melting of time-series

[michaelaye]

I have a standard DataFrame indexed by

<class 'pandas.tseries.index.DatetimeIndex'>
[2009-07-08 07:00:00.134000, ..., 2009-07-08 10:00:01.297000]
Length: 84384, Freq: None, Timezone: None

and 189 columns named like this:

Index([u'a3_01', u'a3_02', u'a3_03', u'a3_04', u'a3_05', u'a3_06', u'a3_07', u'a3_08', u'a3_09', u'a3_10', u'a3_11', u'a3_12', u'a3_13', u'a3_14', u'a3_15', u'a3_16', u'a3_17', u'a3_18', u'a3_19', u'a3_20', u'a3_21', u'a4_01', u'a4_02',...], dtype='object')

Why is it that the following melt operation creates a KeyError:

pd.melt(rad, id_vars=[rad.index])
KeyError: "['2009-07-08T00:00:00.134000000-0700' '2009-07-08T00:00:00.262000000-0700'\n '2009-07-08T00:00:00.390000000-0700' ...,\n '2009-07-08T03:00:01.041000000-0700' '2009-07-08T03:00:01.169000000-0700'\n '2009-07-08T03:00:01.297000000-0700'] not in index"

but if I reset_index() first it works:

pd.melt(rad.reset_index(), id_vars=['index'])

I am aware that the time-index will become non-unique after melting, but shouldn't this actually work? Or maybe it is a feature request that the id_vars accepts a DatetimeIndex as input?

[jreback]

this is the same reason as #6582, so will link to that one