Working with missing data#

Values considered “missing”#

pandas uses different sentinel values to represent a missing (also referred to as NA) depending on the data type.

numpy.nan for NumPy data types. The disadvantage of using NumPy data types is that the original data type will be coerced to np.float64 or object.

In [1]: pd.Series([1, 2], dtype=np.int64).reindex([0, 1, 2])
Out[1]: 
0    1.0
1    2.0
2    NaN
dtype: float64

In [2]: pd.Series([True, False], dtype=np.bool_).reindex([0, 1, 2])
Out[2]: 
0     True
1    False
2      NaN
dtype: object

NaT for NumPy np.datetime64, np.timedelta64, and PeriodDtype. For typing applications, use api.types.NaTType.

In [3]: pd.Series([1, 2], dtype=np.dtype("timedelta64[ns]")).reindex([0, 1, 2])
Out[3]: 
0   0 days 00:00:00.000000001
1   0 days 00:00:00.000000002
2                         NaT
dtype: timedelta64[ns]

In [4]: pd.Series([1, 2], dtype=np.dtype("datetime64[ns]")).reindex([0, 1, 2])
Out[4]: 
0   1970-01-01 00:00:00.000000001
1   1970-01-01 00:00:00.000000002
2                             NaT
dtype: datetime64[ns]

In [5]: pd.Series(["2020", "2020"], dtype=pd.PeriodDtype("D")).reindex([0, 1, 2])
Out[5]: 
0    2020-01-01
1    2020-01-01
2           NaT
dtype: period[D]

NA for StringDtype, Int64Dtype (and other bit widths), Float64Dtype`(and other bit widths), :class:`BooleanDtype and ArrowDtype. These types will maintain the original data type of the data. For typing applications, use api.types.NAType.

In [6]: pd.Series([1, 2], dtype="Int64").reindex([0, 1, 2])
Out[6]: 
0       1
1       2
2    <NA>
dtype: Int64

In [7]: pd.Series([True, False], dtype="boolean[pyarrow]").reindex([0, 1, 2])
Out[7]: 
0     True
1    False
2     <NA>
dtype: bool[pyarrow]

To detect these missing value, use the isna() or notna() methods.

In [8]: ser = pd.Series([pd.Timestamp("2020-01-01"), pd.NaT])

In [9]: ser
Out[9]: 
0   2020-01-01
1          NaT
dtype: datetime64[ns]

In [10]: pd.isna(ser)
Out[10]: 
0    False
1     True
dtype: bool

Note

isna() or notna() will also consider None a missing value.

In [11]: ser = pd.Series([1, None], dtype=object)

In [12]: ser
Out[12]: 
0       1
1    None
dtype: object

In [13]: pd.isna(ser)
Out[13]: 
0    False
1     True
dtype: bool

Warning

Equality compaisons between np.nan, NaT, and NA do not act like None

In [14]: None == None  # noqa: E711
Out[14]: True

In [15]: np.nan == np.nan
Out[15]: False

In [16]: pd.NaT == pd.NaT
Out[16]: False

In [17]: pd.NA == pd.NA
Out[17]: <NA>

Therefore, an equality comparison between a DataFrame or Series with one of these missing values does not provide the same information as isna() or notna().

In [18]: ser = pd.Series([True, None], dtype="boolean[pyarrow]")

In [19]: ser == pd.NA
Out[19]: 
0    <NA>
1    <NA>
dtype: bool[pyarrow]

In [20]: pd.isna(ser)
Out[20]: 
0    False
1     True
dtype: bool

NA semantics#

Warning

Experimental: the behaviour of NA` can still change without warning.

Starting from pandas 1.0, an experimental NA value (singleton) is available to represent scalar missing values. The goal of NA is provide a “missing” indicator that can be used consistently across data types (instead of np.nan, None or pd.NaT depending on the data type).

For example, when having missing values in a Series with the nullable integer dtype, it will use NA:

In [21]: s = pd.Series([1, 2, None], dtype="Int64")

In [22]: s
Out[22]: 
0       1
1       2
2    <NA>
dtype: Int64

In [23]: s[2]
Out[23]: <NA>

In [24]: s[2] is pd.NA
Out[24]: True

Currently, pandas does not yet use those data types using NA by default a DataFrame or Series, so you need to specify the dtype explicitly. An easy way to convert to those dtypes is explained in the conversion section.

Propagation in arithmetic and comparison operations#

In general, missing values propagate in operations involving NA. When one of the operands is unknown, the outcome of the operation is also unknown.

