pandas 0.7.3 documentation

Merge, join, and concatenate

pandas provides various facilities for easily combining together Series, DataFrame, and Panel objects with various kinds of set logic for the indexes and relational algebra functionality in the case of join / merge-type operations.

Concatenating objects

The concat function (in the main pandas namespace) does all of the heavy lifting of performing concatenation operations along an axis while performing optional set logic (union or intersection) of the indexes (if any) on the other axes. Note that I say “if any” because there is only a single possible axis of concatenation for Series.

Before diving into all of the details of concat and what it can do, here is a simple example:

In [636]: df = DataFrame(np.random.randn(10, 4))

In [637]: df
Out[637]: 
          0         1         2         3
0  0.469112 -0.282863 -1.509059 -1.135632
1  1.212112 -0.173215  0.119209 -1.044236
2 -0.861849 -2.104569 -0.494929  1.071804
3  0.721555 -0.706771 -1.039575  0.271860
4 -0.424972  0.567020  0.276232 -1.087401
5 -0.673690  0.113648 -1.478427  0.524988
6  0.404705  0.577046 -1.715002 -1.039268
7 -0.370647 -1.157892 -1.344312  0.844885
8  1.075770 -0.109050  1.643563 -1.469388
9  0.357021 -0.674600 -1.776904 -0.968914

# break it into pieces
In [638]: pieces = [df[:3], df[3:7], df[7:]]

In [639]: concatenated = concat(pieces)

In [640]: concatenated
Out[640]: 
          0         1         2         3
0  0.469112 -0.282863 -1.509059 -1.135632
1  1.212112 -0.173215  0.119209 -1.044236
2 -0.861849 -2.104569 -0.494929  1.071804
3  0.721555 -0.706771 -1.039575  0.271860
4 -0.424972  0.567020  0.276232 -1.087401
5 -0.673690  0.113648 -1.478427  0.524988
6  0.404705  0.577046 -1.715002 -1.039268
7 -0.370647 -1.157892 -1.344312  0.844885
8  1.075770 -0.109050  1.643563 -1.469388
9  0.357021 -0.674600 -1.776904 -0.968914

Like its sibling function on ndarrays, numpy.concatenate, pandas.concat takes a list or dict of homogeneously-typed objects and concatenates them with some configurable handling of “what to do with the other axes”:

concat(objs, axis=0, join='outer', join_axes=None, ignore_index=False,
       keys=None, levels=None, names=None, verify_integrity=False)
  • objs: list or dict of Series, DataFrame, or Panel objects. If a dict is passed, the sorted keys will be used as the keys argument, unless it is passed, in which case the values will be selected (see below)
  • axis: {0, 1, ...}, default 0. The axis to concatenate along
  • join: {‘inner’, ‘outer’}, default ‘outer’. How to handle indexes on other axis(es). Outer for union and inner for intersection
  • join_axes: list of Index objects. Specific indexes to use for the other n - 1 axes instead of performing inner/outer set logic
  • keys: sequence, default None. Construct hierarchical index using the passed keys as the outermost level If multiple levels passed, should contain tuples.
  • levels : list of sequences, default None. If keys passed, specific levels to use for the resulting MultiIndex. Otherwise they will be inferred from the keys
  • names: list, default None. Names for the levels in the resulting hierarchical index
  • verify_integrity: boolean, default False. Check whether the new concatenated axis contains duplicates. This can be very expensive relative to the actual data concatenation
  • ignore_index : boolean, default False. If True, do not use the index values on the concatenation axis. The resulting axis will be labeled 0, ..., n - 1. This is useful if you are concatenating objects where the concatenation axis does not have meaningful indexing information.

Without a little bit of context and example many of these arguments don’t make much sense. Let’s take the above example. Suppose we wanted to associate specific keys with each of the pieces of the chopped up DataFrame. We can do this using the keys argument:

In [641]: concatenated = concat(pieces, keys=['first', 'second', 'third'])

In [642]: concatenated
Out[642]: 
                 0         1         2         3
first  0  0.469112 -0.282863 -1.509059 -1.135632
       1  1.212112 -0.173215  0.119209 -1.044236
       2 -0.861849 -2.104569 -0.494929  1.071804
second 3  0.721555 -0.706771 -1.039575  0.271860
       4 -0.424972  0.567020  0.276232 -1.087401
       5 -0.673690  0.113648 -1.478427  0.524988
       6  0.404705  0.577046 -1.715002 -1.039268
third  7 -0.370647 -1.157892 -1.344312  0.844885
       8  1.075770 -0.109050  1.643563 -1.469388
       9  0.357021 -0.674600 -1.776904 -0.968914

As you can see (if you’ve read the rest of the documentation), the resulting object’s index has a hierarchical index. This means that we can now do stuff like select out each chunk by key:

In [643]: concatenated.ix['second']
Out[643]: 
          0         1         2         3
3  0.721555 -0.706771 -1.039575  0.271860
4 -0.424972  0.567020  0.276232 -1.087401
5 -0.673690  0.113648 -1.478427  0.524988
6  0.404705  0.577046 -1.715002 -1.039268

It’s not a stretch to see how this can be very useful. More detail on this functionality below.

