What happened + What you expected to happen

With a time series with half-hourly granularity (48 samples per day), I am trying to run an ARIMA model but am running into limitations on the order of usable terms.

with (among other variations that raised the same error)

ARIMA(
    order=(0, 0, 0),
    season_length=48,
    seasonal_order=(1, 1, 7),
)

When trying to fit, I get the error

❱   52 │   _arima.getQ0(phi, theta, res) 
...
ValueError: cannot create std::vector larger than max_size()

I tried to dig into the arima.cpp code to understand exactly where it's happening but that's outside of my wheelhouse. max_size() is generally a huge vector though so not sure what may be happening here.

I understand a half-hourly granularity is probably far beyond what ARIMA was meant to do, but I was surprised to run into such specific limits from the underlying code.

Versions / Dependencies