search

tsinghua-fib-lab / UniST

Official implementation for "UniST: A Prompt-Empowered Universal Model for Urban Spatio-Temporal Prediction" (KDD 2024)
55 stars 7 forks source link

Temporal Duration and Temporal interval #6

Open kasanatte opened 4 days ago

kasanatte commented 4 days ago

Hi, thank you for your nice work and for sharing the code.

I want to know how to deal with the data of different time intervals and different dates. the timestamps in your data only include the day of the week and time of day. But the data has different dates and different time intervals.

The range of timestamps for TrafficNJ and BikeNYC-1 in dataloader seems to be [0-6,0-47]? When his_len=6, TrafficCS predicts 6 frames(5 minute intervals) and TDrive predicts 6 frames(1 hour intervals)？ And why doesn't the timestamp include the year, month, and day?

Look forward to your reply. Thanks in advance!

YuanYuan98 commented 3 days ago

Your understanding is correct: the prediction frames indeed vary across different datasets.

In our approach, we focus solely on the time of day and the day of the week. Given that our dataset spans only a few months, we intentionally omit the year, month, and exact date. This simplification allows us to better capture daily and weekly patterns, which are often more significant for the predictions.