Working with missing data

In this section, we will discuss missing (also referred to as NA) values in pandas.

Note

The choice of using NaN internally to denote missing data was largely for simplicity and performance reasons. It differs from the MaskedArray approach of, for example, scikits.timeseries. We are hopeful that NumPy will soon be able to provide a native NA type solution (similar to R) performant enough to be used in pandas.

See the cookbook for some advanced strategies

Missing data basics

When / why does data become missing?

Some might quibble over our usage of missing. By “missing” we simply mean null or “not present for whatever reason”. Many data sets simply arrive with missing data, either because it exists and was not collected or it never existed. For example, in a collection of financial time series, some of the time series might start on different dates. Thus, values prior to the start date would generally be marked as missing.

In pandas, one of the most common ways that missing data is introduced into a data set is by reindexing. For example

In [1]: df = DataFrame(randn(5, 3), index=['a', 'c', 'e', 'f', 'h'],
   ...:                columns=['one', 'two', 'three'])
   ...: 

In [2]: df['four'] = 'bar'

In [3]: df['five'] = df['one'] > 0

In [4]: df
Out[4]: 
        one       two     three four   five
a -1.420361 -0.015601 -1.150641  bar  False
c -0.798334 -0.557697  0.381353  bar  False
e  1.337122 -1.531095  1.331458  bar   True
f -0.571329 -0.026671 -1.085663  bar  False
h -1.114738 -0.058216 -0.486768  bar  False

In [5]: df2 = df.reindex(['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h'])

In [6]: df2
Out[6]: 
        one       two     three four   five
a -1.420361 -0.015601 -1.150641  bar  False
b       NaN       NaN       NaN  NaN    NaN
c -0.798334 -0.557697  0.381353  bar  False
d       NaN       NaN       NaN  NaN    NaN
e  1.337122 -1.531095  1.331458  bar   True
f -0.571329 -0.026671 -1.085663  bar  False
g       NaN       NaN       NaN  NaN    NaN
h -1.114738 -0.058216 -0.486768  bar  False

Values considered “missing”

As data comes in many shapes and forms, pandas aims to be flexible with regard to handling missing data. While NaN is the default missing value marker for reasons of computational speed and convenience, we need to be able to easily detect this value with data of different types: floating point, integer, boolean, and general object. In many cases, however, the Python None will arise and we wish to also consider that “missing” or “null”.

Prior to version v0.10.0 inf and -inf were also considered to be “null” in computations. This is no longer the case by default; use the mode.use_inf_as_null option to recover it.

To make detecting missing values easier (and across different array dtypes), pandas provides the isnull() and notnull() functions, which are also methods on Series objects:

In [7]: df2['one']
Out[7]: 
a   -1.420361
b         NaN
c   -0.798334
d         NaN
e    1.337122
f   -0.571329
g         NaN
h   -1.114738
Name: one, dtype: float64

In [8]: isnull(df2['one'])
Out[8]: 
a    False
b     True
c    False
d     True
e    False
f    False
g     True
h    False
Name: one, dtype: bool

In [9]: df2['four'].notnull()
Out[9]: 
a     True
b    False
c     True
d    False
e     True
f     True
g    False
h     True
Name: four, dtype: bool

Summary: NaN and None (in object arrays) are considered missing by the isnull and notnull functions. inf and -inf are no longer considered missing by default.

Datetimes

For datetime64[ns] types, NaT represents missing values. This is a pseudo-native sentinel value that can be represented by numpy in a singular dtype (datetime64[ns]). pandas objects provide intercompatibility between NaT and NaN.

In [10]: df2 = df.copy()

In [11]: df2['timestamp'] = Timestamp('20120101')

In [12]: df2
Out[12]: 
        one       two     three four   five  timestamp
a -1.420361 -0.015601 -1.150641  bar  False 2012-01-01
c -0.798334 -0.557697  0.381353  bar  False 2012-01-01
e  1.337122 -1.531095  1.331458  bar   True 2012-01-01
f -0.571329 -0.026671 -1.085663  bar  False 2012-01-01
h -1.114738 -0.058216 -0.486768  bar  False 2012-01-01