Set logic on the other axes

When gluing together multiple DataFrames (or Panels or...), for example, you have a choice of how to handle the other axes (other than the one being concatenated). This can be done in three ways:

  • Take the (sorted) union of them all, join='outer'. This is the default option as it results in zero information loss.
  • Take the intersection, join='inner'.
  • Use a specific index (in the case of DataFrame) or indexes (in the case of Panel or future higher dimensional objects), i.e. the join_axes argument

Here is a example of each of these methods. First, the default join='outer' behavior:

In [644]: from pandas.util.testing import rands

In [645]: df = DataFrame(np.random.randn(10, 4), columns=['a', 'b', 'c', 'd'],
   .....:                index=[rands(5) for _ in xrange(10)])

In [646]: df
Out[646]: 
              a         b         c         d
RQlXD -1.294524  0.413738  0.276662 -0.472035
p9y7F -0.013960 -0.362543 -0.006154 -0.923061
XbyT9  0.895717  0.805244 -1.206412  2.565646
8PnTm  1.431256  1.340309 -1.170299 -0.226169
JX6kv  0.410835  0.813850  0.132003 -0.827317
15BgG -0.076467 -1.187678  1.130127 -1.436737
UhTku -1.413681  1.607920  1.024180  0.569605
rJ9U4  0.875906 -2.211372  0.974466 -2.006747
TMmI9 -0.410001 -0.078638  0.545952 -1.219217
0ij9W -1.226825  0.769804 -1.281247 -0.727707

In [647]: concat([df.ix[:7, ['a', 'b']], df.ix[2:-2, ['c']],
   .....:         df.ix[-7:, ['d']]], axis=1)
Out[647]: 
              a         b         c         d
0ij9W       NaN       NaN       NaN -0.727707
15BgG -0.076467 -1.187678  1.130127 -1.436737
8PnTm  1.431256  1.340309 -1.170299 -0.226169
JX6kv  0.410835  0.813850  0.132003 -0.827317
RQlXD -1.294524  0.413738       NaN       NaN
TMmI9       NaN       NaN       NaN -1.219217
UhTku -1.413681  1.607920  1.024180  0.569605
XbyT9  0.895717  0.805244 -1.206412       NaN
p9y7F -0.013960 -0.362543       NaN       NaN
rJ9U4       NaN       NaN  0.974466 -2.006747

Note that the row indexes have been unioned and sorted. Here is the same thing with join='inner':

In [648]: concat([df.ix[:7, ['a', 'b']], df.ix[2:-2, ['c']],
   .....:         df.ix[-7:, ['d']]], axis=1, join='inner')
Out[648]: 
              a         b         c         d
8PnTm  1.431256  1.340309 -1.170299 -0.226169
JX6kv  0.410835  0.813850  0.132003 -0.827317
15BgG -0.076467 -1.187678  1.130127 -1.436737
UhTku -1.413681  1.607920  1.024180  0.569605

Lastly, suppose we just wanted to reuse the exact index from the original DataFrame:

In [649]: concat([df.ix[:7, ['a', 'b']], df.ix[2:-2, ['c']],
   .....:         df.ix[-7:, ['d']]], axis=1, join_axes=[df.index])
Out[649]: 
              a         b         c         d
RQlXD -1.294524  0.413738       NaN       NaN
p9y7F -0.013960 -0.362543       NaN       NaN
XbyT9  0.895717  0.805244 -1.206412       NaN
8PnTm  1.431256  1.340309 -1.170299 -0.226169
JX6kv  0.410835  0.813850  0.132003 -0.827317
15BgG -0.076467 -1.187678  1.130127 -1.436737
UhTku -1.413681  1.607920  1.024180  0.569605
rJ9U4       NaN       NaN  0.974466 -2.006747
TMmI9       NaN       NaN       NaN -1.219217
0ij9W       NaN       NaN       NaN -0.727707

Concatenating using append

A useful shortcut to concat are the append instance methods on Series and DataFrame. These methods actually predated concat. They concatenate along axis=0, namely the index:

In [650]: s = Series(randn(10), index=np.arange(10))

In [651]: s1 = s[:5] # note we're slicing with labels here, so 5 is included

In [652]: s2 = s[6:]

In [653]: s1.append(s2)
Out[653]: 
0   -0.121306
1   -0.097883
2    0.695775
3    0.341734
4    0.959726
6   -0.619976
7    0.149748
8   -0.732339
9    0.687738

In the case of DataFrame, the indexes must be disjoint but the columns do not need to be:

