Open mbauer288 opened 1 year ago
really good question. I'd assume we can safely remove the little-endianness enforcement. It should not make a difference. I think we really just wanted to test if those are nanoseconds; not the byte-order. Most lazy approach to clear this up would be adding tests.
Also, am I correct that STATEPandas does not want/support datatimes with timezone (UTC) encoding (i.e., not '<M8[ns, UTC]', 'datetime64[ns UTC]')?
Just not yet! But we could easily tweak that. I.e. just enforcing conversion to datetime64[ns UTC].
To my understanding the numpy dtype
<M8[ns]
is the byte-order equivalent todatetime64[ns]
on little-endian hardware, whereas on big-endian machinesdatetime64[ns] == >M8[ns]
.So we are enforcing little-endianess of our dataframes in
tivs_from_timeseries()
?Not that I can even think of a big-endian machine nowadays, but perhaps we should add a note (or issue a warning) to this?
I say this because by default panadas and datetime make
datetime64[ns]
objects which defaults to that machine's byte-order.Also, am I correct that STATEPandas does not want/support datatimes with timezone (UTC) encoding (i.e., not '<M8[ns, UTC]', 'datetime64[ns UTC]')?