In [13]: df2.ix[['a','c','h'],['one','timestamp']] = np.nan

In [14]: df2
Out[14]: 
        one       two     three four   five  timestamp
a       NaN -0.015601 -1.150641  bar  False        NaT
c       NaN -0.557697  0.381353  bar  False        NaT
e  1.337122 -1.531095  1.331458  bar   True 2012-01-01
f -0.571329 -0.026671 -1.085663  bar  False 2012-01-01
h       NaN -0.058216 -0.486768  bar  False        NaT

In [15]: df2.get_dtype_counts()
Out[15]: 
bool              1
datetime64[ns]    1
float64           3
object            1
dtype: int64

Inserting missing data

You can insert missing values by simply assigning to containers. The actual missing value used will be chosen based on the dtype.

For example, numeric containers will always use NaN regardless of the missing value type chosen:

In [16]: s = Series([1, 2, 3])

In [17]: s.loc[0] = None

In [18]: s
Out[18]: 
0   NaN
1     2
2     3
dtype: float64

Likewise, datetime containers will always use NaT.

For object containers, pandas will use the value given:

In [19]: s = Series(["a", "b", "c"])

In [20]: s.loc[0] = None

In [21]: s.loc[1] = np.nan

In [22]: s
Out[22]: 
0    None
1     NaN
2       c
dtype: object

Calculations with missing data

Missing values propagate naturally through arithmetic operations between pandas objects.

In [23]: a
Out[23]: 
        one       two
a       NaN -0.015601
c       NaN -0.557697
e  1.337122 -1.531095
f -0.571329 -0.026671
h -0.571329 -0.058216

In [24]: b
Out[24]: 
        one       two     three
a       NaN -0.015601 -1.150641
c       NaN -0.557697  0.381353
e  1.337122 -1.531095  1.331458
f -0.571329 -0.026671 -1.085663
h       NaN -0.058216 -0.486768

In [25]: a + b
Out[25]: 
        one  three       two
a       NaN    NaN -0.031202
c       NaN    NaN -1.115393
e  2.674243    NaN -3.062190
f -1.142658    NaN -0.053342
h       NaN    NaN -0.116432

The descriptive statistics and computational methods discussed in the data structure overview (and listed here and here) are all written to account for missing data. For example:

  • When summing data, NA (missing) values will be treated as zero
  • If the data are all NA, the result will be NA
  • Methods like cumsum and cumprod ignore NA values, but preserve them in the resulting arrays
In [26]: df
Out[26]: 
        one       two     three
a       NaN -0.015601 -1.150641
c       NaN -0.557697  0.381353
e  1.337122 -1.531095  1.331458
f -0.571329 -0.026671 -1.085663
h       NaN -0.058216 -0.486768

In [27]: df['one'].sum()
Out[27]: 0.76579267910953364

In [28]: df.mean(1)
Out[28]: 
a   -0.583121
c   -0.088172
e    0.379162
f   -0.561221
h   -0.272492
dtype: float64

In [29]: df.cumsum()
Out[29]: 
        one       two     three
a       NaN -0.015601 -1.150641
c       NaN -0.573297 -0.769288
e  1.337122 -2.104392  0.562171
f  0.765793 -2.131063 -0.523492
h       NaN -2.189279 -1.010260

NA values in GroupBy

NA groups in GroupBy are automatically excluded. This behavior is consistent with R, for example.

Cleaning / filling missing data

pandas objects are equipped with various data manipulation methods for dealing with missing data.

Filling missing values: fillna

The fillna function can “fill in” NA values with non-null data in a couple of ways, which we illustrate:

Replace NA with a scalar value

In [30]: df2
Out[30]: 
        one       two     three four   five  timestamp
a       NaN -0.015601 -1.150641  bar  False        NaT
c       NaN -0.557697  0.381353  bar  False        NaT
e  1.337122 -1.531095  1.331458  bar   True 2012-01-01
f -0.571329 -0.026671 -1.085663  bar  False 2012-01-01
h       NaN -0.058216 -0.486768  bar  False        NaT