In [654]: df = DataFrame(randn(6, 4), index=DateRange('1/1/2000', periods=6),
   .....:                columns=['A', 'B', 'C', 'D'])

In [655]: df1 = df.ix[:3]

In [656]: df2 = df.ix[3:, :3]

In [657]: df1
Out[657]: 
                   A         B         C         D
2000-01-03  0.176444  0.403310 -0.154951  0.301624
2000-01-04 -2.179861 -1.369849 -0.954208  1.462696
2000-01-05 -1.743161 -0.826591 -0.345352  1.314232

In [658]: df2
Out[658]: 
                   A         B         C
2000-01-06  0.690579  0.995761  2.396780
2000-01-07  3.357427 -0.317441 -1.236269
2000-01-10 -0.487602 -0.082240 -2.182937

In [659]: df1.append(df2)
Out[659]: 
                   A         B         C         D
2000-01-03  0.176444  0.403310 -0.154951  0.301624
2000-01-04 -2.179861 -1.369849 -0.954208  1.462696
2000-01-05 -1.743161 -0.826591 -0.345352  1.314232
2000-01-06  0.690579  0.995761  2.396780       NaN
2000-01-07  3.357427 -0.317441 -1.236269       NaN
2000-01-10 -0.487602 -0.082240 -2.182937       NaN

append may take multiple objects to concatenate:

In [660]: df1 = df.ix[:2]

In [661]: df2 = df.ix[2:4]

In [662]: df3 = df.ix[4:]

In [663]: df1.append([df2,df3])
Out[663]: 
                   A         B         C         D
2000-01-03  0.176444  0.403310 -0.154951  0.301624
2000-01-04 -2.179861 -1.369849 -0.954208  1.462696
2000-01-05 -1.743161 -0.826591 -0.345352  1.314232
2000-01-06  0.690579  0.995761  2.396780  0.014871
2000-01-07  3.357427 -0.317441 -1.236269  0.896171
2000-01-10 -0.487602 -0.082240 -2.182937  0.380396

Note

Unlike list.append method, which appends to the original list and returns nothing, append here does not modify df1 and returns its copy with df2 appended.

Ignoring indexes on the concatenation axis

For DataFrames which don’t have a meaningful index, you may wish to append them and ignore the fact that they may have overlapping indexes:

In [664]: df1 = DataFrame(randn(6, 4), columns=['A', 'B', 'C', 'D'])

In [665]: df2 = DataFrame(randn(3, 4), columns=['A', 'B', 'C', 'D'])

In [666]: df1
Out[666]: 
          A         B         C         D
0  0.084844  0.432390  1.519970 -0.493662
1  0.600178  0.274230  0.132885 -0.023688
2  2.410179  1.450520  0.206053 -0.251905
3 -2.213588  1.063327  1.266143  0.299368
4 -0.863838  0.408204 -1.048089 -0.025747
5 -0.988387  0.094055  1.262731  1.289997

In [667]: df2
Out[667]: 
          A         B         C         D
0  0.082423 -0.055758  0.536580 -0.489682
1  0.369374 -0.034571 -2.484478 -0.281461
2  0.030711  0.109121  1.126203 -0.977349

To do this, use the ignore_index argument:

In [668]: concat([df1, df2], ignore_index=True)
Out[668]: 
          A         B         C         D
0  0.084844  0.432390  1.519970 -0.493662
1  0.600178  0.274230  0.132885 -0.023688
2  2.410179  1.450520  0.206053 -0.251905
3 -2.213588  1.063327  1.266143  0.299368
4 -0.863838  0.408204 -1.048089 -0.025747
5 -0.988387  0.094055  1.262731  1.289997
6  0.082423 -0.055758  0.536580 -0.489682
7  0.369374 -0.034571 -2.484478 -0.281461
8  0.030711  0.109121  1.126203 -0.977349

This is also a valid argument to DataFrame.append:

In [669]: df1.append(df2, ignore_index=True)
Out[669]: 
          A         B         C         D
0  0.084844  0.432390  1.519970 -0.493662
1  0.600178  0.274230  0.132885 -0.023688
2  2.410179  1.450520  0.206053 -0.251905
3 -2.213588  1.063327  1.266143  0.299368
4 -0.863838  0.408204 -1.048089 -0.025747
5 -0.988387  0.094055  1.262731  1.289997
6  0.082423 -0.055758  0.536580 -0.489682
7  0.369374 -0.034571 -2.484478 -0.281461
8  0.030711  0.109121  1.126203 -0.977349

More concatenating with group keys

Let’s consider a variation on the first example presented:

In [670]: df = DataFrame(np.random.randn(10, 4))

