search

facebook / prophet

Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.
https://facebook.github.io/prophet
MIT License
18.47k stars 4.53k forks source link

Using time in HMS as a regressor #1985

Open MSasta opened 3 years ago

MSasta commented 3 years ago

For my dataset, I am trying to use the time an event occurred as a regressor term. Lets say I have a dataset of people who rent bikes. I'm trying to forecast how many people will rent a bike.

If people rent bikes at different times during the day, can that be used as a regressor? Let's say bikes can only be rented at specific times during the day, e.g. 8:15am, 10:30am. So using this as a regressor term for number of people who rent a bike in a day

If possible, how would you treat such a term, I have been trying but i've been getting the issue You have provided a name that is not syntactically valid in R

Thank you!

tcuongd commented 3 years ago

Hey there, any patterns relating to time-of-day are probably better modelled using the daily_seasonality feature rather than additional regressors. Here is an example from the docs: https://facebook.github.io/prophet/docs/non-daily_data.html#sub-daily-data

If you have aggregated the bike rental data to be at a sub-daily granularity (e.g. 15-minute intervals), then the time-of-day seasonality should be fit automatically.

MSasta commented 3 years ago

Ah okay thank you for your reply, I don't exactly have symmetric intervals in my actual dataset is why I was wondering if it would be possible to use xreg instead of daily seasonality. Maybe I have to change my approach.

tcuongd commented 3 years ago

How did you go with this @MSasta? Were you able to re-index the time series data and use daily_seasonality?