AssertionError with df.resample(how="median")

[eloraburns]

I've reproduced something using how="median", perhaps #1648. It seems to hit when there are discontinuities in the resampling (i.e. minutes with no records when downsampling).

Both pandas 0.8.1 and 0.8.2.dev-f5a74d4 don't like it:

import pandas as pd
from datetime import datetime
df = pd.DataFrame([1, 2], index=[datetime(2012,1,1,0,0,0), datetime(2012,1,1,0,5,0)])
df.resample("T", how="median")
# Throws AssertionError

[eloraburns]

how="mean" works:

In [5]: df.resample("T", how="mean")
Out[5]: 
                      0
2012-01-01 00:00:00   1
2012-01-01 00:01:00 NaN
2012-01-01 00:02:00 NaN
2012-01-01 00:03:00 NaN
2012-01-01 00:04:00 NaN
2012-01-01 00:05:00   2

I have no idea what numpy is up to, though:

In [6]: import numpy as np

In [7]: df.resample("T", how=lambda x: np.percentile(x, 50))
Out[7]: 
Empty DataFrame
Columns: array([], dtype=int64)
Index: array([], dtype=object)

I can't seem to make it misbehave here, but with some Actual Data with data absent for several of the resampled intervals, I can get numpy to explode with the same stack trace as:

np.percentile([], 50)
# ...
# ValueError: operands could not be broadcast together with shapes (0) (2) 

I can certainly work around that in the lambda by short-circuiting on an empty list-like thing being passed in, but I don't think I should have to.

For this particular case, I can just do this with expected results:

In [8]: df.resample("T", how=np.median)
Out[8]: 
                      0
2012-01-01 00:00:00   1
2012-01-01 00:01:00 NaN
2012-01-01 00:02:00 NaN
2012-01-01 00:03:00 NaN
2012-01-01 00:04:00 NaN
2012-01-01 00:05:00   2

I can't fathom why np.median(…) is different from np.percentile(…, 50). Maybe that's just a numpy thing? :)

Thanks!

[wesm]

thanks for the test case. fixed the underlying cause

[eloraburns]

Awesome, glad it helped. Thanks for fixing it! :) When trying to learn numpy/pandas concepts, sometimes it's hard to tell user error from a True bug. :)