In [671]: df
Out[671]: 
          0         1         2         3
0  1.474071 -0.064034 -1.282782  0.781836
1 -1.071357  0.441153  2.353925  0.583787
2  0.221471 -0.744471  0.758527  1.729689
3 -0.964980 -0.845696 -1.340896  1.846883
4 -1.328865  1.682706 -1.717693  0.888782
5  0.228440  0.901805  1.171216  0.520260
6 -1.197071 -1.066969 -0.303421 -0.858447
7  0.306996 -0.028665  0.384316  1.574159
8  1.588931  0.476720  0.473424 -0.242861
9 -0.014805 -0.284319  0.650776 -1.461665

# break it into pieces
In [672]: pieces = [df.ix[:, [0, 1]], df.ix[:, [2]], df.ix[:, [3]]]

In [673]: result = concat(pieces, axis=1, keys=['one', 'two', 'three'])

In [674]: result
Out[674]: 
        one                 two     three
          0         1         2         3
0  1.474071 -0.064034 -1.282782  0.781836
1 -1.071357  0.441153  2.353925  0.583787
2  0.221471 -0.744471  0.758527  1.729689
3 -0.964980 -0.845696 -1.340896  1.846883
4 -1.328865  1.682706 -1.717693  0.888782
5  0.228440  0.901805  1.171216  0.520260
6 -1.197071 -1.066969 -0.303421 -0.858447
7  0.306996 -0.028665  0.384316  1.574159
8  1.588931  0.476720  0.473424 -0.242861
9 -0.014805 -0.284319  0.650776 -1.461665

You can also pass a dict to concat in which case the dict keys will be used for the keys argument (unless other keys are specified):

In [675]: pieces = {'one': df.ix[:, [0, 1]],
   .....:           'two': df.ix[:, [2]],
   .....:           'three': df.ix[:, [3]]}

In [676]: concat(pieces, axis=1)
Out[676]: 
        one               three       two
          0         1         3         2
0  1.474071 -0.064034  0.781836 -1.282782
1 -1.071357  0.441153  0.583787  2.353925
2  0.221471 -0.744471  1.729689  0.758527
3 -0.964980 -0.845696  1.846883 -1.340896
4 -1.328865  1.682706  0.888782 -1.717693
5  0.228440  0.901805  0.520260  1.171216
6 -1.197071 -1.066969 -0.858447 -0.303421
7  0.306996 -0.028665  1.574159  0.384316
8  1.588931  0.476720 -0.242861  0.473424
9 -0.014805 -0.284319 -1.461665  0.650776

In [677]: concat(pieces, keys=['three', 'two'])
Out[677]: 
                2         3
three 0       NaN  0.781836
      1       NaN  0.583787
      2       NaN  1.729689
      3       NaN  1.846883
      4       NaN  0.888782
      5       NaN  0.520260
      6       NaN -0.858447
      7       NaN  1.574159
      8       NaN -0.242861
      9       NaN -1.461665
two   0 -1.282782       NaN
      1  2.353925       NaN
      2  0.758527       NaN
      3 -1.340896       NaN
      4 -1.717693       NaN
      5  1.171216       NaN
      6 -0.303421       NaN
      7  0.384316       NaN
      8  0.473424       NaN
      9  0.650776       NaN

The MultiIndex created has levels that are constructed from the passed keys and the columns of the DataFrame pieces:

In [678]: result.columns.levels
Out[678]: [Index([one, two, three], dtype=object), Int64Index([0, 1, 2, 3])]

If you wish to specify other levels (as will occasionally be the case), you can do so using the levels argument:

In [679]: result = concat(pieces, axis=1, keys=['one', 'two', 'three'],
   .....:                 levels=[['three', 'two', 'one', 'zero']],
   .....:                 names=['group_key'])

In [680]: result
Out[680]: 
group_key       one                 two     three
                  0         1         2         3
0          1.474071 -0.064034 -1.282782  0.781836
1         -1.071357  0.441153  2.353925  0.583787
2          0.221471 -0.744471  0.758527  1.729689
3         -0.964980 -0.845696 -1.340896  1.846883
4         -1.328865  1.682706 -1.717693  0.888782
5          0.228440  0.901805  1.171216  0.520260
6         -1.197071 -1.066969 -0.303421 -0.858447
7          0.306996 -0.028665  0.384316  1.574159
8          1.588931  0.476720  0.473424 -0.242861
9         -0.014805 -0.284319  0.650776 -1.461665

In [681]: result.columns.levels
Out[681]: [Index([three, two, one, zero], dtype=object), Int64Index([0, 1, 2, 3])]

Yes, this is fairly esoteric, but is actually necessary for implementing things like GroupBy where the order of a categorical variable is meaningful.

Appending rows to a DataFrame

While not especially efficient (since a new object must be created), you can append a single row to a DataFrame by passing a Series or dict to append, which returns a new DataFrame as above.

