Table Of Contents

Search

Enter search terms or a module, class or function name.

Essential Basic Functionality

Here we discuss a lot of the essential functionality common to the pandas data structures. Here’s how to create some of the objects used in the examples from the previous section:

In [1]: index = date_range('1/1/2000', periods=8)

In [2]: s = Series(randn(5), index=['a', 'b', 'c', 'd', 'e'])

In [3]: df = DataFrame(randn(8, 3), index=index,
   ...:                columns=['A', 'B', 'C'])
   ...: 

In [4]: wp = Panel(randn(2, 5, 4), items=['Item1', 'Item2'],
   ...:            major_axis=date_range('1/1/2000', periods=5),
   ...:            minor_axis=['A', 'B', 'C', 'D'])
   ...: 

Head and Tail

To view a small sample of a Series or DataFrame object, use the head and tail methods. The default number of elements to display is five, but you may pass a custom number.

In [5]: long_series = Series(randn(1000))

In [6]: long_series.head()
Out[6]: 
0   -0.199038
1    1.095864
2   -0.200875
3    0.162291
4   -0.430489
dtype: float64

In [7]: long_series.tail(3)
Out[7]: 
997   -1.198693
998    1.238029
999   -1.344716
dtype: float64

Attributes and the raw ndarray(s)

pandas objects have a number of attributes enabling you to access the metadata

  • shape: gives the axis dimensions of the object, consistent with ndarray
  • Axis labels
    • Series: index (only axis)
    • DataFrame: index (rows) and columns
    • Panel: items, major_axis, and minor_axis

Note, these attributes can be safely assigned to!

In [8]: df[:2]
Out[8]: 
                   A         B         C
2000-01-01  0.232465 -0.789552 -0.364308
2000-01-02 -0.534541  0.822239 -0.443109

In [9]: df.columns = [x.lower() for x in df.columns]

In [10]: df
Out[10]: 
                   a         b         c
2000-01-01  0.232465 -0.789552 -0.364308
2000-01-02 -0.534541  0.822239 -0.443109
2000-01-03 -2.119990 -0.460149  1.813962
2000-01-04 -1.053571  0.009412 -0.165966
2000-01-05 -0.848662 -0.495553 -0.176421
2000-01-06 -0.423595 -1.035433 -1.035374
2000-01-07 -2.369079  0.524408 -0.871120
2000-01-08  1.585433  0.039501  2.274101

To get the actual data inside a data structure, one need only access the values property:

In [11]: s.values
Out[11]: array([ 1.1292,  0.2313, -0.1847, -0.1386, -0.9243])

In [12]: df.values
Out[12]: 
array([[ 0.2325, -0.7896, -0.3643],
       [-0.5345,  0.8222, -0.4431],
       [-2.12  , -0.4601,  1.814 ],
       [-1.0536,  0.0094, -0.166 ],
       [-0.8487, -0.4956, -0.1764],
       [-0.4236, -1.0354, -1.0354],
       [-2.3691,  0.5244, -0.8711],
       [ 1.5854,  0.0395,  2.2741]])

In [13]: wp.values
Out[13]: 
array([[[-1.1181,  0.4313,  0.5547, -1.3336],
        [-0.3322, -0.4859,  1.7259,  1.7993],
        [-0.9689, -0.7795, -2.0007, -1.8666],
        [-1.1013,  1.9575,  0.0589,  0.7581],
        [ 0.0766, -0.5485, -0.1605, -0.3778]],

       [[ 0.2499, -0.3413, -0.2726, -0.2774],
        [-1.1029,  0.1003, -1.6028,  0.9201],
        [-0.6439,  0.0603, -0.4349, -0.4943],
        [ 0.738 ,  0.4516,  0.3341, -0.7871],
        [ 0.6514, -0.7419,  1.1939, -2.3958]]])

If a DataFrame or Panel contains homogeneously-typed data, the ndarray can actually be modified in-place, and the changes will be reflected in the data structure. For heterogeneous data (e.g. some of the DataFrame’s columns are not all the same dtype), this will not be the case. The values attribute itself, unlike the axis labels, cannot be assigned to.

Note

When working with heterogeneous data, the dtype of the resulting ndarray will be chosen to accommodate all of the data involved. For example, if strings are involved, the result will be of object dtype. If there are only floats and integers, the resulting array will be of float dtype.

Accelerated operations

pandas has support for accelerating certain types of binary numerical and boolean operations using the numexpr library (starting in 0.11.0) and the bottleneck libraries.

These libraries are especially useful when dealing with large data sets, and provide large speedups. numexpr uses smart chunking, caching, and multiple cores. bottleneck is a set of specialized cython routines that are especially fast when dealing with arrays that have nans.

Here is a sample (using 100 column x 100,000 row DataFrames):

Operation 0.11.0 (ms) Prior Version (ms) Ratio to Prior
df1 > df2 13.32 125.35 0.1063
df1 * df2 21.71 36.63 0.5928
df1 + df2 22.04 36.50 0.6039

You are highly encouraged to install both libraries. See the section Recommended Dependencies for more installation info.

Flexible binary operations

With binary operations between pandas data structures, there are two key points of interest:

  • Broadcasting behavior between higher- (e.g. DataFrame) and lower-dimensional (e.g. Series) objects.
  • Missing data in computations

We will demonstrate how to manage these issues independently, though they can be handled simultaneously.

Matching / broadcasting behavior

DataFrame has the methods add, sub, mul, div and related functions radd, rsub, ... for carrying out binary operations. For broadcasting behavior, Series input is of primary interest. Using these functions, you can use to either match on the index or columns via the axis keyword:

In [14]: df = DataFrame({'one' : Series(randn(3), index=['a', 'b', 'c']),
   ....:                 'two' : Series(randn(4), index=['a', 'b', 'c', 'd']),
   ....:                 'three' : Series(randn(3), index=['b', 'c', 'd'])})
   ....: 

In [15]: df
Out[15]: 
        one     three       two
a -0.701368       NaN -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [16]: row = df.ix[1]

In [17]: column = df['two']

In [18]: df.sub(row, axis='columns')
Out[18]: 
        one     three       two
a -0.810701       NaN -0.724777
b  0.000000  0.000000  0.000000
c -0.340950  0.205973 -0.640340
d       NaN  0.186952 -0.533630

In [19]: df.sub(row, axis=1)
Out[19]: 
        one     three       two
a -0.810701       NaN -0.724777
b  0.000000  0.000000  0.000000
c -0.340950  0.205973 -0.640340
d       NaN  0.186952 -0.533630

In [20]: df.sub(column, axis='index')
Out[20]: 
        one     three  two
a -0.614265       NaN    0
b -0.528341 -0.992033    0
c -0.228950 -0.145720    0
d       NaN -0.271451    0

In [21]: df.sub(column, axis=0)
Out[21]: 
        one     three  two
a -0.614265       NaN    0
b -0.528341 -0.992033    0
c -0.228950 -0.145720    0
d       NaN -0.271451    0

Furthermore you can align a level of a multi-indexed DataFrame with a Series.

In [22]: dfmi = df.copy()

In [23]: dfmi.index = MultiIndex.from_tuples([(1,'a'),(1,'b'),(1,'c'),(2,'a')],
   ....:                                     names=['first','second'])
   ....: 

In [24]: dfmi.sub(column, axis=0, level='second')
Out[24]: 
                   one     three       two
first second                              
1     a      -0.614265       NaN  0.000000
      b      -0.528341 -0.992033  0.000000
      c      -0.228950 -0.145720  0.000000
2     a            NaN -0.080304  0.191147

With Panel, describing the matching behavior is a bit more difficult, so the arithmetic methods instead (and perhaps confusingly?) give you the option to specify the broadcast axis. For example, suppose we wished to demean the data over a particular axis. This can be accomplished by taking the mean over an axis and broadcasting over the same axis:

In [25]: major_mean = wp.mean(axis='major')

In [26]: major_mean
Out[26]: 
      Item1     Item2
A -0.688773 -0.021497
B  0.114982 -0.094183
C  0.035674 -0.156470
D -0.204142 -0.606887

In [27]: wp.sub(major_mean, axis='major')
Out[27]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 2 (items) x 5 (major_axis) x 4 (minor_axis)
Items axis: Item1 to Item2
Major_axis axis: 2000-01-01 00:00:00 to 2000-01-05 00:00:00
Minor_axis axis: A to D

And similarly for axis="items" and axis="minor".

Note

I could be convinced to make the axis argument in the DataFrame methods match the broadcasting behavior of Panel. Though it would require a transition period so users can change their code...

Missing data / operations with fill values

In Series and DataFrame (though not yet in Panel), the arithmetic functions have the option of inputting a fill_value, namely a value to substitute when at most one of the values at a location are missing. For example, when adding two DataFrame objects, you may wish to treat NaN as 0 unless both DataFrames are missing that value, in which case the result will be NaN (you can later replace NaN with some other value using fillna if you wish).

