? or we need to predict from that particular row data? In test data will it be provided the some future time stamp values( as it is given in sample test file here). I mean to say on a particular date at 10:30 AM can I use feature data of 10:31 AM
", "rawMarkdown": "For predicting a value( of responder_6) , can we use past timestamp values ? can we use fututre time stamp values ? or we need to predict from that particular row data? In test data will it be provided the some future time stamp values( as it is given in sample test file here). I mean to say on a particular date at 10:30 AM can I use feature data of 10:31 AM", "votes": null }, { "id": "3032587", "postDate": "10/31/2024 05:58:58", "content": "you typically predict the target (responder_6) based on information available up to the current timestamp of the data point you are predicting. Using future timestamp values for prediction is usually lead to data leakage.
\nin this competition the submission is made by the evaluation API through a predict function which ensures that models do not peek forward in time. for each call of the API makes to the predict function it passes only 39 row from the test dataset.
To be precise, there’s a possibility that not all symbol IDs are present, and new ones might be added, so there may be times when it’s not exactly 39 rows!
", "rawMarkdown": "To be precise, there’s a possibility that not all symbol IDs are present, and new ones might be added, so there may be times when it’s not exactly 39 rows!", "votes": null }, { "id": "3032630", "postDate": "10/31/2024 07:00:15", "content": "@chumajin what about the time limit between the calls ? is it one or 10 minutes?
", "rawMarkdown": "chumajin what about the time limit between the calls ? is it one or 10 minutes?", "votes": null }, { "id": "3032679", "postDate": "10/31/2024 08:10:15", "content": "@younesbenalia So far, I understand 1 minutes. I don’t think it’s been updated since here! https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/discussion/542022#3027305
", "rawMarkdown": "younesbenalia So far, I understand 1 minutes. I don’t think it’s been updated since here! https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/discussion/542022#3027305", "votes": null } ], "comments": [ { "id": 3032587, "author_name": "younesbenalia", "author_url": "", "post_date": "10/31/2024 05:58:58", "content": "you typically predict the target (responder_6) based on information available up to the current timestamp of the data point you are predicting. Using future timestamp values for prediction is usually lead to data leakage.
\nin this competition the submission is made by the evaluation API through a predict function which ensures that models do not peek forward in time. for each call of the API makes to the predict function it passes only 39 row from the test dataset.
To be precise, there’s a possibility that not all symbol IDs are present, and new ones might be added, so there may be times when it’s not exactly 39 rows!
", "votes": null, "replies": [ { "id": 3032630, "author_name": "younesbenalia", "author_url": "", "post_date": "10/31/2024 07:00:15", "content": "@chumajin what about the time limit between the calls ? is it one or 10 minutes?
", "votes": null, "replies": [ { "id": 3032679, "author_name": "chumajin", "author_url": "", "post_date": "10/31/2024 08:10:15", "content": "@younesbenalia So far, I understand 1 minutes. I don’t think it’s been updated since here! https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/discussion/542022#3027305
", "votes": null, "replies": [] } ] } ] } ] } ], "raw_markdown_by_id": { "3032571": "For predicting a value( of responder_6) , can we use past timestamp values ? can we use fututre time stamp values ? or we need to predict from that particular row data? In test data will it be provided the some future time stamp values( as it is given in sample test file here). I mean to say on a particular date at 10:30 AM can I use feature data of 10:31 AM", "3032587": "you typically predict the target (responder_6) based on information available up to the current timestamp of the data point you are predicting. Using future timestamp values for prediction is usually lead to data leakage.\nin this competition the submission is made by the evaluation API through a predict function which ensures that models do not peek forward in time. for each call of the API makes to the predict function it passes only 39 row from the test dataset.", "3032606": "To be precise, there’s a possibility that not all symbol IDs are present, and new ones might be added, so there may be times when it’s not exactly 39 rows!", "3032630": "chumajin what about the time limit between the calls ? is it one or 10 minutes?", "3032679": "younesbenalia So far, I understand 1 minutes. I don’t think it’s been updated since here! https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/discussion/542022#3027305" }, "source": "meta" }