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
18.01k stars 4.48k forks source link

Prophet Training Speed: Intel vs Apple M1/M2 #2572

Open ppiotrow opened 2 months ago

ppiotrow commented 2 months ago

I've noticed a significant speed gap in training time series data with Prophet: my code is roughly 5 times faster on Apple M1/M2 CPUs than on modern Intel processors. This contrast was consistent across four different setups running similar Python versions (3.11 and 3.12), with CPU architecture being the key variable.

My main question is whether this speed difference between Apple and Intel CPUs is a recognized issue? If it's not expected, could you please advise me on which dependencies I should investigate or possibly recompile to understand why Prophet runs slower on Intel CPUs compared to Apple ones?

aakashnand commented 1 week ago

@ppiotrow have you found any new information regarding this? I recently observed this and it seems apple silicon is definitely faster than Intel and lot of it has to do with underlying CPU architecture I guess.