In [28]: df
Out[28]: 
        one     three       two
a -0.701368       NaN -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [29]: df2
Out[29]: 
        one     three       two
a -0.701368  1.000000 -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [30]: df + df2
Out[30]: 
        one     three       two
a -1.402736       NaN -0.174206
b  0.218666 -0.708719  1.275347
c -0.463233 -0.296773 -0.005333
d       NaN -0.334814  0.208088

In [31]: df.add(df2, fill_value=0)
Out[31]: 
        one     three       two
a -1.402736  1.000000 -0.174206
b  0.218666 -0.708719  1.275347
c -0.463233 -0.296773 -0.005333
d       NaN -0.334814  0.208088

Flexible Comparisons

Starting in v0.8, pandas introduced binary comparison methods eq, ne, lt, gt, le, and ge to Series and DataFrame whose behavior is analogous to the binary arithmetic operations described above:

In [32]: df.gt(df2)
Out[32]: 
     one  three    two
a  False  False  False
b  False  False  False
c  False  False  False
d  False  False  False

In [33]: df2.ne(df)
Out[33]: 
     one  three    two
a  False   True  False
b  False  False  False
c  False  False  False
d   True  False  False

These operations produce a pandas object the same type as the left-hand-side input that if of dtype bool. These boolean objects can be used in indexing operations, see here

Boolean Reductions

You can apply the reductions: empty, any(), all(), and bool() to provide a way to summarize a boolean result.

In [34]: (df>0).all()
Out[34]: 
one      False
three    False
two      False
dtype: bool

In [35]: (df>0).any()
Out[35]: 
one       True
three    False
two       True
dtype: bool

You can reduce to a final boolean value.

In [36]: (df>0).any().any()
Out[36]: True

You can test if a pandas object is empty, via the empty property.

In [37]: df.empty
Out[37]: False

In [38]: DataFrame(columns=list('ABC')).empty
Out[38]: True

To evaluate single-element pandas objects in a boolean context, use the method .bool():

In [39]: Series([True]).bool()
Out[39]: True

In [40]: Series([False]).bool()
Out[40]: False

In [41]: DataFrame([[True]]).bool()
Out[41]: True

In [42]: DataFrame([[False]]).bool()
Out[42]: False

Warning

You might be tempted to do the following:

>>>if df:
     ...

Or

>>> df and df2

These both will raise as you are trying to compare multiple values.

ValueError: The truth value of an array is ambiguous. Use a.empty, a.any() or a.all().

See gotchas for a more detailed discussion.

Comparing if objects are equivalent

Often you may find there is more than one way to compute the same result. As a simple example, consider df+df and df*2. To test that these two computations produce the same result, given the tools shown above, you might imagine using (df+df == df*2).all(). But in fact, this expression is False:

In [43]: df+df == df*2
Out[43]: 
     one  three   two
a   True  False  True
b   True   True  True
c   True   True  True
d  False   True  True

In [44]: (df+df == df*2).all()
Out[44]: 
one      False
three    False
two       True
dtype: bool

Notice that the boolean DataFrame df+df == df*2 contains some False values! That is because NaNs do not compare as equals:

In [45]: np.nan == np.nan
Out[45]: False

So, as of v0.13.1, NDFrames (such as Series, DataFrames, and Panels) have an equals method for testing equality, with NaNs in corresponding locations treated as equal.

In [46]: (df+df).equals(df*2)
Out[46]: True

Combining overlapping data sets

A problem occasionally arising is the combination of two similar data sets where values in one are preferred over the other. An example would be two data series representing a particular economic indicator where one is considered to be of “higher quality”. However, the lower quality series might extend further back in history or have more complete data coverage. As such, we would like to combine two DataFrame objects where missing values in one DataFrame are conditionally filled with like-labeled values from the other DataFrame. The function implementing this operation is combine_first, which we illustrate:

In [47]: df1 = DataFrame({'A' : [1., np.nan, 3., 5., np.nan],
   ....:                  'B' : [np.nan, 2., 3., np.nan, 6.]})
   ....: 

In [48]: df2 = DataFrame({'A' : [5., 2., 4., np.nan, 3., 7.],
   ....:                  'B' : [np.nan, np.nan, 3., 4., 6., 8.]})
   ....: 

In [49]: df1
Out[49]: 
    A   B
0   1 NaN
1 NaN   2
2   3   3
3   5 NaN
4 NaN   6

In [50]: df2
Out[50]: 
    A   B
0   5 NaN
1   2 NaN
2   4   3
3 NaN   4
4   3   6
5   7   8

In [51]: df1.combine_first(df2)
Out[51]: 
   A   B
0  1 NaN
1  2   2
2  3   3
3  5   4
4  3   6
5  7   8

General DataFrame Combine

The combine_first method above calls the more general DataFrame method combine. This method takes another DataFrame and a combiner function, aligns the input DataFrame and then passes the combiner function pairs of Series (ie, columns whose names are the same).

So, for instance, to reproduce combine_first as above:

In [52]: combiner = lambda x, y: np.where(isnull(x), y, x)

In [53]: df1.combine(df2, combiner)
Out[53]: 
   A   B
0  1 NaN
1  2   2
2  3   3
3  5   4
4  3   6
5  7   8

Descriptive statistics

A large number of methods for computing descriptive statistics and other related operations on Series, DataFrame, and Panel. Most of these are aggregations (hence producing a lower-dimensional result) like sum, mean, and quantile, but some of them, like cumsum and cumprod, produce an object of the same size. Generally speaking, these methods take an axis argument, just like ndarray.{sum, std, ...}, but the axis can be specified by name or integer:

  • Series: no axis argument needed
  • DataFrame: “index” (axis=0, default), “columns” (axis=1)
  • Panel: “items” (axis=0), “major” (axis=1, default), “minor” (axis=2)

For example:

In [54]: df
Out[54]: 
        one     three       two
a -0.701368       NaN -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [55]: df.mean(0)
Out[55]: 
one     -0.274551
three   -0.223384
two      0.162987
dtype: float64

In [56]: df.mean(1)
Out[56]: 
a   -0.394235
b    0.130882
c   -0.127557
d   -0.031682
dtype: float64

All such methods have a skipna option signaling whether to exclude missing data (True by default):

In [57]: df.sum(0, skipna=False)
Out[57]: 
one           NaN
three         NaN
two      0.651948
dtype: float64

In [58]: df.sum(axis=1, skipna=True)
Out[58]: 
a   -0.788471
b    0.392647
c   -0.382670
d   -0.063363
dtype: float64

Combined with the broadcasting / arithmetic behavior, one can describe various statistical procedures, like standardization (rendering data zero mean and standard deviation 1), very concisely:

In [59]: ts_stand = (df - df.mean()) / df.std()

In [60]: ts_stand.std()
Out[60]: 
one      1
three    1
two      1
dtype: float64

In [61]: xs_stand = df.sub(df.mean(1), axis=0).div(df.std(1), axis=0)

In [62]: xs_stand.std(1)
Out[62]: 
a    1
b    1
c    1
d    1
dtype: float64

Note that methods like cumsum and cumprod preserve the location of NA values:

In [63]: df.cumsum()
Out[63]: 
        one     three       two
a -0.701368       NaN -0.087103
b -0.592035 -0.354359  0.550570
c -0.823652 -0.502746  0.547904
d       NaN -0.670153  0.651948

Here is a quick reference summary table of common functions. Each also takes an optional level parameter which applies only if the object has a hierarchical index.

Function Description
count Number of non-null observations
sum Sum of values
mean Mean of values
mad Mean absolute deviation
median Arithmetic median of values
min Minimum
max Maximum
mode Mode
abs Absolute Value
prod Product of values
std Unbiased standard deviation
var Unbiased variance
sem Unbiased standard error of the mean
skew Unbiased skewness (3rd moment)
kurt Unbiased kurtosis (4th moment)
quantile Sample quantile (value at %)
cumsum Cumulative sum
cumprod Cumulative product
cummax Cumulative maximum
cummin Cumulative minimum

Note that by chance some NumPy methods, like mean, std, and sum, will exclude NAs on Series input by default:

In [64]: np.mean(df['one'])
Out[64]: -0.27455055654271204

In [65]: np.mean(df['one'].values)
Out[65]: nan

Series also has a method nunique which will return the number of unique non-null values:

In [66]: series = Series(randn(500))

In [67]: series[20:500] = np.nan

In [68]: series[10:20]  = 5

In [69]: series.nunique()
Out[69]: 11

Summarizing data: describe

There is a convenient describe function which computes a variety of summary statistics about a Series or the columns of a DataFrame (excluding NAs of course):

In [70]: series = Series(randn(1000))

In [71]: series[::2] = np.nan

In [72]: series.describe()
Out[72]: 
count    500.000000
mean      -0.019898
std        1.019180
min       -2.628792
25%       -0.649795
50%       -0.059405
75%        0.651932
max        3.240991
dtype: float64

In [73]: frame = DataFrame(randn(1000, 5), columns=['a', 'b', 'c', 'd', 'e'])

In [74]: frame.ix[::2] = np.nan

In [75]: frame.describe()
Out[75]: 
                a           b           c           d           e
count  500.000000  500.000000  500.000000  500.000000  500.000000
mean     0.051388    0.053476   -0.035612    0.015388    0.057804
std      0.989217    0.995961    0.977047    0.968385    1.022528
min     -3.224136   -2.606460   -2.762875   -2.961757   -2.829100
25%     -0.657420   -0.597123   -0.688961   -0.695019   -0.738097
50%      0.042928    0.018837   -0.071830   -0.011326    0.073287
75%      0.702445    0.693542    0.600454    0.680924    0.807670
max      3.034008    3.104512    2.812028    2.623914    3.542846

You can select specific percentiles to include in the output:

In [76]: series.describe(percentiles=[.05, .25, .75, .95])
Out[76]: 
count    500.000000
mean      -0.019898
std        1.019180
min       -2.628792
5%        -1.670021
25%       -0.649795
50%       -0.059405
75%        0.651932
95%        1.584100
max        3.240991
dtype: float64

By default, the median is always included.