In [31]: df2.fillna(0)
Out[31]: 
        one       two     three four   five  timestamp
a  0.000000 -0.015601 -1.150641  bar  False 1970-01-01
c  0.000000 -0.557697  0.381353  bar  False 1970-01-01
e  1.337122 -1.531095  1.331458  bar   True 2012-01-01
f -0.571329 -0.026671 -1.085663  bar  False 2012-01-01
h  0.000000 -0.058216 -0.486768  bar  False 1970-01-01

In [32]: df2['four'].fillna('missing')
Out[32]: 
a    bar
c    bar
e    bar
f    bar
h    bar
Name: four, dtype: object

Fill gaps forward or backward

Using the same filling arguments as reindexing, we can propagate non-null values forward or backward:

In [33]: df
Out[33]: 
        one       two     three
a       NaN -0.015601 -1.150641
c       NaN -0.557697  0.381353
e  1.337122 -1.531095  1.331458
f -0.571329 -0.026671 -1.085663
h       NaN -0.058216 -0.486768

In [34]: df.fillna(method='pad')
Out[34]: 
        one       two     three
a       NaN -0.015601 -1.150641
c       NaN -0.557697  0.381353
e  1.337122 -1.531095  1.331458
f -0.571329 -0.026671 -1.085663
h -0.571329 -0.058216 -0.486768

Limit the amount of filling

If we only want consecutive gaps filled up to a certain number of data points, we can use the limit keyword:

In [35]: df
Out[35]: 
   one       two     three
a  NaN -0.015601 -1.150641
c  NaN -0.557697  0.381353
e  NaN       NaN       NaN
f  NaN       NaN       NaN
h  NaN -0.058216 -0.486768

In [36]: df.fillna(method='pad', limit=1)
Out[36]: 
   one       two     three
a  NaN -0.015601 -1.150641
c  NaN -0.557697  0.381353
e  NaN -0.557697  0.381353
f  NaN       NaN       NaN
h  NaN -0.058216 -0.486768

To remind you, these are the available filling methods:

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

With time series data, using pad/ffill is extremely common so that the “last known value” is available at every time point.

The ffill() function is equivalent to fillna(method='ffill') and bfill() is equivalent to fillna(method='bfill')

Filling with a PandasObject

New in version 0.12.

You can also fillna using a dict or Series that is alignable. The labels of the dict or index of the Series must match the columns of the frame you wish to fill. The use case of this is to fill a DataFrame with the mean of that column.

In [37]: dff = DataFrame(np.random.randn(10,3),columns=list('ABC'))

In [38]: dff.iloc[3:5,0] = np.nan

In [39]: dff.iloc[4:6,1] = np.nan

In [40]: dff.iloc[5:8,2] = np.nan

In [41]: dff
Out[41]: 
          A         B         C
0  1.685148  0.112572 -1.495309
1  0.898435 -0.148217 -1.596070
2  0.159653  0.262136  0.036220
3       NaN -0.255069 -0.271020
4       NaN       NaN -1.165787
5  0.846974       NaN       NaN
6 -0.303961  0.625555       NaN
7  0.249698  1.103949       NaN
8  1.998044 -0.244548  0.136235
9  0.886313 -1.350722 -0.886348

In [42]: dff.fillna(dff.mean())
Out[42]: 
          A         B         C
0  1.685148  0.112572 -1.495309
1  0.898435 -0.148217 -1.596070
2  0.159653  0.262136  0.036220
3  0.802538 -0.255069 -0.271020
4  0.802538  0.013207 -1.165787
5  0.846974  0.013207 -0.748868
6 -0.303961  0.625555 -0.748868
7  0.249698  1.103949 -0.748868
8  1.998044 -0.244548  0.136235
9  0.886313 -1.350722 -0.886348

In [43]: dff.fillna(dff.mean()['B':'C'])
Out[43]: 
          A         B         C
0  1.685148  0.112572 -1.495309
1  0.898435 -0.148217 -1.596070
2  0.159653  0.262136  0.036220
3       NaN -0.255069 -0.271020
4       NaN  0.013207 -1.165787
5  0.846974  0.013207 -0.748868
6 -0.303961  0.625555 -0.748868
7  0.249698  1.103949 -0.748868
8  1.998044 -0.244548  0.136235
9  0.886313 -1.350722 -0.886348

New in version 0.13.

Same result as above, but is aligning the ‘fill’ value which is a Series in this case.

