4 ms·

We've enjoyed using Prophet for forecasting in our work with electric utilities. It's pretty fire-and-forget. We wish it had an option for fast retraining of th

We've enjoyed using Prophet for forecasting in our work with electric utilities. It's pretty fire-and-forget. We wish it had an option for fast retraining of the model when we get new hourly data, though. That's pushing us towards other options in scipy.

wenc 8y ago
I looked at Prophet a few months ago because we needed a fire-and-forget library similar to 'auto.arima' (R "forecast" package) for Python, but no good candidates existed. However I found Prophet to be computationally a little heavier than auto.arima because it uses "stan" (Bayesian) underneath, which in turn uses an MCMC type approach and has quite a few dependencies. We needed fast model retraining as well, and at the time, it didn't seem like that was something it excelled at. (might have changed, I'm not sure) I ended up putting together a simple ensemble forecast model class with "statsmodels" which automatically selected/averaged the best models over a collection of model types via heuristics and cross-validation. It works ok, but I'm still waiting for someone to port the R auto.arima over to Python. (I tried rpy, which in theory should have worked, but I struggled with the impedance mismatch)
- Tarq0n 8y ago
  You only need to do MCMC if you want simulated confidence intervals, it doesn't do anything for model estimation.
  - wenc 8y ago
    You’re right. It uses Stan to do MAP for model estimation and HMC for (optional) computation of confidence intervals. [1] Thanks for the correction. I need to look at Prophet again. Maybe it will work for fast model retraining. I had trouble the last time I tried it. [1] https://research.fb.com/prophet-forecasting-at-scale/ https://research.fb.com/prophet-forecasting-at-scale/
- venuur 8y ago
  I found this port of auto-arima for Python. I haven’t used it in production, but it was easy to test on some demo data. https://pypi.org/project/pyramid-arima/ https://pypi.org/project/pyramid-arima/
  - wenc 8y ago
    Thanks, I'll take a look.

wenc 8y ago

I looked at Prophet a few months ago because we needed a fire-and-forget library similar to 'auto.arima' (R "forecast" package) for Python, but no good candidates existed. However I found Prophet to be computationally a little heavier than auto.arima because it uses "stan" (Bayesian) underneath, which in turn uses an MCMC type approach and has quite a few dependencies. We needed fast model retraining as well, and at the time, it didn't seem like that was something it excelled at. (might have changed, I'm not sure) I ended up putting together a simple ensemble forecast model class with "statsmodels" which automatically selected/averaged the best models over a collection of model types via heuristics and cross-validation. It works ok, but I'm still waiting for someone to port the R auto.arima over to Python. (I tried rpy, which in theory should have worked, but I struggled with the impedance mismatch)

Tarq0n 8y ago
You only need to do MCMC if you want simulated confidence intervals, it doesn't do anything for model estimation.
- wenc 8y ago
  You’re right. It uses Stan to do MAP for model estimation and HMC for (optional) computation of confidence intervals. [1] Thanks for the correction. I need to look at Prophet again. Maybe it will work for fast model retraining. I had trouble the last time I tried it. [1] https://research.fb.com/prophet-forecasting-at-scale/ https://research.fb.com/prophet-forecasting-at-scale/
venuur 8y ago
I found this port of auto-arima for Python. I haven’t used it in production, but it was easy to test on some demo data. https://pypi.org/project/pyramid-arima/ https://pypi.org/project/pyramid-arima/
- wenc 8y ago
  Thanks, I'll take a look.