Hi! This week we released the fastest autoarima implementation ever for Python. It is a mirror implementation of Hyndman’s auto.arima (R) and it is optimized using Numba. 
In the experiments we performed, our implementation is a lot faster than the current alternative (pmdarima) and than the R implementation. Also, it is more accurate than those implementations and than FB’s prophet. 
For this use case, it is faster to compile the code in Numba rather than calling C functions, which is very interesting. 
Here’s the link to the project: GitHub - Nixtla/statsforecast: Lightning ⚡️ fast forecasting with statistical and econometric models. If you like it, please give us a star. 