In [44]: dff.where(notnull(dff),dff.mean(),axis='columns')
Out[44]: 
          A         B         C
0  1.685148  0.112572 -1.495309
1  0.898435 -0.148217 -1.596070
2  0.159653  0.262136  0.036220
3  0.802538 -0.255069 -0.271020
4  0.802538  0.013207 -1.165787
5  0.846974  0.013207 -0.748868
6 -0.303961  0.625555 -0.748868
7  0.249698  1.103949 -0.748868
8  1.998044 -0.244548  0.136235
9  0.886313 -1.350722 -0.886348

Dropping axis labels with missing data: dropna

You may wish to simply exclude labels from a data set which refer to missing data. To do this, use the dropna method:

In [45]: df
Out[45]: 
   one       two     three
a  NaN -0.015601 -1.150641
c  NaN -0.557697  0.381353
e  NaN  0.000000  0.000000
f  NaN  0.000000  0.000000
h  NaN -0.058216 -0.486768

In [46]: df.dropna(axis=0)
Out[46]: 
Empty DataFrame
Columns: [one, two, three]
Index: []

In [47]: df.dropna(axis=1)
Out[47]: 
        two     three
a -0.015601 -1.150641
c -0.557697  0.381353
e  0.000000  0.000000
f  0.000000  0.000000
h -0.058216 -0.486768

In [48]: df['one'].dropna()
Out[48]: Series([], name: one, dtype: float64)

Series.dropna is a simpler method as it only has one axis to consider. DataFrame.dropna has considerably more options than Series.dropna, which can be examined in the API.

Interpolation

New in version 0.13.0.

Both Series and Dataframe objects have an interpolate method that, by default, performs linear interpolation at missing datapoints.

In [49]: ts
Out[49]: 
2000-01-31    0.469112
2000-02-29         NaN
2000-03-31         NaN
2000-04-28         NaN
2000-05-31         NaN
...
2007-11-30   -5.485119
2007-12-31   -6.854968
2008-01-31   -7.809176
2008-02-29   -6.346480
2008-03-31   -8.089641
2008-04-30   -8.916232
Freq: BM, Length: 100

In [50]: ts.count()
Out[50]: 61

In [51]: ts.interpolate().count()
Out[51]: 100

In [52]: plt.figure()
Out[52]: <matplotlib.figure.Figure at 0xa90763cc>

In [53]: ts.interpolate().plot()
Out[53]: <matplotlib.axes._subplots.AxesSubplot at 0xa8e8942c>
_images/series_interpolate.png

Index aware interpolation is available via the method keyword:

In [54]: ts2
Out[54]: 
2000-01-31    0.469112
2000-02-29         NaN
2002-07-31   -5.689738
2005-01-31         NaN
2008-04-30   -8.916232
dtype: float64

In [55]: ts2.interpolate()
Out[55]: 
2000-01-31    0.469112
2000-02-29   -2.610313
2002-07-31   -5.689738
2005-01-31   -7.302985
2008-04-30   -8.916232
dtype: float64

In [56]: ts2.interpolate(method='time')
Out[56]: 
2000-01-31    0.469112
2000-02-29    0.273272
2002-07-31   -5.689738
2005-01-31   -7.095568
2008-04-30   -8.916232
dtype: float64

For a floating-point index, use method='values':

In [57]: ser
Out[57]: 
0      0
1    NaN
10    10
dtype: float64

In [58]: ser.interpolate()
Out[58]: 
0      0
1      5
10    10
dtype: float64

In [59]: ser.interpolate(method='values')
Out[59]: 
0      0
1      1
10    10
dtype: float64

You can also interpolate with a DataFrame:

In [60]: df = DataFrame({'A': [1, 2.1, np.nan, 4.7, 5.6, 6.8],
   ....:                 'B': [.25, np.nan, np.nan, 4, 12.2, 14.4]})
   ....: 

In [61]: df
Out[61]: 
     A      B
0  1.0   0.25
1  2.1    NaN
2  NaN    NaN
3  4.7   4.00
4  5.6  12.20
5  6.8  14.40

In [62]: df.interpolate()
Out[62]: 
     A      B
0  1.0   0.25
1  2.1   1.50
2  3.4   2.75
3  4.7   4.00
4  5.6  12.20
5  6.8  14.40

The method argument gives access to fancier interpolation methods. If you have scipy installed, you can set pass the name of a 1-d interpolation routine to method. You’ll want to consult the full scipy interpolation documentation and reference guide for details. The appropriate interpolation method will depend on the type of data you are working with. For example, if you are dealing with a time series that is growing at an increasing rate, method='quadratic' may be appropriate. If you have values approximating a cumulative distribution function, then method='pchip' should work well.