For example, NA propagates in arithmetic operations, similarly to np.nan:

In [25]: pd.NA + 1
Out[25]: <NA>

In [26]: "a" * pd.NA
Out[26]: <NA>

There are a few special cases when the result is known, even when one of the operands is NA.

In [27]: pd.NA ** 0
Out[27]: 1

In [28]: 1 ** pd.NA
Out[28]: 1

In equality and comparison operations, NA also propagates. This deviates from the behaviour of np.nan, where comparisons with np.nan always return False.

In [29]: pd.NA == 1
Out[29]: <NA>

In [30]: pd.NA == pd.NA
Out[30]: <NA>

In [31]: pd.NA < 2.5
Out[31]: <NA>

To check if a value is equal to NA, use isna()

In [32]: pd.isna(pd.NA)
Out[32]: True

Note

An exception on this basic propagation rule are reductions (such as the mean or the minimum), where pandas defaults to skipping missing values. See the calculation section for more.

Logical operations#

For logical operations, NA follows the rules of the three-valued logic (or Kleene logic, similarly to R, SQL and Julia). This logic means to only propagate missing values when it is logically required.

For example, for the logical “or” operation (|), if one of the operands is True, we already know the result will be True, regardless of the other value (so regardless the missing value would be True or False). In this case, NA does not propagate:

In [33]: True | False
Out[33]: True

In [34]: True | pd.NA
Out[34]: True

In [35]: pd.NA | True
Out[35]: True

On the other hand, if one of the operands is False, the result depends on the value of the other operand. Therefore, in this case NA propagates:

In [36]: False | True
Out[36]: True

In [37]: False | False
Out[37]: False

In [38]: False | pd.NA
Out[38]: <NA>

The behaviour of the logical “and” operation (&) can be derived using similar logic (where now NA will not propagate if one of the operands is already False):

In [39]: False & True
Out[39]: False

In [40]: False & False
Out[40]: False

In [41]: False & pd.NA
Out[41]: False
In [42]: True & True
Out[42]: True

In [43]: True & False
Out[43]: False

In [44]: True & pd.NA
Out[44]: <NA>

NA in a boolean context#

Since the actual value of an NA is unknown, it is ambiguous to convert NA to a boolean value.

In [45]: bool(pd.NA)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[45], line 1
----> 1 bool(pd.NA)

File missing.pyx:392, in pandas._libs.missing.NAType.__bool__()

TypeError: boolean value of NA is ambiguous

This also means that NA cannot be used in a context where it is evaluated to a boolean, such as if condition: ... where condition can potentially be NA. In such cases, isna() can be used to check for NA or condition being NA can be avoided, for example by filling missing values beforehand.

A similar situation occurs when using Series or DataFrame objects in if statements, see Using if/truth statements with pandas.

NumPy ufuncs#

pandas.NA implements NumPy’s __array_ufunc__ protocol. Most ufuncs work with NA, and generally return NA:

In [46]: np.log(pd.NA)
Out[46]: <NA>

In [47]: np.add(pd.NA, 1)
Out[47]: <NA>

Warning

Currently, ufuncs involving an ndarray and NA will return an object-dtype filled with NA values.

In [48]: a = np.array([1, 2, 3])

In [49]: np.greater(a, pd.NA)
Out[49]: array([<NA>, <NA>, <NA>], dtype=object)

The return type here may change to return a different array type in the future.

See DataFrame interoperability with NumPy functions for more on ufuncs.

Conversion#

If you have a DataFrame or Series using np.nan, Series.convert_dtypes() and DataFrame.convert_dtypes() in DataFrame that can convert data to use the data types that use NA such as Int64Dtype or ArrowDtype. This is especially helpful after reading in data sets from IO methods where data types were inferred.

In this example, while the dtypes of all columns are changed, we show the results for the first 10 columns.

In [50]: import io

In [51]: data = io.StringIO("a,b\n,True\n2,")

In [52]: df = pd.read_csv(data)

In [53]: df.dtypes
Out[53]: 
a    float64
b     object
dtype: object

In [54]: df_conv = df.convert_dtypes()

In [55]: df_conv
Out[55]: 
      a     b
0  <NA>  True
1     2  <NA>

In [56]: df_conv.dtypes
Out[56]: 
a      Int64
b    boolean
dtype: object

Inserting missing data#

You can insert missing values by simply assigning to a Series or DataFrame. The missing value sentinel used will be chosen based on the dtype.

In [57]: ser = pd.Series([1., 2., 3.])

