problems with reindexing dataframes: Reindexing only valid with uniquely valued Index objects

Question

I am having a real strange behaviour when trying to reindex a dataframe in pandas. My version of Pandas is 0.10.0 and I use Python 2.7. Basically, when I load a dataframe:

eurusd = pd.DataFrame.load('EUR_USD_30Min.df').drop_duplicates().dropna()

eurusd

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 119710 entries, 2003-02-02 17:30:00 to 2012-12-28 17:00:00
Data columns:
open     119710  non-null values
high     119710  non-null values
low      119710  non-null values
close    119710  non-null values
dtypes: float64(4)

and then I try to reindex inside a larger date range:

newindex  = pd.DateRange(datetime.datetime(2002,1,1), datetime.datetime(2012,12,31), offset=pd.datetools.Minute(30))

newindex

<class 'pandas.tseries.index.DatetimeIndex'>
[2002-01-01 00:00:00, ..., 2012-12-31 00:00:00]
Length: 192817, Freq: 30T, Timezone: None

I get strange behaviour when trying to reindex the dataframe. If I reindex one larger part of the dataset I get this error:

eurusd[29558:29560].reindex(index=newindex)

Exception: Reindexing only valid with uniquely valued Index objects

But, if I do the same for two subsets of the data above, I don't get the error:

Here's the first subset, with no problems,

eurusd[29558:29559].reindex(index=newindex)

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 192817 entries, 2002-01-01 00:00:00 to 2012-12-31 00:00:00
Freq: 30T
Data columns:
open     1  non-null values
high     1  non-null values
low      1  non-null values
close    1  non-null values
dtypes: float64(4)

and here's the second subset, still no problems,

eurusd[29559:29560].reindex(index=newindex)

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 192817 entries, 2002-01-01 00:00:00 to 2012-12-31 00:00:00
Freq: 30T
Data columns:
open     1  non-null values
high     1  non-null values
low      1  non-null values
close    1  non-null values
dtypes: float64(4)

I am really going crazy about this, and cannot understand the reasons of this. It seems like the dataframe is 'clean' from duplicates, and duplicated indexes.... I can provide the pickle file for the dataframe if you want.

Answer 1

You could groupby the index and take the first entry (see docs):

df.groupby(level=0).first()

Example:

In [1]: df = pd.DataFrame([[1], [2]], index=[1, 1])

In [2]: df
Out[2]: 
   0
1  1
1  2

In [3]: df.groupby(level=0).first()
Out[3]: 
   0
1  1