Warning

These methods require scipy.

In [63]: df.interpolate(method='barycentric')
Out[63]: 
      A       B
0  1.00   0.250
1  2.10  -7.660
2  3.53  -4.515
3  4.70   4.000
4  5.60  12.200
5  6.80  14.400

In [64]: df.interpolate(method='pchip')
Out[64]: 
          A          B
0  1.000000   0.250000
1  2.100000   1.130135
2  3.429309   2.337586
3  4.700000   4.000000
4  5.600000  12.200000
5  6.800000  14.400000

When interpolating via a polynomial or spline approximation, you must also specify the degree or order of the approximation:

In [65]: df.interpolate(method='spline', order=2)
Out[65]: 
          A          B
0  1.000000   0.250000
1  2.100000  -0.428598
2  3.404545   1.206900
3  4.700000   4.000000
4  5.600000  12.200000
5  6.800000  14.400000

In [66]: df.interpolate(method='polynomial', order=2)
Out[66]: 
          A          B
0  1.000000   0.250000
1  2.100000  -4.161538
2  3.547059  -2.911538
3  4.700000   4.000000
4  5.600000  12.200000
5  6.800000  14.400000

Compare several methods:

In [67]: np.random.seed(2)

In [68]: ser = Series(np.arange(1, 10.1, .25)**2 + np.random.randn(37))

In [69]: bad = np.array([4, 13, 14, 15, 16, 17, 18, 20, 29])

In [70]: ser[bad] = np.nan

In [71]: methods = ['linear', 'quadratic', 'cubic']

In [72]: df = DataFrame({m: ser.interpolate(method=m) for m in methods})

In [73]: plt.figure()
Out[73]: <matplotlib.figure.Figure at 0xa8eadb4c>

In [74]: df.plot()
Out[74]: <matplotlib.axes._subplots.AxesSubplot at 0xa906faac>
_images/compare_interpolations.png

Another use case is interpolation at new values. Suppose you have 100 observations from some distribution. And let’s suppose that you’re particularly interested in what’s happening around the middle. You can mix pandas’ reindex and interpolate methods to interpolate at the new values.

In [75]: ser = Series(np.sort(np.random.uniform(size=100)))

# interpolate at new_index
In [76]: new_index = ser.index | Index([49.25, 49.5, 49.75, 50.25, 50.5, 50.75])

In [77]: interp_s = ser.reindex(new_index).interpolate(method='pchip')

In [78]: interp_s[49:51]
Out[78]: 
49.00    0.471410
49.25    0.476841
49.50    0.481780
49.75    0.485998
50.00    0.489266
50.25    0.491814
50.50    0.493995
50.75    0.495763
51.00    0.497074
dtype: float64

Like other pandas fill methods, interpolate accepts a limit keyword argument. Use this to limit the number of consecutive interpolations, keeping NaN values for interpolations that are too far from the last valid observation:

In [79]: ser = Series([1, 3, np.nan, np.nan, np.nan, 11])

In [80]: ser.interpolate(limit=2)
Out[80]: 
0     1
1     3
2     5
3     7
4   NaN
5    11
dtype: float64

Replacing Generic Values

Often times we want to replace arbitrary values with other values. New in v0.8 is the replace method in Series/DataFrame that provides an efficient yet flexible way to perform such replacements.

For a Series, you can replace a single value or a list of values by another value:

In [81]: ser = Series([0., 1., 2., 3., 4.])