In [682]: df = DataFrame(np.random.randn(8, 4), columns=['A','B','C','D'])

In [683]: df
Out[683]: 
          A         B         C         D
0 -1.137707 -0.891060 -0.693921  1.613616
1  0.464000  0.227371 -0.496922  0.306389
2 -2.290613 -1.134623 -1.561819 -0.260838
3  0.281957  1.523962 -0.902937  0.068159
4 -0.057873 -0.368204 -1.144073  0.861209
5  0.800193  0.782098 -1.069094 -1.099248
6  0.255269  0.009750  0.661084  0.379319
7 -0.008434  1.952541 -1.056652  0.533946

In [684]: s = df.xs(3)

In [685]: df.append(s, ignore_index=True)
Out[685]: 
          A         B         C         D
0 -1.137707 -0.891060 -0.693921  1.613616
1  0.464000  0.227371 -0.496922  0.306389
2 -2.290613 -1.134623 -1.561819 -0.260838
3  0.281957  1.523962 -0.902937  0.068159
4 -0.057873 -0.368204 -1.144073  0.861209
5  0.800193  0.782098 -1.069094 -1.099248
6  0.255269  0.009750  0.661084  0.379319
7 -0.008434  1.952541 -1.056652  0.533946
8  0.281957  1.523962 -0.902937  0.068159

You should use ignore_index with this method to instruct DataFrame to discard its index. If you wish to preserve the index, you should construct an appropriately-indexed DataFrame and append or concatenate those objects.

You can also pass a list of dicts or Series:

In [686]: df = DataFrame(np.random.randn(5, 4),
   .....:                columns=['foo', 'bar', 'baz', 'qux'])

In [687]: dicts = [{'foo': 1, 'bar': 2, 'baz': 3, 'peekaboo': 4},
   .....:          {'foo': 5, 'bar': 6, 'baz': 7, 'peekaboo': 8}]

In [688]: result = df.append(dicts, ignore_index=True)

In [689]: result
Out[689]: 
        bar       baz       foo  peekaboo       qux
0  0.040403 -0.507516 -1.226970       NaN -0.230096
1 -1.934370 -1.652499  0.394500       NaN  1.488753
2  0.576897  1.146000 -0.896484       NaN  1.487349
3  2.121453  0.597701  0.604603       NaN  0.563700
4 -1.057909  1.375020  0.967661       NaN -0.928797
5  2.000000  3.000000  1.000000         4       NaN
6  6.000000  7.000000  5.000000         8       NaN

Database-style DataFrame joining/merging

pandas has full-featured, high performance in-memory join operations idiomatically very similar to relational databases like SQL. These methods perform significantly better (in some cases well over an order of magnitude better) than other open source implementations (like base::merge.data.frame in R). The reason for this is careful algorithmic design and internal layout of the data in DataFrame.

pandas provides a single function, merge, as the entry point for all standard database join operations between DataFrame objects:

merge(left, right, how='left', on=None, left_on=None, right_on=None,
      left_index=False, right_index=False, sort=True,
      suffixes=('.x', '.y'), copy=True)

Here’s a description of what each argument is for:

  • left: A DataFrame object
  • right: Another DataFrame object
  • on: Columns (names) to join on. Must be found in both the left and right DataFrame objects. If not passed and left_index and right_index are False, the intersection of the columns in the DataFrames will be inferred to be the join keys
  • left_on: Columns from the left DataFrame to use as keys. Can either be column names or arrays with length equal to the length of the DataFrame
  • right_on: Columns from the right DataFrame to use as keys. Can either be column names or arrays with length equal to the length of the DataFrame
  • left_index: If True, use the index (row labels) from the left DataFrame as its join key(s). In the case of a DataFrame with a MultiIndex (hierarchical), the number of levels must match the number of join keys from the right DataFrame
  • right_index: Same usage as left_index for the right DataFrame
  • how: One of 'left', 'right', 'outer', 'inner'. Defaults to inner. See below for more detailed description of each method
  • sort: Sort the result DataFrame by the join keys in lexicographical order. Defaults to True, setting to False will improve performance substantially in many cases
  • suffixes: A tuple of string suffixes to apply to overlapping columns. Defaults to ('.x', '.y').
  • copy: Always copy data (default True) from the passed DataFrame objects, even when reindexing is not necessary. Cannot be avoided in many cases but may improve performance / memory usage. The cases where copying can be avoided are somewhat pathological but this option is provided nonetheless.

merge is a function in the pandas namespace, and it is also available as a DataFrame instance method, with the calling DataFrame being implicitly considered the left object in the join.

The related DataFrame.join method, uses merge internally for the index-on-index and index-on-column(s) joins, but joins on indexes by default rather than trying to join on common columns (the default behavior for merge). If you are joining on index, you may wish to use DataFrame.join to save yourself some typing.

Brief primer on merge methods (relational algebra)

Experienced users of relational databases like SQL will be familiar with the terminology used to describe join operations between two SQL-table like structures (DataFrame objects). There are several cases to consider which are very important to understand:

  • one-to-one joins: for example when joining two DataFrame objects on their indexes (which must contain unique values)
  • many-to-one joins: for example when joining an index (unique) to one or more columns in a DataFrame
  • many-to-many joins: joining columns on columns.

