search

decisionintelligence / TFB

[PVLDB 2024 Best Paper Nomination] TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods
https://www.vldb.org/pvldb/vol17/p2363-hu.pdf
MIT License
321 stars 36 forks source link

How to evaluate over multiple csv files #32

Open tionry opened 4 weeks ago

tionry commented 4 weeks ago

Hi, Thanks for your work! I tried to use your code for our dataset model training/validating. I meet several issues:

  1. how to train over multiple csv files, I see that the sh command --data-name-list gives file_name, how to write the command if I want to train over multiple files, such as AQShunyi-1.csv, AQShunyi-2.csv simutenously?
  2. How may I train my prediction model as the hyperparameters are predefined in the command line, or I must train the mode lat somewhere else?
  3. Can the framework deal with data of multiple features, e.g., if I want to forecast the nationwide air quality, assume I get data from different regions, the data may look like [region, timestamp, quality], it is time series with 'region' feature, so the question is how to train and also validate the prediction model, or it is infeasible in the context of time series forecasting. Looking forward to your reply, thanks so much!
qiu69 commented 4 weeks ago

Thank you for your attention and questions! Regarding the first question, we currently do not support it, but we had plans to implement this function (global forecasting). I didn't understand the last two questions! Can we exchange WeChat for easier communication? My WeChat: qxf780310. Look forward to hearing from you.