In [82]: ser.replace(0, 5)
Out[82]: 
0    5
1    1
2    2
3    3
4    4
dtype: float64

You can replace a list of values by a list of other values:

In [83]: ser.replace([0, 1, 2, 3, 4], [4, 3, 2, 1, 0])
Out[83]: 
0    4
1    3
2    2
3    1
4    0
dtype: float64

You can also specify a mapping dict:

In [84]: ser.replace({0: 10, 1: 100})
Out[84]: 
0     10
1    100
2      2
3      3
4      4
dtype: float64

For a DataFrame, you can specify individual values by column:

In [85]: df = DataFrame({'a': [0, 1, 2, 3, 4], 'b': [5, 6, 7, 8, 9]})

In [86]: df.replace({'a': 0, 'b': 5}, 100)
Out[86]: 
     a    b
0  100  100
1    1    6
2    2    7
3    3    8
4    4    9

Instead of replacing with specified values, you can treat all given values as missing and interpolate over them:

In [87]: ser.replace([1, 2, 3], method='pad')
Out[87]: 
0    0
1    0
2    0
3    0
4    4
dtype: float64

String/Regular Expression Replacement

Note

Python strings prefixed with the r character such as r'hello world' are so-called “raw” strings. They have different semantics regarding backslashes than strings without this prefix. Backslashes in raw strings will be interpreted as an escaped backslash, e.g., r'\' == '\\'. You should read about them if this is unclear.

Replace the ‘.’ with nan (str -> str)

In [88]: d = {'a': list(range(4)), 'b': list('ab..'), 'c': ['a', 'b', nan, 'd']}

In [89]: df = DataFrame(d)

In [90]: df.replace('.', nan)
Out[90]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

Now do it with a regular expression that removes surrounding whitespace (regex -> regex)

In [91]: df.replace(r'\s*\.\s*', nan, regex=True)
Out[91]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

Replace a few different values (list -> list)

In [92]: df.replace(['a', '.'], ['b', nan])
Out[92]: 
   a    b    c
0  0    b    b
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

list of regex -> list of regex

In [93]: df.replace([r'\.', r'(a)'], ['dot', '\1stuff'], regex=True)
Out[93]: 
   a       b       c
0  0  stuff  stuff
1  1       b       b
2  2     dot     NaN
3  3     dot       d

Only search in column 'b' (dict -> dict)

In [94]: df.replace({'b': '.'}, {'b': nan})
Out[94]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

Same as the previous example, but use a regular expression for searching instead (dict of regex -> dict)

In [95]: df.replace({'b': r'\s*\.\s*'}, {'b': nan}, regex=True)
Out[95]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

You can pass nested dictionaries of regular expressions that use regex=True

In [96]: df.replace({'b': {'b': r''}}, regex=True)
Out[96]: 
   a  b    c
0  0  a    a
1  1       b
2  2  .  NaN
3  3  .    d

or you can pass the nested dictionary like so

In [97]: df.replace(regex={'b': {r'\s*\.\s*': nan}})
Out[97]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

You can also use the group of a regular expression match when replacing (dict of regex -> dict of regex), this works for lists as well

In [98]: df.replace({'b': r'\s*(\.)\s*'}, {'b': r'\1ty'}, regex=True)
Out[98]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  .ty  NaN
3  3  .ty    d

You can pass a list of regular expressions, of which those that match will be replaced with a scalar (list of regex -> regex)

In [99]: df.replace([r'\s*\.\s*', r'a|b'], nan, regex=True)
Out[99]: 
   a   b    c
0  0 NaN  NaN
1  1 NaN  NaN
2  2 NaN  NaN
3  3 NaN    d

All of the regular expression examples can also be passed with the to_replace argument as the regex argument. In this case the value argument must be passed explicitly by name or regex must be a nested dictionary. The previous example, in this case, would then be

In [100]: df.replace(regex=[r'\s*\.\s*', r'a|b'], value=nan)
Out[100]: 
   a   b    c
0  0 NaN  NaN
1  1 NaN  NaN
2  2 NaN  NaN
3  3 NaN    d

This can be convenient if you do not want to pass regex=True every time you want to use a regular expression.

Note

Anywhere in the above replace examples that you see a regular expression a compiled regular expression is valid as well.