Note

When joining columns on columns (potentially a many-to-many join), any indexes on the passed DataFrame objects will be discarded.

It is worth spending some time understanding the result of the many-to-many join case. In SQL / standard relational algebra, if a key combination appears more than once in both tables, the resulting table will have the Cartesian product of the associated data. Here is a very basic example with one unique key combination:

In [690]: left = DataFrame({'key': ['foo', 'foo'], 'lval': [1, 2]})

In [691]: right = DataFrame({'key': ['foo', 'foo'], 'rval': [4, 5]})

In [692]: left
Out[692]: 
   key  lval
0  foo     1
1  foo     2

In [693]: right
Out[693]: 
   key  rval
0  foo     4
1  foo     5

In [694]: merge(left, right, on='key')
Out[694]: 
   key  lval  rval
0  foo     1     4
1  foo     1     5
2  foo     2     4
3  foo     2     5

Here is a more complicated example with multiple join keys:

In [695]: left = DataFrame({'key1': ['foo', 'foo', 'bar'],
   .....:                   'key2': ['one', 'two', 'one'],
   .....:                   'lval': [1, 2, 3]})

In [696]: right = DataFrame({'key1': ['foo', 'foo', 'bar', 'bar'],
   .....:                    'key2': ['one', 'one', 'one', 'two'],
   .....:                    'rval': [4, 5, 6, 7]})

In [697]: merge(left, right, how='outer')
Out[697]: 
  key1 key2  lval  rval
0  bar  one     3     6
1  bar  two   NaN     7
2  foo  one     1     4
3  foo  one     1     5
4  foo  two     2   NaN

In [698]: merge(left, right, how='inner')
Out[698]: 
  key1 key2  lval  rval
0  bar  one     3     6
1  foo  one     1     4
2  foo  one     1     5

The how argument to merge specifies how to determine which keys are to be included in the resulting table. If a key combination does not appear in either the left or right tables, the values in the joined table will be NA. Here is a summary of the how options and their SQL equivalent names:

Merge method SQL Join Name Description
left LEFT OUTER JOIN Use keys from left frame only
right RIGHT OUTER JOIN Use keys from right frame only
outer FULL OUTER JOIN Use union of keys from both frames
inner INNER JOIN Use intersection of keys from both frames

Note that if using the index from either the left or right DataFrame (or both) using the left_index / right_index options, the join operation is no longer a many-to-many join by construction, as the index values are necessarily unique. There will be some examples of this below.

Joining on index

DataFrame.join is a convenient method for combining the columns of two potentially differently-indexed DataFrames into a single result DataFrame. Here is a very basic example:

In [699]: df = DataFrame(np.random.randn(8, 4), columns=['A','B','C','D'])

In [700]: df1 = df.ix[1:, ['A', 'B']]

In [701]: df2 = df.ix[:5, ['C', 'D']]

In [702]: df1
Out[702]: 
          A         B
1 -2.461467 -1.553902
2  1.771740 -0.670027
3 -3.201750  0.792716
4 -0.747169 -0.309038
5  0.936527  1.255746
6  0.062297 -0.110388
7  0.077849  0.629498

In [703]: df2
Out[703]: 
          C         D
0  0.377953  0.493672
1  2.015523 -1.833722
2  0.049307 -0.521493
3  0.146111  1.903247
4  0.393876  1.861468
5 -2.655452  1.219492

In [704]: df1.join(df2)
Out[704]: 
          A         B         C         D
1 -2.461467 -1.553902  2.015523 -1.833722
2  1.771740 -0.670027  0.049307 -0.521493
3 -3.201750  0.792716  0.146111  1.903247
4 -0.747169 -0.309038  0.393876  1.861468
5  0.936527  1.255746 -2.655452  1.219492
6  0.062297 -0.110388       NaN       NaN
7  0.077849  0.629498       NaN       NaN

In [705]: df1.join(df2, how='outer')
Out[705]: 
          A         B         C         D
0       NaN       NaN  0.377953  0.493672
1 -2.461467 -1.553902  2.015523 -1.833722
2  1.771740 -0.670027  0.049307 -0.521493
3 -3.201750  0.792716  0.146111  1.903247
4 -0.747169 -0.309038  0.393876  1.861468
5  0.936527  1.255746 -2.655452  1.219492
6  0.062297 -0.110388       NaN       NaN
7  0.077849  0.629498       NaN       NaN

In [706]: df1.join(df2, how='inner')
Out[706]: 
          A         B         C         D
1 -2.461467 -1.553902  2.015523 -1.833722
2  1.771740 -0.670027  0.049307 -0.521493
3 -3.201750  0.792716  0.146111  1.903247
4 -0.747169 -0.309038  0.393876  1.861468
5  0.936527  1.255746 -2.655452  1.219492