For a non-numerical Series object, describe will give a simple summary of the number of unique values and most frequently occurring values:

In [77]: s = Series(['a', 'a', 'b', 'b', 'a', 'a', np.nan, 'c', 'd', 'a'])

In [78]: s.describe()
Out[78]: 
count     9
unique    4
top       a
freq      5
dtype: object

There also is a utility function, value_range which takes a DataFrame and returns a series with the minimum/maximum values in the DataFrame.

Index of Min/Max Values

The idxmin and idxmax functions on Series and DataFrame compute the index labels with the minimum and maximum corresponding values:

In [79]: s1 = Series(randn(5))

In [80]: s1
Out[80]: 
0   -0.574018
1    0.668292
2    0.303418
3   -1.190271
4    0.138399
dtype: float64

In [81]: s1.idxmin(), s1.idxmax()
Out[81]: (3, 1)

In [82]: df1 = DataFrame(randn(5,3), columns=['A','B','C'])

In [83]: df1
Out[83]: 
          A         B         C
0 -0.184355 -1.054354 -1.613138
1 -0.050807 -2.130168 -1.852271
2  0.455674  2.571061 -1.152538
3 -1.638940 -0.364831 -0.348520
4  0.202856  0.777088 -0.358316

In [84]: df1.idxmin(axis=0)
Out[84]: 
A    3
B    1
C    1
dtype: int64

In [85]: df1.idxmax(axis=1)
Out[85]: 
0    A
1    A
2    B
3    C
4    B
dtype: object

When there are multiple rows (or columns) matching the minimum or maximum value, idxmin and idxmax return the first matching index:

In [86]: df3 = DataFrame([2, 1, 1, 3, np.nan], columns=['A'], index=list('edcba'))

In [87]: df3
Out[87]: 
    A
e   2
d   1
c   1
b   3
a NaN

In [88]: df3['A'].idxmin()
Out[88]: 'd'

Note

idxmin and idxmax are called argmin and argmax in NumPy.

Value counts (histogramming) / Mode

The value_counts Series method and top-level function computes a histogram of a 1D array of values. It can also be used as a function on regular arrays:

In [89]: data = np.random.randint(0, 7, size=50)

In [90]: data
Out[90]: 
array([4, 6, 6, 1, 2, 1, 0, 5, 3, 2, 4, 3, 1, 3, 5, 3, 0, 0, 4, 4, 6, 1, 0,
       4, 3, 2, 1, 3, 1, 5, 6, 3, 1, 2, 4, 4, 3, 3, 2, 2, 2, 3, 2, 3, 0, 1,
       2, 4, 5, 5])

In [91]: s = Series(data)

In [92]: s.value_counts()
Out[92]: 
3    11
2     9
4     8
1     8
5     5
0     5
6     4
dtype: int64

In [93]: value_counts(data)
Out[93]: 
3    11
2     9
4     8
1     8
5     5
0     5
6     4
dtype: int64

Similarly, you can get the most frequently occurring value(s) (the mode) of the values in a Series or DataFrame:

In [94]: s5 = Series([1, 1, 3, 3, 3, 5, 5, 7, 7, 7])

In [95]: s5.mode()
Out[95]: 
0    3
1    7
dtype: int64

In [96]: df5 = DataFrame({"A": np.random.randint(0, 7, size=50),
   ....:                  "B": np.random.randint(-10, 15, size=50)})
   ....: 

In [97]: df5.mode()
Out[97]: 
   A   B
0  5  -4
1  6 NaN

Discretization and quantiling

Continuous values can be discretized using the cut (bins based on values) and qcut (bins based on sample quantiles) functions:

In [98]: arr = np.random.randn(20)

In [99]: factor = cut(arr, 4)

In [100]: factor
Out[100]: 
 (-0.886, -0.0912]
 (-0.886, -0.0912]
 (-0.886, -0.0912]
    (1.493, 2.285]
    (0.701, 1.493]
...
  (-0.0912, 0.701]
 (-0.886, -0.0912]
    (0.701, 1.493]
    (0.701, 1.493]
  (-0.0912, 0.701]
    (1.493, 2.285]
Levels (4): Index(['(-0.886, -0.0912]', '(-0.0912, 0.701]',
                   '(0.701, 1.493]', '(1.493, 2.285]'], dtype=object)
Length: 20

In [101]: factor = cut(arr, [-5, -1, 0, 1, 5])

In [102]: factor
Out[102]: 
 (-1, 0]
 (-1, 0]
 (-1, 0]
  (1, 5]
  (1, 5]
...
  (0, 1]
 (-1, 0]
  (0, 1]
  (0, 1]
  (0, 1]
  (1, 5]
Levels (4): Index(['(-5, -1]', '(-1, 0]', '(0, 1]', '(1, 5]'], dtype=object)
Length: 20

qcut computes sample quantiles. For example, we could slice up some normally distributed data into equal-size quartiles like so:

In [103]: arr = np.random.randn(30)

In [104]: factor = qcut(arr, [0, .25, .5, .75, 1])

In [105]: factor
Out[105]: 
 [-1.861, -0.487]
  (0.0554, 0.658]
   (0.658, 2.259]
 [-1.861, -0.487]
   (0.658, 2.259]
...
  (0.0554, 0.658]
  (0.0554, 0.658]
   (0.658, 2.259]
 [-1.861, -0.487]
  (0.0554, 0.658]
 (-0.487, 0.0554]
Levels (4): Index(['[-1.861, -0.487]', '(-0.487, 0.0554]',
                   '(0.0554, 0.658]', '(0.658, 2.259]'], dtype=object)
Length: 30

In [106]: value_counts(factor)
Out[106]: 
(0.658, 2.259]      8
[-1.861, -0.487]    8
(0.0554, 0.658]     7
(-0.487, 0.0554]    7
dtype: int64

We can also pass infinite values to define the bins:

In [107]: arr = np.random.randn(20)

In [108]: factor = cut(arr, [-np.inf, 0, np.inf])

In [109]: factor
Out[109]: 
  (0, inf]
  (0, inf]
 (-inf, 0]
  (0, inf]
 (-inf, 0]
...
 (-inf, 0]
  (0, inf]
  (0, inf]
 (-inf, 0]
  (0, inf]
 (-inf, 0]
Levels (2): Index(['(-inf, 0]', '(0, inf]'], dtype=object)
Length: 20

Function application

Arbitrary functions can be applied along the axes of a DataFrame or Panel using the apply method, which, like the descriptive statistics methods, take an optional axis argument:

In [110]: df.apply(np.mean)
Out[110]: 
one     -0.274551
three   -0.223384
two      0.162987
dtype: float64

In [111]: df.apply(np.mean, axis=1)
Out[111]: 
a   -0.394235
b    0.130882
c   -0.127557
d   -0.031682
dtype: float64

In [112]: df.apply(lambda x: x.max() - x.min())
Out[112]: 
one      0.810701
three    0.205973
two      0.724777
dtype: float64

In [113]: df.apply(np.cumsum)
Out[113]: 
        one     three       two
a -0.701368       NaN -0.087103
b -0.592035 -0.354359  0.550570
c -0.823652 -0.502746  0.547904
d       NaN -0.670153  0.651948

In [114]: df.apply(np.exp)
Out[114]: 
        one     three       two
a  0.495907       NaN  0.916583
b  1.115534  0.701623  1.892074
c  0.793250  0.862098  0.997337
d       NaN  0.845855  1.109649

Depending on the return type of the function passed to apply, the result will either be of lower dimension or the same dimension.

apply combined with some cleverness can be used to answer many questions about a data set. For example, suppose we wanted to extract the date where the maximum value for each column occurred:

In [115]: tsdf = DataFrame(randn(1000, 3), columns=['A', 'B', 'C'],
   .....:                  index=date_range('1/1/2000', periods=1000))
   .....: 

In [116]: tsdf.apply(lambda x: x.idxmax())
Out[116]: 
A   2002-08-19
B   2000-11-30
C   2002-01-10
dtype: datetime64[ns]

You may also pass additional arguments and keyword arguments to the apply method. For instance, consider the following function you would like to apply:

def subtract_and_divide(x, sub, divide=1):
    return (x - sub) / divide

You may then apply this function as follows:

df.apply(subtract_and_divide, args=(5,), divide=3)

Another useful feature is the ability to pass Series methods to carry out some Series operation on each column or row:

In [117]: tsdf
Out[117]: 
                   A         B         C
2000-01-01 -1.226159  0.173875 -0.798063
2000-01-02  0.127076  0.141070 -2.186743
2000-01-03 -1.804229  0.879800  0.465165
2000-01-04       NaN       NaN       NaN
2000-01-05       NaN       NaN       NaN
2000-01-06       NaN       NaN       NaN
2000-01-07       NaN       NaN       NaN
2000-01-08  1.542261  0.524780  1.445690
2000-01-09 -1.104998 -0.470200  0.336180
2000-01-10 -0.947692 -0.262122 -0.423769

In [118]: tsdf.apply(Series.interpolate)
Out[118]: 
                   A         B         C
2000-01-01 -1.226159  0.173875 -0.798063
2000-01-02  0.127076  0.141070 -2.186743
2000-01-03 -1.804229  0.879800  0.465165
2000-01-04 -1.134931  0.808796  0.661270
2000-01-05 -0.465633  0.737792  0.857375
2000-01-06  0.203665  0.666788  1.053480
2000-01-07  0.872963  0.595784  1.249585
2000-01-08  1.542261  0.524780  1.445690
2000-01-09 -1.104998 -0.470200  0.336180
2000-01-10 -0.947692 -0.262122 -0.423769

Finally, apply takes an argument raw which is False by default, which converts each row or column into a Series before applying the function. When set to True, the passed function will instead receive an ndarray object, which has positive performance implications if you do not need the indexing functionality.

See also

The section on GroupBy demonstrates related, flexible functionality for grouping by some criterion, applying, and combining the results into a Series, DataFrame, etc.

Applying elementwise Python functions

Since not all functions can be vectorized (accept NumPy arrays and return another array or value), the methods applymap on DataFrame and analogously map on Series accept any Python function taking a single value and returning a single value. For example:

In [119]: df4
Out[119]: 
        one     three       two
a -0.701368       NaN -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [120]: f = lambda x: len(str(x))

In [121]: df4['one'].map(f)
Out[121]: 
a    15
b    14
c    15
d     3
Name: one, dtype: int64

In [122]: df4.applymap(f)
Out[122]: 
   one  three  two
a   15      3   16
b   14     15   14
c   15     15   17
d    3     15   14

Series.map has an additional feature which is that it can be used to easily “link” or “map” values defined by a secondary series. This is closely related to merging/joining functionality:

In [123]: s = Series(['six', 'seven', 'six', 'seven', 'six'],
   .....:            index=['a', 'b', 'c', 'd', 'e'])
   .....: 

In [124]: t = Series({'six' : 6., 'seven' : 7.})

In [125]: s
Out[125]: 
a      six
b    seven
c      six
d    seven
e      six
dtype: object

In [126]: s.map(t)
Out[126]: 
a    6
b    7
c    6
d    7
e    6
dtype: float64

Applying with a Panel

Applying with a Panel will pass a Series to the applied function. If the applied function returns a Series, the result of the application will be a Panel. If the applied function reduces to a scalar, the result of the application will be a DataFrame.

Note

Prior to 0.13.1 apply on a Panel would only work on ufuncs (e.g. np.sum/np.max).

In [127]: import pandas.util.testing as tm

In [128]: panel = tm.makePanel(5)

In [129]: panel
Out[129]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 3 (items) x 5 (major_axis) x 4 (minor_axis)
Items axis: ItemA to ItemC
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-07 00:00:00
Minor_axis axis: A to D

In [130]: panel['ItemA']
Out[130]: 
                   A         B         C         D
2000-01-03  0.166882 -0.597361 -1.200639  0.174260
2000-01-04 -1.759496 -1.514940 -1.872993 -0.581163
2000-01-05  0.901336 -1.640398  0.825210  0.087916
2000-01-06 -0.317478 -1.130643 -0.392715  0.416971
2000-01-07 -0.681335 -0.245890 -1.994150  0.666084

A transformational apply.

In [131]: result = panel.apply(lambda x: x*2, axis='items')

In [132]: result
Out[132]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 3 (items) x 5 (major_axis) x 4 (minor_axis)
Items axis: ItemA to ItemC
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-07 00:00:00
Minor_axis axis: A to D

In [133]: result['ItemA']
Out[133]: 
                   A         B         C         D
2000-01-03  0.333764 -1.194722 -2.401278  0.348520
2000-01-04 -3.518991 -3.029880 -3.745986 -1.162326
2000-01-05  1.802673 -3.280796  1.650421  0.175832
2000-01-06 -0.634955 -2.261286 -0.785430  0.833943
2000-01-07 -1.362670 -0.491779 -3.988300  1.332168

A reduction operation.

In [134]: panel.apply(lambda x: x.dtype, axis='items')
Out[134]: 
                  A        B        C        D
2000-01-03  float64  float64  float64  float64
2000-01-04  float64  float64  float64  float64
2000-01-05  float64  float64  float64  float64
2000-01-06  float64  float64  float64  float64
2000-01-07  float64  float64  float64  float64

A similar reduction type operation

In [135]: panel.apply(lambda x: x.sum(), axis='major_axis')
Out[135]: 
      ItemA     ItemB     ItemC
A -1.690090  1.840259  0.010754
B -5.129232  0.860182  0.178018
C -4.635286  0.545328  2.456520
D  0.764068 -3.623586  1.761541

This last reduction is equivalent to

In [136]: panel.sum('major_axis')
Out[136]: 
      ItemA     ItemB     ItemC
A -1.690090  1.840259  0.010754
B -5.129232  0.860182  0.178018
C -4.635286  0.545328  2.456520
D  0.764068 -3.623586  1.761541

A transformation operation that returns a Panel, but is computing the z-score across the major_axis.

In [137]: result = panel.apply(
   .....:            lambda x: (x-x.mean())/x.std(),
   .....:            axis='major_axis')
   .....: 

In [138]: result
Out[138]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 3 (items) x 5 (major_axis) x 4 (minor_axis)
Items axis: ItemA to ItemC
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-07 00:00:00
Minor_axis axis: A to D

In [139]: result['ItemA']
Out[139]: 
                   A         B         C         D
2000-01-03  0.509389  0.719204 -0.234072  0.045812
2000-01-04 -1.434116 -0.820934 -0.809328 -1.567858
2000-01-05  1.250373 -1.031513  1.499214 -0.138629
2000-01-06  0.020723 -0.175899  0.457175  0.564271
2000-01-07 -0.346370  1.309142 -0.912988  1.096405

Apply can also accept multiple axes in the axis argument. This will pass a DataFrame of the cross-section to the applied function.

In [140]: f = lambda x: ((x.T-x.mean(1))/x.std(1)).T

In [141]: result = panel.apply(f, axis = ['items','major_axis'])

In [142]: result
Out[142]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 4 (items) x 5 (major_axis) x 3 (minor_axis)
Items axis: A to D
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-07 00:00:00
Minor_axis axis: ItemA to ItemC

In [143]: result.loc[:,:,'ItemA']
Out[143]: 
                   A         B         C         D
2000-01-03  0.783778 -0.648605 -0.903128  0.450190
2000-01-04 -0.884670 -1.046087 -1.096521 -0.900467
2000-01-05  1.140729 -1.124651  0.716895  0.754324
2000-01-06 -1.043494  0.029043 -0.991042  0.845339
2000-01-07 -1.125870 -0.536928 -1.152240 -0.182526

This is equivalent to the following

In [144]: result = Panel(dict([ (ax,f(panel.loc[:,:,ax]))
   .....:                         for ax in panel.minor_axis ]))
   .....: 

In [145]: result
Out[145]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 4 (items) x 5 (major_axis) x 3 (minor_axis)
Items axis: A to D
Major_axis axis: 2000-01-03 00:00:00 to 2000-01-07 00:00:00
Minor_axis axis: ItemA to ItemC

In [146]: result.loc[:,:,'ItemA']
Out[146]: 
                   A         B         C         D
2000-01-03  0.783778 -0.648605 -0.903128  0.450190
2000-01-04 -0.884670 -1.046087 -1.096521 -0.900467
2000-01-05  1.140729 -1.124651  0.716895  0.754324
2000-01-06 -1.043494  0.029043 -0.991042  0.845339
2000-01-07 -1.125870 -0.536928 -1.152240 -0.182526

Reindexing and altering labels

reindex is the fundamental data alignment method in pandas. It is used to implement nearly all other features relying on label-alignment functionality. To reindex means to conform the data to match a given set of labels along a particular axis. This accomplishes several things:

  • Reorders the existing data to match a new set of labels
  • Inserts missing value (NA) markers in label locations where no data for that label existed
  • If specified, fill data for missing labels using logic (highly relevant to working with time series data)

Here is a simple example:

In [147]: s = Series(randn(5), index=['a', 'b', 'c', 'd', 'e'])

In [148]: s
Out[148]: 
a    1.112686
b   -1.069046
c   -1.218080
d   -0.944778
e    0.005240
dtype: float64

In [149]: s.reindex(['e', 'b', 'f', 'd'])
Out[149]: 
e    0.005240
b   -1.069046
f         NaN
d   -0.944778
dtype: float64

Here, the f label was not contained in the Series and hence appears as NaN in the result.