In [58]: ser.loc[0] = None

In [59]: ser
Out[59]: 
0    NaN
1    2.0
2    3.0
dtype: float64

In [60]: ser = pd.Series([pd.Timestamp("2021"), pd.Timestamp("2021")])

In [61]: ser.iloc[0] = np.nan

In [62]: ser
Out[62]: 
0          NaT
1   2021-01-01
dtype: datetime64[ns]

In [63]: ser = pd.Series([True, False], dtype="boolean[pyarrow]")

In [64]: ser.iloc[0] = None

In [65]: ser
Out[65]: 
0     <NA>
1    False
dtype: bool[pyarrow]

For object types, pandas will use the value given:

In [66]: s = pd.Series(["a", "b", "c"], dtype=object)

In [67]: s.loc[0] = None

In [68]: s.loc[1] = np.nan

In [69]: s
Out[69]: 
0    None
1     NaN
2       c
dtype: object

Calculations with missing data#

Missing values propagate through arithmetic operations between pandas objects.

In [70]: ser1 = pd.Series([np.nan, np.nan, 2, 3])

In [71]: ser2 = pd.Series([np.nan, 1, np.nan, 4])

In [72]: ser1
Out[72]: 
0    NaN
1    NaN
2    2.0
3    3.0
dtype: float64

In [73]: ser2
Out[73]: 
0    NaN
1    1.0
2    NaN
3    4.0
dtype: float64

In [74]: ser1 + ser2
Out[74]: 
0    NaN
1    NaN
2    NaN
3    7.0
dtype: float64

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

When summing data, NA values or empty data will be treated as zero.

In [75]: pd.Series([np.nan]).sum()
Out[75]: 0.0

In [76]: pd.Series([], dtype="float64").sum()
Out[76]: 0.0

When taking the product, NA values or empty data will be treated as 1.

In [77]: pd.Series([np.nan]).prod()
Out[77]: 1.0

In [78]: pd.Series([], dtype="float64").prod()
Out[78]: 1.0

Cumulative methods like cumsum() and cumprod() ignore NA values by default preserve them in the result. This behavior can be changed with skipna

  • Cumulative methods like cumsum() and cumprod() ignore NA values by default, but preserve them in the resulting arrays. To override this behaviour and include NA values, use skipna=False.

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

In [80]: ser
Out[80]: 
0    1.0
1    NaN
2    3.0
3    NaN
dtype: float64

In [81]: ser.cumsum()
Out[81]: 
0    1.0
1    NaN
2    4.0
3    NaN
dtype: float64

In [82]: ser.cumsum(skipna=False)
Out[82]: 
0    1.0
1    NaN
2    NaN
3    NaN
dtype: float64

Dropping missing data#

dropna() dropa rows or columns with missing data.

In [83]: df = pd.DataFrame([[np.nan, 1, 2], [1, 2, np.nan], [1, 2, 3]])

In [84]: df
Out[84]: 
     0  1    2
0  NaN  1  2.0
1  1.0  2  NaN
2  1.0  2  3.0

In [85]: df.dropna()
Out[85]: 
     0  1    2
2  1.0  2  3.0

In [86]: df.dropna(axis=1)
Out[86]: 
   1
0  1
1  2
2  2

In [87]: ser = pd.Series([1, pd.NA], dtype="int64[pyarrow]")

In [88]: ser.dropna()
Out[88]: 
0    1
dtype: int64[pyarrow]

Filling missing data#

Filling by value#

fillna() replaces NA values with non-NA data.

Replace NA with a scalar value

In [89]: data = {"np": [1.0, np.nan, np.nan, 2], "arrow": pd.array([1.0, pd.NA, pd.NA, 2], dtype="float64[pyarrow]")}

In [90]: df = pd.DataFrame(data)

In [91]: df
Out[91]: 
    np  arrow
0  1.0    1.0
1  NaN   <NA>
2  NaN   <NA>
3  2.0    2.0

In [92]: df.fillna(0)
Out[92]: 
    np  arrow
0  1.0    1.0
1  0.0    0.0
2  0.0    0.0
3  2.0    2.0

Fill gaps forward or backward

In [93]: df.ffill()
Out[93]: 
    np  arrow
0  1.0    1.0
1  1.0    1.0
2  1.0    1.0
3  2.0    2.0

In [94]: df.bfill()
Out[94]: 
    np  arrow
0  1.0    1.0
1  2.0    2.0
2  2.0    2.0
3  2.0    2.0

Limit the number of NA values filled

In [95]: df.ffill(limit=1)
Out[95]: 
    np  arrow
0  1.0    1.0
1  1.0    1.0
2  NaN   <NA>
3  2.0    2.0

NA values can be replaced with corresponding value from a Series or DataFrame where the index and column aligns between the original object and the filled object.