The data alignment here is on the indexes (row labels). This same behavior can be achieved using merge plus additional arguments instructing it to use the indexes:

In [707]: merge(df1, df2, left_index=True, right_index=True, how='outer')
Out[707]: 
          A         B         C         D
0       NaN       NaN  0.377953  0.493672
1 -2.461467 -1.553902  2.015523 -1.833722
2  1.771740 -0.670027  0.049307 -0.521493
3 -3.201750  0.792716  0.146111  1.903247
4 -0.747169 -0.309038  0.393876  1.861468
5  0.936527  1.255746 -2.655452  1.219492
6  0.062297 -0.110388       NaN       NaN
7  0.077849  0.629498       NaN       NaN

Joining key columns on an index

join takes an optional on argument which may be a column or multiple column names, which specifies that the passed DataFrame is to be aligned on that column in the DataFrame. These two function calls are completely equivalent:

left.join(right, on=key_or_keys)
merge(left, right, left_on=key_or_keys, right_index=True,
      how='left', sort=False)

Obviously you can choose whichever form you find more convenient. For many-to-one joins (where one of the DataFrame’s is already indexed by the join key), using join may be more convenient. Here is a simple example:

In [708]: df['key'] = ['foo', 'bar'] * 4

In [709]: to_join = DataFrame(randn(2, 2), index=['bar', 'foo'],
   .....:                     columns=['j1', 'j2'])

In [710]: df
Out[710]: 
          A         B         C         D  key
0 -0.308853 -0.681087  0.377953  0.493672  foo
1 -2.461467 -1.553902  2.015523 -1.833722  bar
2  1.771740 -0.670027  0.049307 -0.521493  foo
3 -3.201750  0.792716  0.146111  1.903247  bar
4 -0.747169 -0.309038  0.393876  1.861468  foo
5  0.936527  1.255746 -2.655452  1.219492  bar
6  0.062297 -0.110388 -1.184357 -0.558081  foo
7  0.077849  0.629498 -1.035260 -0.438229  bar

In [711]: to_join
Out[711]: 
           j1        j2
bar  0.503703  0.413086
foo -1.139050  0.660342

In [712]: df.join(to_join, on='key')
Out[712]: 
          A         B         C         D  key        j1        j2
0 -0.308853 -0.681087  0.377953  0.493672  foo -1.139050  0.660342
1 -2.461467 -1.553902  2.015523 -1.833722  bar  0.503703  0.413086
2  1.771740 -0.670027  0.049307 -0.521493  foo -1.139050  0.660342
3 -3.201750  0.792716  0.146111  1.903247  bar  0.503703  0.413086
4 -0.747169 -0.309038  0.393876  1.861468  foo -1.139050  0.660342
5  0.936527  1.255746 -2.655452  1.219492  bar  0.503703  0.413086
6  0.062297 -0.110388 -1.184357 -0.558081  foo -1.139050  0.660342
7  0.077849  0.629498 -1.035260 -0.438229  bar  0.503703  0.413086

In [713]: merge(df, to_join, left_on='key', right_index=True,
   .....:       how='left', sort=False)
Out[713]: 
          A         B         C         D  key        j1        j2
0 -0.308853 -0.681087  0.377953  0.493672  foo -1.139050  0.660342
1 -2.461467 -1.553902  2.015523 -1.833722  bar  0.503703  0.413086
2  1.771740 -0.670027  0.049307 -0.521493  foo -1.139050  0.660342
3 -3.201750  0.792716  0.146111  1.903247  bar  0.503703  0.413086
4 -0.747169 -0.309038  0.393876  1.861468  foo -1.139050  0.660342
5  0.936527  1.255746 -2.655452  1.219492  bar  0.503703  0.413086
6  0.062297 -0.110388 -1.184357 -0.558081  foo -1.139050  0.660342
7  0.077849  0.629498 -1.035260 -0.438229  bar  0.503703  0.413086

To join on multiple keys, the passed DataFrame must have a MultiIndex:

In [714]: index = MultiIndex(levels=[['foo', 'bar', 'baz', 'qux'],
   .....:                            ['one', 'two', 'three']],
   .....:                    labels=[[0, 0, 0, 1, 1, 2, 2, 3, 3, 3],
   .....:                            [0, 1, 2, 0, 1, 1, 2, 0, 1, 2]],
   .....:                    names=['first', 'second'])

In [715]: to_join = DataFrame(np.random.randn(10, 3), index=index,
   .....:                     columns=['j_one', 'j_two', 'j_three'])

# a little relevant example with NAs
In [716]: key1 = ['bar', 'bar', 'bar', 'foo', 'foo', 'baz', 'baz', 'qux',
   .....:         'qux', 'snap']

In [717]: key2 = ['two', 'one', 'three', 'one', 'two', 'one', 'two', 'two',
   .....:         'three', 'one']