With a DataFrame, you can simultaneously reindex the index and columns:

In [150]: df
Out[150]: 
        one     three       two
a -0.701368       NaN -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [151]: df.reindex(index=['c', 'f', 'b'], columns=['three', 'two', 'one'])
Out[151]: 
      three       two       one
c -0.148387 -0.002666 -0.231617
f       NaN       NaN       NaN
b -0.354359  0.637674  0.109333

For convenience, you may utilize the reindex_axis method, which takes the labels and a keyword axis parameter.

Note that the Index objects containing the actual axis labels can be shared between objects. So if we have a Series and a DataFrame, the following can be done:

In [152]: rs = s.reindex(df.index)

In [153]: rs
Out[153]: 
a    1.112686
b   -1.069046
c   -1.218080
d   -0.944778
dtype: float64

In [154]: rs.index is df.index
Out[154]: True

This means that the reindexed Series’s index is the same Python object as the DataFrame’s index.

See also

Advanced indexing is an even more concise way of doing reindexing.

Note

When writing performance-sensitive code, there is a good reason to spend some time becoming a reindexing ninja: many operations are faster on pre-aligned data. Adding two unaligned DataFrames internally triggers a reindexing step. For exploratory analysis you will hardly notice the difference (because reindex has been heavily optimized), but when CPU cycles matter sprinkling a few explicit reindex calls here and there can have an impact.

Reindexing to align with another object

You may wish to take an object and reindex its axes to be labeled the same as another object. While the syntax for this is straightforward albeit verbose, it is a common enough operation that the reindex_like method is available to make this simpler:

In [155]: df2
Out[155]: 
        one       two
a -0.701368 -0.087103
b  0.109333  0.637674
c -0.231617 -0.002666

In [156]: df3
Out[156]: 
        one       two
a -0.426817 -0.269738
b  0.383883  0.455039
c  0.042934 -0.185301

In [157]: df.reindex_like(df2)
Out[157]: 
        one       two
a -0.701368 -0.087103
b  0.109333  0.637674
c -0.231617 -0.002666

Reindexing with reindex_axis

Aligning objects with each other with align

The align method is the fastest way to simultaneously align two objects. It supports a join argument (related to joining and merging):

  • join='outer': take the union of the indexes
  • join='left': use the calling object’s index
  • join='right': use the passed object’s index
  • join='inner': intersect the indexes

It returns a tuple with both of the reindexed Series:

In [158]: s = Series(randn(5), index=['a', 'b', 'c', 'd', 'e'])

In [159]: s1 = s[:4]

In [160]: s2 = s[1:]

In [161]: s1.align(s2)
Out[161]: 
(a    0.479090
 b    0.686579
 c   -0.949750
 d   -0.257472
 e         NaN
 dtype: float64, a         NaN
 b    0.686579
 c   -0.949750
 d   -0.257472
 e   -0.568459
 dtype: float64)

In [162]: s1.align(s2, join='inner')
Out[162]: 
(b    0.686579
 c   -0.949750
 d   -0.257472
 dtype: float64, b    0.686579
 c   -0.949750
 d   -0.257472
 dtype: float64)

In [163]: s1.align(s2, join='left')
Out[163]: 
(a    0.479090
 b    0.686579
 c   -0.949750
 d   -0.257472
 dtype: float64, a         NaN
 b    0.686579
 c   -0.949750
 d   -0.257472
 dtype: float64)

For DataFrames, the join method will be applied to both the index and the columns by default:

In [164]: df.align(df2, join='inner')
Out[164]: 
(        one       two
 a -0.701368 -0.087103
 b  0.109333  0.637674
 c -0.231617 -0.002666,         one       two
 a -0.701368 -0.087103
 b  0.109333  0.637674
 c -0.231617 -0.002666)

You can also pass an axis option to only align on the specified axis:

In [165]: df.align(df2, join='inner', axis=0)
Out[165]: 
(        one     three       two
 a -0.701368       NaN -0.087103
 b  0.109333 -0.354359  0.637674
 c -0.231617 -0.148387 -0.002666,         one       two
 a -0.701368 -0.087103
 b  0.109333  0.637674
 c -0.231617 -0.002666)

If you pass a Series to DataFrame.align, you can choose to align both objects either on the DataFrame’s index or columns using the axis argument:

In [166]: df.align(df2.ix[0], axis=1)
Out[166]: 
(        one     three       two
 a -0.701368       NaN -0.087103
 b  0.109333 -0.354359  0.637674
 c -0.231617 -0.148387 -0.002666
 d       NaN -0.167407  0.104044, one     -0.701368
 three         NaN
 two     -0.087103
 Name: a, dtype: float64)

Filling while reindexing

reindex takes an optional parameter method which is a filling method chosen from the following table:

Method Action
pad / ffill Fill values forward
bfill / backfill Fill values backward

Other fill methods could be added, of course, but these are the two most commonly used for time series data. In a way they only make sense for time series or otherwise ordered data, but you may have an application on non-time series data where this sort of “interpolation” logic is the correct thing to do. More sophisticated interpolation of missing values would be an obvious extension.

We illustrate these fill methods on a simple TimeSeries:

In [167]: rng = date_range('1/3/2000', periods=8)

In [168]: ts = Series(randn(8), index=rng)

In [169]: ts2 = ts[[0, 3, 6]]

In [170]: ts
Out[170]: 
2000-01-03   -0.059786
2000-01-04    0.936271
2000-01-05    0.040623
2000-01-06    0.836517
2000-01-07    1.849649
2000-01-08   -1.198994
2000-01-09    0.688500
2000-01-10   -0.696903
Freq: D, dtype: float64

In [171]: ts2
Out[171]: 
2000-01-03   -0.059786
2000-01-06    0.836517
2000-01-09    0.688500
dtype: float64

In [172]: ts2.reindex(ts.index)
Out[172]: 
2000-01-03   -0.059786
2000-01-04         NaN
2000-01-05         NaN
2000-01-06    0.836517
2000-01-07         NaN
2000-01-08         NaN
2000-01-09    0.688500
2000-01-10         NaN
Freq: D, dtype: float64

In [173]: ts2.reindex(ts.index, method='ffill')
Out[173]: 
2000-01-03   -0.059786
2000-01-04   -0.059786
2000-01-05   -0.059786
2000-01-06    0.836517
2000-01-07    0.836517
2000-01-08    0.836517
2000-01-09    0.688500
2000-01-10    0.688500
Freq: D, dtype: float64

In [174]: ts2.reindex(ts.index, method='bfill')
Out[174]: 
2000-01-03   -0.059786
2000-01-04    0.836517
2000-01-05    0.836517
2000-01-06    0.836517
2000-01-07    0.688500
2000-01-08    0.688500
2000-01-09    0.688500
2000-01-10         NaN
Freq: D, dtype: float64

Note these methods require that the indexes are order increasing.

Note the same result could have been achieved using fillna:

In [175]: ts2.reindex(ts.index).fillna(method='ffill')
Out[175]: 
2000-01-03   -0.059786
2000-01-04   -0.059786
2000-01-05   -0.059786
2000-01-06    0.836517
2000-01-07    0.836517
2000-01-08    0.836517
2000-01-09    0.688500
2000-01-10    0.688500
Freq: D, dtype: float64

Note that reindex will raise a ValueError if the index is not monotonic. fillna will not make any checks on the order of the index.

Dropping labels from an axis

A method closely related to reindex is the drop function. It removes a set of labels from an axis:

In [176]: df
Out[176]: 
        one     three       two
a -0.701368       NaN -0.087103
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666
d       NaN -0.167407  0.104044

In [177]: df.drop(['a', 'd'], axis=0)
Out[177]: 
        one     three       two
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666

In [178]: df.drop(['one'], axis=1)
Out[178]: 
      three       two
a       NaN -0.087103
b -0.354359  0.637674
c -0.148387 -0.002666
d -0.167407  0.104044

Note that the following also works, but is a bit less obvious / clean:

In [179]: df.reindex(df.index - ['a', 'd'])
Out[179]: 
        one     three       two
b  0.109333 -0.354359  0.637674
c -0.231617 -0.148387 -0.002666

Renaming / mapping labels

The rename method allows you to relabel an axis based on some mapping (a dict or Series) or an arbitrary function.

In [180]: s
Out[180]: 
a    0.479090
b    0.686579
c   -0.949750
d   -0.257472
e   -0.568459
dtype: float64

In [181]: s.rename(str.upper)
Out[181]: 
A    0.479090
B    0.686579
C   -0.949750
D   -0.257472
E   -0.568459
dtype: float64

If you pass a function, it must return a value when called with any of the labels (and must produce a set of unique values). But if you pass a dict or Series, it need only contain a subset of the labels as keys:

In [182]: df.rename(columns={'one' : 'foo', 'two' : 'bar'},
   .....:           index={'a' : 'apple', 'b' : 'banana', 'd' : 'durian'})
   .....: 
Out[182]: 
             foo     three       bar
apple  -0.701368       NaN -0.087103
banana  0.109333 -0.354359  0.637674
c      -0.231617 -0.148387 -0.002666
durian       NaN -0.167407  0.104044

The rename method also provides an inplace named parameter that is by default False and copies the underlying data. Pass inplace=True to rename the data in place.