In [96]: dff = pd.DataFrame(np.arange(30, dtype=np.float64).reshape(10, 3), columns=list("ABC"))

In [97]: dff.iloc[3:5, 0] = np.nan

In [98]: dff.iloc[4:6, 1] = np.nan

In [99]: dff.iloc[5:8, 2] = np.nan

In [100]: dff
Out[100]: 
      A     B     C
0   0.0   1.0   2.0
1   3.0   4.0   5.0
2   6.0   7.0   8.0
3   NaN  10.0  11.0
4   NaN   NaN  14.0
5  15.0   NaN   NaN
6  18.0  19.0   NaN
7  21.0  22.0   NaN
8  24.0  25.0  26.0
9  27.0  28.0  29.0

In [101]: dff.fillna(dff.mean())
Out[101]: 
       A     B          C
0   0.00   1.0   2.000000
1   3.00   4.0   5.000000
2   6.00   7.0   8.000000
3  14.25  10.0  11.000000
4  14.25  14.5  14.000000
5  15.00  14.5  13.571429
6  18.00  19.0  13.571429
7  21.00  22.0  13.571429
8  24.00  25.0  26.000000
9  27.00  28.0  29.000000

Note

DataFrame.where() can also be used to fill NA values.Same result as above.

In [102]: dff.where(pd.notna(dff), dff.mean(), axis="columns")
Out[102]: 
       A     B          C
0   0.00   1.0   2.000000
1   3.00   4.0   5.000000
2   6.00   7.0   8.000000
3  14.25  10.0  11.000000
4  14.25  14.5  14.000000
5  15.00  14.5  13.571429
6  18.00  19.0  13.571429
7  21.00  22.0  13.571429
8  24.00  25.0  26.000000
9  27.00  28.0  29.000000

Interpolation#

DataFrame.interpolate() and Series.interpolate() fills NA values using various interpolation methods.

In [103]: df = pd.DataFrame(
   .....:     {
   .....:         "A": [1, 2.1, np.nan, 4.7, 5.6, 6.8],
   .....:         "B": [0.25, np.nan, np.nan, 4, 12.2, 14.4],
   .....:     }
   .....: )
   .....: 

In [104]: df
Out[104]: 
     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 [105]: df.interpolate()
Out[105]: 
     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

In [106]: idx = pd.date_range("2020-01-01", periods=10, freq="D")

In [107]: data = np.random.default_rng(2).integers(0, 10, 10).astype(np.float64)

In [108]: ts = pd.Series(data, index=idx)

In [109]: ts.iloc[[1, 2, 5, 6, 9]] = np.nan

In [110]: ts
Out[110]: 
2020-01-01    8.0
2020-01-02    NaN
2020-01-03    NaN
2020-01-04    2.0
2020-01-05    4.0
2020-01-06    NaN
2020-01-07    NaN
2020-01-08    0.0
2020-01-09    3.0
2020-01-10    NaN
Freq: D, dtype: float64

In [111]: ts.plot()
Out[111]: <Axes: >
../_images/series_before_interpolate.png
In [112]: ts.interpolate()
Out[112]: 
2020-01-01    8.000000
2020-01-02    6.000000
2020-01-03    4.000000
2020-01-04    2.000000
2020-01-05    4.000000
2020-01-06    2.666667
2020-01-07    1.333333
2020-01-08    0.000000
2020-01-09    3.000000
2020-01-10    3.000000
Freq: D, dtype: float64

In [113]: ts.interpolate().plot()
Out[113]: <Axes: >
../_images/series_interpolate.png

Interpolation relative to a Timestamp in the DatetimeIndex is available by setting method="time"

In [114]: ts2 = ts.iloc[[0, 1, 3, 7, 9]]

In [115]: ts2
Out[115]: 
2020-01-01    8.0
2020-01-02    NaN
2020-01-04    2.0
2020-01-08    0.0
2020-01-10    NaN
dtype: float64

In [116]: ts2.interpolate()
Out[116]: 
2020-01-01    8.0
2020-01-02    5.0
2020-01-04    2.0
2020-01-08    0.0
2020-01-10    0.0
dtype: float64

