issue_comments

The reason for not using numeric only for max/min is that they should be well defined even for strings and dates -- unlike aggregations like mean, sum, variance (actually, in principle most should be able to work OK for dates but the numpy codes has some bugs we would need to work around).

.

The bytes handling in to_datetime is arguably a pandas bug. Alternatively we could decode character arrays from netcdf as unicode instead of bytes, but I'm not sure that's unambiguously the right thing to do. This is a place where the legacy Python 2 distinction of strings/unicode is a closer match for netcdf (and scientific file formats more generally) than the Python 3 behavior.

On Fri, Jul 3, 2015 at 4:40 PM, Will Holmgren notifications@github.com wrote:

Thanks for the tips. This may be Python 3 specific, but I needed to convert to convert to strings first

python times_strings = list(map(lambda x: x.decode('utf-8'), ds['Times'].values)) ds['Times'] = ('Time', pd.to_datetime(times_strings, format='%Y-%m-%d_%H:%M:%S'))

Is there a reason why you don't use numeric_only=True for the min and max functions? I was just recommending more consistency across the min/max/mean/std/etc functions. Might also be good to be explicit about that in the doc strings.

Reply to this email directly or view it on GitHub: https://github.com/xray/xray/issues/453#issuecomment-118437972

Thanks for the tips. This may be Python 3 specific, but I needed to convert to convert to strings first

python times_strings = list(map(lambda x: x.decode('utf-8'), ds['Times'].values)) ds['Times'] = ('Time', pd.to_datetime(times_strings, format='%Y-%m-%d_%H:%M:%S'))

Is there a reason why you don't use numeric_only=True for the min and max functions? I was just recommending more consistency across the min/max/mean/std/etc functions. Might also be good to be explicit about that in the doc strings.

I agree, it's not friendly to give an error message here.

Something you could do about this -- you probably want to convert your times into the numpy datetime64 type. That makes your operations much more efficient, and would make .min() work:

ds['Times'] = ('Time', pd.to_datetime(ds['Times'], format='%Y-%m-%d_%H:%M:%S'))

You also probably want to make this Times variable the dimension variable -- that will let you select times with datetime objects or strings instead of integers: ds.swap_dims({"Times": "Time"}).

Or in one line:

ds = ds.assign(Time=pd.to_datetime(ds['Times'], format='%Y-%m-%d_%H:%M:%S')).drop('Times')

Something xray do could about this -- we could convert string/unicode arrays into the numpy object dtype prior to attempting operations like min/argmin. That way, the min operation would still work.