Numeric Replacement

Similar to DataFrame.fillna

In [101]: df = DataFrame(randn(10, 2))

In [102]: df[rand(df.shape[0]) > 0.5] = 1.5

In [103]: df.replace(1.5, nan)
Out[103]: 
          0         1
0 -0.844214 -1.021415
1  0.432396 -0.323580
2  0.423825  0.799180
3  1.262614  0.751965
4       NaN       NaN
5       NaN       NaN
6 -0.498174 -1.060799
7  0.591667 -0.183257
8  1.019855 -1.482465
9       NaN       NaN

Replacing more than one value via lists works as well

In [104]: df00 = df.values[0, 0]

In [105]: df.replace([1.5, df00], [nan, 'a'])
Out[105]: 
           0         1
0          a -1.021415
1  0.4323957 -0.323580
2  0.4238247  0.799180
3   1.262614  0.751965
4        NaN       NaN
5        NaN       NaN
6 -0.4981742 -1.060799
7  0.5916665 -0.183257
8   1.019855 -1.482465
9        NaN       NaN

In [106]: df[1].dtype
Out[106]: dtype('float64')

You can also operate on the DataFrame in place

In [107]: df.replace(1.5, nan, inplace=True)

Warning

When replacing multiple bool or datetime64 objects, the first argument to replace (to_replace) must match the type of the value being replaced type. For example,

s = Series([True, False, True])
s.replace({'a string': 'new value', True: False})  # raises

TypeError: Cannot compare types 'ndarray(dtype=bool)' and 'str'

will raise a TypeError because one of the dict keys is not of the correct type for replacement.

However, when replacing a single object such as,

In [108]: s = Series([True, False, True])

In [109]: s.replace('a string', 'another string')
Out[109]: 
0     True
1    False
2     True
dtype: bool

the original NDFrame object will be returned untouched. We’re working on unifying this API, but for backwards compatibility reasons we cannot break the latter behavior. See GH6354 for more details.

Missing data casting rules and indexing

While pandas supports storing arrays of integer and boolean type, these types are not capable of storing missing data. Until we can switch to using a native NA type in NumPy, we’ve established some “casting rules” when reindexing will cause missing data to be introduced into, say, a Series or DataFrame. Here they are:

data type Cast to
integer float
boolean object
float no cast
object no cast

For example:

In [110]: s = Series(randn(5), index=[0, 2, 4, 6, 7])

In [111]: s > 0
Out[111]: 
0    True
2    True
4    True
6    True
7    True
dtype: bool

In [112]: (s > 0).dtype
Out[112]: dtype('bool')

In [113]: crit = (s > 0).reindex(list(range(8)))

In [114]: crit
Out[114]: 
0    True
1     NaN
2    True
3     NaN
4    True
5     NaN
6    True
7    True
dtype: object

In [115]: crit.dtype
Out[115]: dtype('O')

Ordinarily NumPy will complain if you try to use an object array (even if it contains boolean values) instead of a boolean array to get or set values from an ndarray (e.g. selecting values based on some criteria). If a boolean vector contains NAs, an exception will be generated:

In [116]: reindexed = s.reindex(list(range(8))).fillna(0)

In [117]: reindexed[crit]
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-117-2da204ed1ac7> in <module>()
----> 1 reindexed[crit]

/home/joris/scipy/pandas/pandas/core/series.pyc in __getitem__(self, key)
    537             key = list(key)
    538 
--> 539         if _is_bool_indexer(key):
    540             key = _check_bool_indexer(self.index, key)
    541 

/home/joris/scipy/pandas/pandas/core/common.pyc in _is_bool_indexer(key)
   2039             if not lib.is_bool_array(key):
   2040                 if isnull(key).any():
-> 2041                     raise ValueError('cannot index with vector containing '
   2042                                      'NA / NaN values')
   2043                 return False

ValueError: cannot index with vector containing NA / NaN values

However, these can be filled in using fillna and it will work fine:

In [118]: reindexed[crit.fillna(False)]
Out[118]: 
0    0.126504
2    0.696198
4    0.697416
6    0.601516
7    0.003659
dtype: float64

In [119]: reindexed[crit.fillna(True)]
Out[119]: 
0    0.126504
1    0.000000
2    0.696198
3    0.000000
4    0.697416
5    0.000000
6    0.601516
7    0.003659
dtype: float64