In [117]: ts2.interpolate(method="time")
Out[117]: 
2020-01-01    8.0
2020-01-02    6.0
2020-01-04    2.0
2020-01-08    0.0
2020-01-10    0.0
dtype: float64

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

In [118]: idx = [0.0, 1.0, 10.0]

In [119]: ser = pd.Series([0.0, np.nan, 10.0], idx)

In [120]: ser
Out[120]: 
0.0      0.0
1.0      NaN
10.0    10.0
dtype: float64

In [121]: ser.interpolate()
Out[121]: 
0.0      0.0
1.0      5.0
10.0    10.0
dtype: float64

In [122]: ser.interpolate(method="values")
Out[122]: 
0.0      0.0
1.0      1.0
10.0    10.0
dtype: float64

If you have scipy installed, you can pass the name of a 1-d interpolation routine to method. as specified in the scipy interpolation documentation and reference guide. The appropriate interpolation method will depend on the data type.

Tip

If you are dealing with a time series that is growing at an increasing rate, use method='barycentric'.

If you have values approximating a cumulative distribution function, use method='pchip'.

To fill missing values with goal of smooth plotting use method='akima'.

In [123]: df = pd.DataFrame(
   .....:    {
   .....:       "A": [1, 2.1, np.nan, 4.7, 5.6, 6.8],
   .....:       "B": [0.25, np.nan, np.nan, 4, 12.2, 14.4],
   .....:    }
   .....: )
   .....: 

In [124]: df
Out[124]: 
     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 [125]: df.interpolate(method="barycentric")
Out[125]: 
      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 [126]: df.interpolate(method="pchip")
Out[126]: 
         A          B
0  1.00000   0.250000
1  2.10000   0.672808
2  3.43454   1.928950
3  4.70000   4.000000
4  5.60000  12.200000
5  6.80000  14.400000

In [127]: df.interpolate(method="akima")
Out[127]: 
          A          B
0  1.000000   0.250000
1  2.100000  -0.873316
2  3.406667   0.320034
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 [128]: df.interpolate(method="spline", order=2)
Out[128]: 
          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 [129]: df.interpolate(method="polynomial", order=2)
Out[129]: 
          A          B
0  1.000000   0.250000
1  2.100000  -2.703846
2  3.451351  -1.453846
3  4.700000   4.000000
4  5.600000  12.200000
5  6.800000  14.400000

Comparing several methods.

In [130]: np.random.seed(2)

In [131]: ser = pd.Series(np.arange(1, 10.1, 0.25) ** 2 + np.random.randn(37))

In [132]: missing = np.array([4, 13, 14, 15, 16, 17, 18, 20, 29])

In [133]: ser.iloc[missing] = np.nan

In [134]: methods = ["linear", "quadratic", "cubic"]

In [135]: df = pd.DataFrame({m: ser.interpolate(method=m) for m in methods})

In [136]: df.plot()
Out[136]: <Axes: >
../_images/compare_interpolations.png

Interpolating new observations from expanding data with Series.reindex().

In [137]: ser = pd.Series(np.sort(np.random.uniform(size=100)))

# interpolate at new_index
In [138]: new_index = ser.index.union(pd.Index([49.25, 49.5, 49.75, 50.25, 50.5, 50.75]))

In [139]: interp_s = ser.reindex(new_index).interpolate(method="pchip")

In [140]: interp_s.loc[49:51]
Out[140]: 
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

Interpolation limits#

interpolate() accepts a limit keyword argument to limit the number of consecutive NaN values filled since the last valid observation

In [141]: ser = pd.Series([np.nan, np.nan, 5, np.nan, np.nan, np.nan, 13, np.nan, np.nan])

In [142]: ser
Out[142]: 
0     NaN
1     NaN
2     5.0
3     NaN
4     NaN
5     NaN
6    13.0
7     NaN
8     NaN
dtype: float64

In [143]: ser.interpolate()
Out[143]: 
0     NaN
1     NaN
2     5.0
3     7.0
4     9.0
5    11.0
6    13.0
7    13.0
8    13.0
dtype: float64

In [144]: ser.interpolate(limit=1)
Out[144]: 
0     NaN
1     NaN
2     5.0
3     7.0
4     NaN
5     NaN
6    13.0
7    13.0
8     NaN
dtype: float64

By default, NaN values are filled in a forward direction. Use limit_direction parameter to fill backward or from both directions.