The Panel class has a related rename_axis class which can rename any of its three axes.

Iteration

Because Series is array-like, basic iteration produces the values. Other data structures follow the dict-like convention of iterating over the “keys” of the objects. In short:

  • Series: values
  • DataFrame: column labels
  • Panel: item labels

Thus, for example:

In [183]: for col in df:
   .....:     print(col)
   .....: 
one
three
two

iteritems

Consistent with the dict-like interface, iteritems iterates through key-value pairs:

  • Series: (index, scalar value) pairs
  • DataFrame: (column, Series) pairs
  • Panel: (item, DataFrame) pairs

For example:

In [184]: for item, frame in wp.iteritems():
   .....:     print(item)
   .....:     print(frame)
   .....: 
Item1
                   A         B         C         D
2000-01-01 -1.118121  0.431279  0.554724 -1.333649
2000-01-02 -0.332174 -0.485882  1.725945  1.799276
2000-01-03 -0.968916 -0.779465 -2.000701 -1.866630
2000-01-04 -1.101268  1.957478  0.058889  0.758071
2000-01-05  0.076612 -0.548502 -0.160485 -0.377780
Item2
                   A         B         C         D
2000-01-01  0.249911 -0.341270 -0.272599 -0.277446
2000-01-02 -1.102896  0.100307 -1.602814  0.920139
2000-01-03 -0.643870  0.060336 -0.434942 -0.494305
2000-01-04  0.737973  0.451632  0.334124 -0.787062
2000-01-05  0.651396 -0.741919  1.193881 -2.395763

iterrows

New in v0.7 is the ability to iterate efficiently through rows of a DataFrame. It returns an iterator yielding each index value along with a Series containing the data in each row:

In [185]: for row_index, row in df2.iterrows():
   .....:     print('%s\n%s' % (row_index, row))
   .....: 
a
one   -0.701368
two   -0.087103
Name: a, dtype: float64
b
one    0.109333
two    0.637674
Name: b, dtype: float64
c
one   -0.231617
two   -0.002666
Name: c, dtype: float64

For instance, a contrived way to transpose the DataFrame would be:

In [186]: df2 = DataFrame({'x': [1, 2, 3], 'y': [4, 5, 6]})

In [187]: print(df2)
   x  y
0  1  4
1  2  5
2  3  6

In [188]: print(df2.T)
   0  1  2
x  1  2  3
y  4  5  6

In [189]: df2_t = DataFrame(dict((idx,values) for idx, values in df2.iterrows()))

In [190]: print(df2_t)
   0  1  2
x  1  2  3
y  4  5  6

Note

iterrows does not preserve dtypes across the rows (dtypes are preserved across columns for DataFrames). For example,

In [191]: df_iter = DataFrame([[1, 1.0]], columns=['x', 'y'])

In [192]: row = next(df_iter.iterrows())[1]

In [193]: print(row['x'].dtype)
float64

In [194]: print(df_iter['x'].dtype)
int64

itertuples

This method will return an iterator yielding a tuple for each row in the DataFrame. The first element of the tuple will be the row’s corresponding index value, while the remaining values are the row values proper.

For instance,

In [195]: for r in df2.itertuples():
   .....:     print(r)
   .....: 
(0, 1, 4)
(1, 2, 5)
(2, 3, 6)

Vectorized string methods

Series is equipped (as of pandas 0.8.1) with a set of string processing methods that make it easy to operate on each element of the array. Perhaps most importantly, these methods exclude missing/NA values automatically. These are accessed via the Series’s str attribute and generally have names matching the equivalent (scalar) build-in string methods:

Splitting and Replacing Strings

In [196]: s = Series(['A', 'B', 'C', 'Aaba', 'Baca', np.nan, 'CABA', 'dog', 'cat'])

In [197]: s.str.lower()
Out[197]: 
0       a
1       b
2       c
3    aaba
4    baca
5     NaN
6    caba
7     dog
8     cat
dtype: object

In [198]: s.str.upper()
Out[198]: 
0       A
1       B
2       C
3    AABA
4    BACA
5     NaN
6    CABA
7     DOG
8     CAT
dtype: object

In [199]: s.str.len()
Out[199]: 
0     1
1     1
2     1
3     4
4     4
5   NaN
6     4
7     3
8     3
dtype: float64

Methods like split return a Series of lists:

In [200]: s2 = Series(['a_b_c', 'c_d_e', np.nan, 'f_g_h'])

In [201]: s2.str.split('_')
Out[201]: 
0    [a, b, c]
1    [c, d, e]
2          NaN
3    [f, g, h]
dtype: object

Elements in the split lists can be accessed using get or [] notation:

In [202]: s2.str.split('_').str.get(1)
Out[202]: 
0      b
1      d
2    NaN
3      g
dtype: object

In [203]: s2.str.split('_').str[1]
Out[203]: 
0      b
1      d
2    NaN
3      g
dtype: object

Methods like replace and findall take regular expressions, too:

In [204]: s3 = Series(['A', 'B', 'C', 'Aaba', 'Baca',
   .....:             '', np.nan, 'CABA', 'dog', 'cat'])
   .....: 

In [205]: s3
Out[205]: 
0       A
1       B
2       C
3    Aaba
4    Baca
5        
6     NaN
7    CABA
8     dog
9     cat
dtype: object

In [206]: s3.str.replace('^.a|dog', 'XX-XX ', case=False)
Out[206]: 
0           A
1           B
2           C
3    XX-XX ba
4    XX-XX ca
5            
6         NaN
7    XX-XX BA
8      XX-XX 
9     XX-XX t
dtype: object

Extracting Substrings

The method extract (introduced in version 0.13) accepts regular expressions with match groups. Extracting a regular expression with one group returns a Series of strings.

In [207]: Series(['a1', 'b2', 'c3']).str.extract('[ab](\d)')
Out[207]: 
0      1
1      2
2    NaN
dtype: object

Elements that do not match return NaN. Extracting a regular expression with more than one group returns a DataFrame with one column per group.

In [208]: Series(['a1', 'b2', 'c3']).str.extract('([ab])(\d)')
Out[208]: 
     0    1
0    a    1
1    b    2
2  NaN  NaN

Elements that do not match return a row filled with NaN. Thus, a Series of messy strings can be “converted” into a like-indexed Series or DataFrame of cleaned-up or more useful strings, without necessitating get() to access tuples or re.match objects.

The results dtype always is object, even if no match is found and the result only contains NaN.

Named groups like

In [209]: Series(['a1', 'b2', 'c3']).str.extract('(?P<letter>[ab])(?P<digit>\d)')
Out[209]: 
  letter digit
0      a     1
1      b     2
2    NaN   NaN

and optional groups like

In [210]: Series(['a1', 'b2', '3']).str.extract('(?P<letter>[ab])?(?P<digit>\d)')
Out[210]: 
  letter digit
0      a     1
1      b     2
2    NaN     3

can also be used.

Testing for Strings that Match or Contain a Pattern

You can check whether elements contain a pattern:

In [211]: pattern = r'[a-z][0-9]'

In [212]: Series(['1', '2', '3a', '3b', '03c']).str.contains(pattern)
Out[212]: 
0    False
1    False
2    False
3    False
4    False
dtype: bool

or match a pattern:

In [213]: Series(['1', '2', '3a', '3b', '03c']).str.match(pattern, as_indexer=True)
Out[213]: 
0    False
1    False
2    False
3    False
4    False
dtype: bool

The distinction between match and contains is strictness: match relies on strict re.match, while contains relies on re.search.

Warning

In previous versions, match was for extracting groups, returning a not-so-convenient Series of tuples. The new method extract (described in the previous section) is now preferred.

This old, deprecated behavior of match is still the default. As demonstrated above, use the new behavior by setting as_indexer=True. In this mode, match is analogous to contains, returning a boolean Series. The new behavior will become the default behavior in a future release.

Methods like match, contains, startswith, and endswith take
an extra na argument so missing values can be considered True or False:
In [214]: s4 = Series(['A', 'B', 'C', 'Aaba', 'Baca', np.nan, 'CABA', 'dog', 'cat'])

In [215]: s4.str.contains('A', na=False)
Out[215]: 
0     True
1    False
2    False
3     True
4    False
5    False
6     True
7    False
8    False
dtype: bool
Method Description
cat Concatenate strings
split Split strings on delimiter
get Index into each element (retrieve i-th element)
join Join strings in each element of the Series with passed separator
contains Return boolean array if each string contains pattern/regex
replace Replace occurrences of pattern/regex with some other string
repeat Duplicate values (s.str.repeat(3) equivalent to x * 3)
pad Add whitespace to left, right, or both sides of strings
center Equivalent to pad(side='both')
wrap Split long strings into lines with length less than a given width
slice Slice each string in the Series
slice_replace Replace slice in each string with passed value
count Count occurrences of pattern
startswith Equivalent to str.startswith(pat) for each element
endswith Equivalent to str.endswith(pat) for each element
findall Compute list of all occurrences of pattern/regex for each string
match Call re.match on each element, returning matched groups as list
extract Call re.match on each element, as match does, but return matched groups as strings for convenience.
len Compute string lengths
strip Equivalent to str.strip
rstrip Equivalent to str.rstrip
lstrip Equivalent to str.lstrip
lower Equivalent to str.lower
upper Equivalent to str.upper

Getting indicator variables from seperated strings

You can extract dummy variables from string columns. For example if they are seperated by a '|':

In [216]: s = pd.Series(['a', 'a|b', np.nan, 'a|c'])

In [217]: s.str.get_dummies(sep='|')
Out[217]: 
   a  b  c
0  1  0  0
1  1  1  0
2  0  0  0
3  1  0  1

See also get_dummies().