In [718]: data = np.random.randn(len(key1))

In [719]: data = DataFrame({'key1' : key1, 'key2' : key2,
   .....:                   'data' : data})

In [720]: data
Out[720]: 
       data  key1   key2
0 -1.004168   bar    two
1 -1.377627   bar    one
2  0.499281   bar  three
3 -1.405256   foo    one
4  0.162565   foo    two
5 -0.067785   baz    one
6 -1.260006   baz    two
7 -1.132896   qux    two
8 -2.006481   qux  three
9  0.301016  snap    one

In [721]: to_join
Out[721]: 
                 j_one     j_two   j_three
first second                              
foo   one     0.464794 -0.309337 -0.649593
      two     0.683758 -0.643834  0.421287
      three   1.032814 -1.290493  0.787872
bar   one     1.515707 -0.276487 -0.223762
      two     1.397431  1.503874 -0.478905
baz   two    -0.135950 -0.730327 -0.033277
      three   0.281151 -1.298915 -2.819487
qux   one    -0.851985 -1.106952 -0.937731
      two    -1.537770  0.555759 -2.277282
      three  -0.390201  1.207122  0.178690

Now this can be joined by passing the two key column names:

In [722]: data.join(to_join, on=['key1', 'key2'])
Out[722]: 
       data  key1   key2     j_one     j_two   j_three
0 -1.004168   bar    two  1.397431  1.503874 -0.478905
1 -1.377627   bar    one  1.515707 -0.276487 -0.223762
2  0.499281   bar  three       NaN       NaN       NaN
3 -1.405256   foo    one  0.464794 -0.309337 -0.649593
4  0.162565   foo    two  0.683758 -0.643834  0.421287
5 -0.067785   baz    one       NaN       NaN       NaN
6 -1.260006   baz    two -0.135950 -0.730327 -0.033277
7 -1.132896   qux    two -1.537770  0.555759 -2.277282
8 -2.006481   qux  three -0.390201  1.207122  0.178690
9  0.301016  snap    one       NaN       NaN       NaN

The default for DataFrame.join is to perform a left join (essentially a “VLOOKUP” operation, for Excel users), which uses only the keys found in the calling DataFrame. Other join types, for example inner join, can be just as easily performed:

In [723]: data.join(to_join, on=['key1', 'key2'], how='inner')
Out[723]: 
       data key1   key2     j_one     j_two   j_three
0 -1.004168  bar    two  1.397431  1.503874 -0.478905
1 -1.377627  bar    one  1.515707 -0.276487 -0.223762
3 -1.405256  foo    one  0.464794 -0.309337 -0.649593
4  0.162565  foo    two  0.683758 -0.643834  0.421287
6 -1.260006  baz    two -0.135950 -0.730327 -0.033277
7 -1.132896  qux    two -1.537770  0.555759 -2.277282
8 -2.006481  qux  three -0.390201  1.207122  0.178690

As you can see, this drops any rows where there was no match.

Overlapping value columns

The merge suffixes argument takes a tuple of list of strings to append to overlapping column names in the input DataFrames to disambiguate the result columns:

In [724]: left = DataFrame({'key': ['foo', 'foo'], 'value': [1, 2]})

In [725]: right = DataFrame({'key': ['foo', 'foo'], 'value': [4, 5]})

In [726]: merge(left, right, on='key', suffixes=['_left', '_right'])
Out[726]: 
   key  value_left  value_right
0  foo           1            4
1  foo           1            5
2  foo           2            4
3  foo           2            5

DataFrame.join has lsuffix and rsuffix arguments which behave similarly.

Joining multiple DataFrame or Panel objects

A list or tuple of DataFrames can also be passed to DataFrame.join to join them together on their indexes. The same is true for Panel.join.

In [727]: df1 = df.ix[:, ['A', 'B']]

In [728]: df2 = df.ix[:, ['C', 'D']]

In [729]: df3 = df.ix[:, ['key']]

In [730]: df1
Out[730]: 
          A         B
0 -0.308853 -0.681087
1 -2.461467 -1.553902
2  1.771740 -0.670027
3 -3.201750  0.792716
4 -0.747169 -0.309038
5  0.936527  1.255746
6  0.062297 -0.110388
7  0.077849  0.629498

In [731]: df1.join([df2, df3])
Out[731]: 
          A         B         C         D  key
0 -0.308853 -0.681087  0.377953  0.493672  foo
1 -2.461467 -1.553902  2.015523 -1.833722  bar
2  1.771740 -0.670027  0.049307 -0.521493  foo
3 -3.201750  0.792716  0.146111  1.903247  bar
4 -0.747169 -0.309038  0.393876  1.861468  foo
5  0.936527  1.255746 -2.655452  1.219492  bar
6  0.062297 -0.110388 -1.184357 -0.558081  foo
7  0.077849  0.629498 -1.035260 -0.438229  bar