Pandas MultiIndex Rolling Group by

Question

Pandas has some very strange group by behavior with Multi_indexing. I don't understand why one of these sites is not being calculated but when called alone works perfectly fine.

Consider the following code

all_data['rolling_avg_supply_temp'] = all_data.groupby(level=0).rolling('1h',on='ts')['supplytemp'].mean()
all_data['rolling_avg_supply_temp'].groupby(level=0).describe()

site	count	mean	std	min	25%	50%	75%	max
1A1	22130.0	21.698209	0.588990	14.611525	21.521668	21.837078	22.001139	23.345806
2B5	22533.0	21.952604	0.535339	18.900000	21.828639	21.976729	22.110053	25.985087
3B8	9515.0	22.060124	0.427317	19.500000	21.822788	22.004277	22.226354	24.400361
3B9	19234.0	21.810098	0.686575	19.530944	21.657984	21.966606	22.142692	23.987056
VC1A1	0.0	NaN	NaN	NaN	NaN	NaN	NaN	NaN

The VC1A site just doesn't get a rolling average applied in the function call.

However consider this function call ON THE SAME dataset

all_data.loc["VC1A1"].rolling('1h',on='ts')['supplytemp'].mean().describe()

Statistic	Value
count	14400.000000
mean	21.492677
std	0.406032
min	20.152830
25%	21.345195
50%	21.430232
75%	21.755384
max	22.265454

There is no mixing of datatypes for reference here is the info on the dataset

all_data.info()

Metadata	Value
Total Entries	5,456,561
Memory Usage	654.1+ MB
MultiIndex From	('1A1', Timestamp('2024-04-07 17:00:00.431000+0000', tz='UTC'))
MultiIndex To	('VC1A1', Timestamp('2024-04-17 09:29:28.405000+0000', tz='UTC'))
Total Columns	11

Column Index	Column Name	Dtype
0	sn	object
1	ts	datetime64[ns, UTC]
2	maxcellt	float64
3	mincellt	float64
4	supplytemp	float64
5	hvac_status_c1runstat	float64
6	hvac_status_c2runstat	float64
7	hvac_status_bypassdamp	float64
8	hvac_sensor_heatcurrent	float64
9	hvac_sensor_avgspacetemp	float64
10	timediff	timedelta64[ns]