Sorting by index and value

There are two obvious kinds of sorting that you may be interested in: sorting by label and sorting by actual values. The primary method for sorting axis labels (indexes) across data structures is the sort_index method.

In [218]: unsorted_df = df.reindex(index=['a', 'd', 'c', 'b'],
   .....:                          columns=['three', 'two', 'one'])
   .....: 

In [219]: unsorted_df.sort_index()
Out[219]: 
      three       two       one
a       NaN -0.087103 -0.701368
b -0.354359  0.637674  0.109333
c -0.148387 -0.002666 -0.231617
d -0.167407  0.104044       NaN

In [220]: unsorted_df.sort_index(ascending=False)
Out[220]: 
      three       two       one
d -0.167407  0.104044       NaN
c -0.148387 -0.002666 -0.231617
b -0.354359  0.637674  0.109333
a       NaN -0.087103 -0.701368

In [221]: unsorted_df.sort_index(axis=1)
Out[221]: 
        one     three       two
a -0.701368       NaN -0.087103
d       NaN -0.167407  0.104044
c -0.231617 -0.148387 -0.002666
b  0.109333 -0.354359  0.637674

DataFrame.sort_index can accept an optional by argument for axis=0 which will use an arbitrary vector or a column name of the DataFrame to determine the sort order:

In [222]: df1 = DataFrame({'one':[2,1,1,1],'two':[1,3,2,4],'three':[5,4,3,2]})

In [223]: df1.sort_index(by='two')
Out[223]: 
   one  three  two
0    2      5    1
2    1      3    2
1    1      4    3
3    1      2    4

The by argument can take a list of column names, e.g.:

In [224]: df1[['one', 'two', 'three']].sort_index(by=['one','two'])
Out[224]: 
   one  two  three
2    1    2      3
1    1    3      4
3    1    4      2
0    2    1      5

Series has the method order (analogous to R’s order function) which sorts by value, with special treatment of NA values via the na_position argument:

In [225]: s[2] = np.nan

In [226]: s.order()
Out[226]: 
0      a
1    a|b
3    a|c
2    NaN
dtype: object

In [227]: s.order(na_position='first')
Out[227]: 
2    NaN
0      a
1    a|b
3    a|c
dtype: object

Note

Series.sort sorts a Series by value in-place. This is to provide compatibility with NumPy methods which expect the ndarray.sort behavior. Series.order returns a copy of the sorted data.

smallest / largest values

New in version 0.14.0.

Series has the nsmallest and nlargest methods which return the smallest or largest n values. For a large Series this can be much faster than sorting the entire Series and calling head(n) on the result.

In [228]: s = Series(np.random.permutation(10))

In [229]: s
Out[229]: 
0    6
1    2
2    7
3    3
4    9
5    4
6    8
7    0
8    1
9    5
dtype: int32

In [230]: s.order()
Out[230]: 
7    0
8    1
1    2
3    3
5    4
9    5
0    6
2    7
6    8
4    9
dtype: int32

In [231]: s.nsmallest(3)
Out[231]: 
7    0
8    1
1    2
dtype: int32

In [232]: s.nlargest(3)
Out[232]: 
4    9
6    8
2    7
dtype: int32

Sorting by a multi-index column

You must be explicit about sorting when the column is a multi-index, and fully specify all levels to by.

In [233]: df1.columns = MultiIndex.from_tuples([('a','one'),('a','two'),('b','three')])

In [234]: df1.sort_index(by=('a','two'))
Out[234]: 
     a           b
   one  two  three
3    1    2      4
2    1    3      2
1    1    4      3
0    2    5      1

Copying

The copy method on pandas objects copies the underlying data (though not the axis indexes, since they are immutable) and returns a new object. Note that it is seldom necessary to copy objects. For example, there are only a handful of ways to alter a DataFrame in-place:

  • Inserting, deleting, or modifying a column
  • Assigning to the index or columns attributes
  • For homogeneous data, directly modifying the values via the values attribute or advanced indexing

To be clear, no pandas methods have the side effect of modifying your data; almost all methods return new objects, leaving the original object untouched. If data is modified, it is because you did so explicitly.

dtypes

The main types stored in pandas objects are float, int, bool, datetime64[ns], timedelta[ns], and object. In addition these dtypes have item sizes, e.g. int64 and int32. A convenient dtypes attribute for DataFrames returns a Series with the data type of each column.

In [235]: dft = DataFrame(dict( A = np.random.rand(3),
   .....:                       B = 1,
   .....:                       C = 'foo',
   .....:                       D = Timestamp('20010102'),
   .....:                       E = Series([1.0]*3).astype('float32'),
   .....:                       F = False,
   .....:                       G = Series([1]*3,dtype='int8')))
   .....: 

In [236]: dft
Out[236]: 
          A  B    C          D  E      F  G
0  0.193366  1  foo 2001-01-02  1  False  1
1  0.013428  1  foo 2001-01-02  1  False  1
2  0.347430  1  foo 2001-01-02  1  False  1

In [237]: dft.dtypes
Out[237]: 
A           float64
B             int64
C            object
D    datetime64[ns]
E           float32
F              bool
G              int8
dtype: object

On a Series use the dtype method.

In [238]: dft['A'].dtype
Out[238]: dtype('float64')

If a pandas object contains data multiple dtypes IN A SINGLE COLUMN, the dtype of the column will be chosen to accommodate all of the data types (object is the most general).

# these ints are coerced to floats
In [239]: Series([1, 2, 3, 4, 5, 6.])
Out[239]: 
0    1
1    2
2    3
3    4
4    5
5    6
dtype: float64

# string data forces an ``object`` dtype
In [240]: Series([1, 2, 3, 6., 'foo'])
Out[240]: 
0      1
1      2
2      3
3      6
4    foo
dtype: object

The method get_dtype_counts will return the number of columns of each type in a DataFrame:

In [241]: dft.get_dtype_counts()
Out[241]: 
bool              1
datetime64[ns]    1
float32           1
float64           1
int64             1
int8              1
object            1
dtype: int64

Numeric dtypes will propagate and can coexist in DataFrames (starting in v0.11.0). If a dtype is passed (either directly via the dtype keyword, a passed ndarray, or a passed Series, then it will be preserved in DataFrame operations. Furthermore, different numeric dtypes will NOT be combined. The following example will give you a taste.

In [242]: df1 = DataFrame(randn(8, 1), columns = ['A'], dtype = 'float32')

In [243]: df1
Out[243]: 
          A
0  1.111528
1 -1.805497
2 -0.125340
3  2.055101
4  0.170350
5 -1.551268
6 -0.503071
7  0.370166

In [244]: df1.dtypes
Out[244]: 
A    float32
dtype: object

In [245]: df2 = DataFrame(dict( A = Series(randn(8),dtype='float16'),
   .....:                       B = Series(randn(8)),
   .....:                       C = Series(np.array(randn(8),dtype='uint8')) ))
   .....: 

In [246]: df2
Out[246]: 
          A         B    C
0  2.220703  0.447712    0
1  0.589355  0.429500    0
2  1.896484 -1.947809  255
3 -0.916992 -0.046360    0
4  0.614746  0.044316    0
5 -0.392578  0.234849    2
6  0.604004 -0.622669    0
7 -0.061737 -0.351207    0

In [247]: df2.dtypes
Out[247]: 
A    float16
B    float64
C      uint8
dtype: object

defaults

By default integer types are int64 and float types are float64, REGARDLESS of platform (32-bit or 64-bit). The following will all result in int64 dtypes.

In [248]: DataFrame([1, 2], columns=['a']).dtypes
Out[248]: 
a    int64
dtype: object

In [249]: DataFrame({'a': [1, 2]}).dtypes
Out[249]: 
a    int64
dtype: object

In [250]: DataFrame({'a': 1 }, index=list(range(2))).dtypes
Out[250]: 
a    int64
dtype: object

Numpy, however will choose platform-dependent types when creating arrays. The following WILL result in int32 on 32-bit platform.

In [251]: frame = DataFrame(np.array([1, 2]))

upcasting

Types can potentially be upcasted when combined with other types, meaning they are promoted from the current type (say int to float)

In [252]: df3 = df1.reindex_like(df2).fillna(value=0.0) + df2

In [253]: df3
Out[253]: 
          A         B    C
0  3.332231  0.447712    0
1 -1.216141  0.429500    0
2  1.771144 -1.947809  255
3  1.138109 -0.046360    0
4  0.785096  0.044316    0
5 -1.943846  0.234849    2
6  0.100933 -0.622669    0
7  0.308429 -0.351207    0

In [254]: df3.dtypes
Out[254]: 
A    float32
B    float64
C    float64
dtype: object

The values attribute on a DataFrame return the lower-common-denominator of the dtypes, meaning the dtype that can accommodate ALL of the types in the resulting homogenous dtyped numpy array. This can force some upcasting.

In [255]: df3.values.dtype
Out[255]: dtype('float64')

astype

You can use the astype method to explicitly convert dtypes from one to another. These will by default return a copy, even if the dtype was unchanged (pass copy=False to change this behavior). In addition, they will raise an exception if the astype operation is invalid.

Upcasting is always according to the numpy rules. If two different dtypes are involved in an operation, then the more general one will be used as the result of the operation.

In [256]: df3
Out[256]: 
          A         B    C
0  3.332231  0.447712    0
1 -1.216141  0.429500    0
2  1.771144 -1.947809  255
3  1.138109 -0.046360    0
4  0.785096  0.044316    0
5 -1.943846  0.234849    2
6  0.100933 -0.622669    0
7  0.308429 -0.351207    0

In [257]: df3.dtypes
Out[257]: 
A    float32
B    float64
C    float64
dtype: object

# conversion of dtypes
In [258]: df3.astype('float32').dtypes
Out[258]: 
A    float32
B    float32
C    float32
dtype: object

object conversion

convert_objects is a method to try to force conversion of types from the object dtype to other types. To force conversion of specific types that are number like, e.g. could be a string that represents a number, pass convert_numeric=True. This will force strings and numbers alike to be numbers if possible, otherwise they will be set to np.nan.

In [259]: df3['D'] = '1.'

In [260]: df3['E'] = '1'

In [261]: df3.convert_objects(convert_numeric=True).dtypes
Out[261]: 
A    float32
B    float64
C    float64
D    float64
E      int64
dtype: object

# same, but specific dtype conversion
In [262]: df3['D'] = df3['D'].astype('float16')

In [263]: df3['E'] = df3['E'].astype('int32')

In [264]: df3.dtypes
Out[264]: 
A    float32
B    float64
C    float64
D    float16
E      int32
dtype: object

To force conversion to datetime64[ns], pass convert_dates='coerce'. This will convert any datetime-like object to dates, forcing other values to NaT. This might be useful if you are reading in data which is mostly dates, but occasionally has non-dates intermixed and you want to represent as missing.

In [265]: s = Series([datetime(2001,1,1,0,0),
   .....:            'foo', 1.0, 1, Timestamp('20010104'),
   .....:            '20010105'],dtype='O')
   .....: 

In [266]: s
Out[266]: 
0    2001-01-01 00:00:00
1                    foo
2                      1
3                      1
4    2001-01-04 00:00:00
5               20010105
dtype: object

In [267]: s.convert_objects(convert_dates='coerce')
Out[267]: 
0   2001-01-01
1          NaT
2          NaT
3          NaT
4   2001-01-04
5   2001-01-05
dtype: datetime64[ns]

In addition, convert_objects will attempt the soft conversion of any object dtypes, meaning that if all the objects in a Series are of the same type, the Series will have that dtype.

gotchas

Performing selection operations on integer type data can easily upcast the data to floating. The dtype of the input data will be preserved in cases where nans are not introduced (starting in 0.11.0) See also integer na gotchas

In [268]: dfi = df3.astype('int32')

In [269]: dfi['E'] = 1

In [270]: dfi
Out[270]: 
   A  B    C  D  E
0  3  0    0  1  1
1 -1  0    0  1  1
2  1 -1  255  1  1
3  1  0    0  1  1
4  0  0    0  1  1
5 -1  0    2  1  1
6  0  0    0  1  1
7  0  0    0  1  1

In [271]: dfi.dtypes
Out[271]: 
A    int32
B    int32
C    int32
D    int32
E    int64
dtype: object

In [272]: casted = dfi[dfi>0]

In [273]: casted
Out[273]: 
    A   B    C  D  E
0   3 NaN  NaN  1  1
1 NaN NaN  NaN  1  1
2   1 NaN  255  1  1
3   1 NaN  NaN  1  1
4 NaN NaN  NaN  1  1
5 NaN NaN    2  1  1
6 NaN NaN  NaN  1  1
7 NaN NaN  NaN  1  1

In [274]: casted.dtypes
Out[274]: 
A    float64
B    float64
C    float64
D      int32
E      int64
dtype: object

While float dtypes are unchanged.

In [275]: dfa = df3.copy()

In [276]: dfa['A'] = dfa['A'].astype('float32')

In [277]: dfa.dtypes
Out[277]: 
A    float32
B    float64
C    float64
D    float16
E      int32
dtype: object

In [278]: casted = dfa[df2>0]

In [279]: casted
Out[279]: 
          A         B    C   D   E
0  3.332231  0.447712  NaN NaN NaN
1 -1.216141  0.429500  NaN NaN NaN
2  1.771144       NaN  255 NaN NaN
3       NaN       NaN  NaN NaN NaN
4  0.785096  0.044316  NaN NaN NaN
5       NaN  0.234849    2 NaN NaN
6  0.100933       NaN  NaN NaN NaN
7       NaN       NaN  NaN NaN NaN

In [280]: casted.dtypes
Out[280]: 
A    float32
B    float64
C    float64
D    float16
E    float64
dtype: object

Selecting columns based on dtype

New in version 0.14.1.

The select_dtypes() method implements subsetting of columns based on their dtype.

First, let’s create a DataFrame with a slew of different dtypes:

In [281]: df = DataFrame({'string': list('abc'),
   .....:                 'int64': list(range(1, 4)),
   .....:                 'uint8': np.arange(3, 6).astype('u1'),
   .....:                 'float64': np.arange(4.0, 7.0),
   .....:                 'bool1': [True, False, True],
   .....:                 'bool2': [False, True, False],
   .....:                 'dates': pd.date_range('now', periods=3).values})
   .....: 

In [282]: df['tdeltas'] = df.dates.diff()

In [283]: df['uint64'] = np.arange(3, 6).astype('u8')

In [284]: df['other_dates'] = pd.date_range('20130101', periods=3).values

In [285]: df
Out[285]: 
   bool1  bool2               dates  float64  int64 string  uint8  tdeltas  \
0   True  False 2014-07-11 09:13:45        4      1      a      3      NaT   
1  False   True 2014-07-12 09:13:45        5      2      b      4   1 days   
2   True  False 2014-07-13 09:13:45        6      3      c      5   1 days   

   uint64 other_dates  
0       3  2013-01-01  
1       4  2013-01-02  
2       5  2013-01-03  

select_dtypes has two parameters include and exclude that allow you to say “give me the columns WITH these dtypes” (include) and/or “give the columns WITHOUT these dtypes” (exclude).

For example, to select bool columns

In [286]: df.select_dtypes(include=[bool])
Out[286]: 
   bool1  bool2
0   True  False
1  False   True
2   True  False

You can also pass the name of a dtype in the numpy dtype hierarchy:

In [287]: df.select_dtypes(include=['bool'])
Out[287]: 
   bool1  bool2
0   True  False
1  False   True
2   True  False

select_dtypes() also works with generic dtypes as well.

For example, to select all numeric and boolean columns while excluding unsigned integers

In [288]: df.select_dtypes(include=['number', 'bool'], exclude=['unsignedinteger'])
Out[288]: 
   bool1  bool2  float64  int64  tdeltas
0   True  False        4      1      NaT
1  False   True        5      2   1 days
2   True  False        6      3   1 days

To select string columns you must use the object dtype:

In [289]: df.select_dtypes(include=['object'])
Out[289]: 
  string
0      a
1      b
2      c

To see all the child dtypes of a generic dtype like numpy.number you can define a function that returns a tree of child dtypes:

In [290]: def subdtypes(dtype):
   .....:     subs = dtype.__subclasses__()
   .....:     if not subs:
   .....:         return dtype
   .....:     return [dtype, [subdtypes(dt) for dt in subs]]
   .....: 

All numpy dtypes are subclasses of numpy.generic:

In [291]: subdtypes(np.generic)
Out[291]: 
[numpy.generic,
 [[numpy.number,
   [[numpy.integer,
     [[numpy.signedinteger,
       [numpy.int8,
        numpy.int16,
        numpy.int32,
        numpy.int32,
        numpy.int64,
        numpy.timedelta64]],
      [numpy.unsignedinteger,
       [numpy.uint8,
        numpy.uint16,
        numpy.uint32,
        numpy.uint32,
        numpy.uint64]]]],
    [numpy.inexact,
     [[numpy.floating,
       [numpy.float16, numpy.float32, numpy.float64, numpy.float96]],
      [numpy.complexfloating,
       [numpy.complex64, numpy.complex128, numpy.complex192]]]]]],
  [numpy.flexible,
   [[numpy.character, [numpy.string_, numpy.unicode_]],
    [numpy.void, [numpy.core.records.record]]]],
  numpy.bool_,
  numpy.datetime64,
  numpy.object_]]

Note

The include and exclude parameters must be non-string sequences.