In [145]: ser.interpolate(limit=1, limit_direction="backward")
Out[145]: 
0     NaN
1     5.0
2     5.0
3     NaN
4     NaN
5    11.0
6    13.0
7     NaN
8     NaN
dtype: float64

In [146]: ser.interpolate(limit=1, limit_direction="both")
Out[146]: 
0     NaN
1     5.0
2     5.0
3     7.0
4     NaN
5    11.0
6    13.0
7    13.0
8     NaN
dtype: float64

In [147]: ser.interpolate(limit_direction="both")
Out[147]: 
0     5.0
1     5.0
2     5.0
3     7.0
4     9.0
5    11.0
6    13.0
7    13.0
8    13.0
dtype: float64

By default, NaN values are filled whether they are surrounded by existing valid values or outside existing valid values. The limit_area parameter restricts filling to either inside or outside values.

# fill one consecutive inside value in both directions
In [148]: ser.interpolate(limit_direction="both", limit_area="inside", limit=1)
Out[148]: 
0     NaN
1     NaN
2     5.0
3     7.0
4     NaN
5    11.0
6    13.0
7     NaN
8     NaN
dtype: float64

# fill all consecutive outside values backward
In [149]: ser.interpolate(limit_direction="backward", limit_area="outside")
Out[149]: 
0     5.0
1     5.0
2     5.0
3     NaN
4     NaN
5     NaN
6    13.0
7     NaN
8     NaN
dtype: float64

# fill all consecutive outside values in both directions
In [150]: ser.interpolate(limit_direction="both", limit_area="outside")
Out[150]: 
0     5.0
1     5.0
2     5.0
3     NaN
4     NaN
5     NaN
6    13.0
7    13.0
8    13.0
dtype: float64

Replacing values#

Series.replace() and DataFrame.replace() can be used similar to Series.fillna() and DataFrame.fillna() to replace or insert missing values.

In [151]: df = pd.DataFrame(np.eye(3))

In [152]: df
Out[152]: 
     0    1    2
0  1.0  0.0  0.0
1  0.0  1.0  0.0
2  0.0  0.0  1.0

In [153]: df_missing = df.replace(0, np.nan)

In [154]: df_missing
Out[154]: 
     0    1    2
0  1.0  NaN  NaN
1  NaN  1.0  NaN
2  NaN  NaN  1.0

In [155]: df_filled = df_missing.replace(np.nan, 2)

In [156]: df_filled
Out[156]: 
     0    1    2
0  1.0  2.0  2.0
1  2.0  1.0  2.0
2  2.0  2.0  1.0

Replacing more than one value is possible by passing a list.

In [157]: df_filled.replace([1, 44], [2, 28])
Out[157]: 
     0    1    2
0  2.0  2.0  2.0
1  2.0  2.0  2.0
2  2.0  2.0  2.0

Replacing using a mapping dict.

In [158]: df_filled.replace({1: 44, 2: 28})
Out[158]: 
      0     1     2
0  44.0  28.0  28.0
1  28.0  44.0  28.0
2  28.0  28.0  44.0

Regular expression replacement#

Note

Python strings prefixed with the r character such as r'hello world' are “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'\' == '\\'.

Replace the ‘.’ with NaN

In [159]: d = {"a": list(range(4)), "b": list("ab.."), "c": ["a", "b", np.nan, "d"]}

In [160]: df = pd.DataFrame(d)

In [161]: df.replace(".", np.nan)
Out[161]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

Replace the ‘.’ with NaN with regular expression that removes surrounding whitespace

In [162]: df.replace(r"\s*\.\s*", np.nan, regex=True)
Out[162]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

Replace with a list of regexes.

In [163]: df.replace([r"\.", r"(a)"], ["dot", r"\1stuff"], regex=True)
Out[163]: 
   a       b       c
0  0  astuff  astuff
1  1       b       b
2  2     dot     NaN
3  3     dot       d

Replace with a regex in a mapping dict.

In [164]: df.replace({"b": r"\s*\.\s*"}, {"b": np.nan}, regex=True)
Out[164]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

Pass nested dictionaries of regular expressions that use the regex keyword.

In [165]: df.replace({"b": {"b": r""}}, regex=True)
Out[165]: 
   a  b    c
0  0  a    a
1  1       b
2  2  .  NaN
3  3  .    d

In [166]: df.replace(regex={"b": {r"\s*\.\s*": np.nan}})
Out[166]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  NaN  NaN
3  3  NaN    d

In [167]: df.replace({"b": r"\s*(\.)\s*"}, {"b": r"\1ty"}, regex=True)
Out[167]: 
   a    b    c
0  0    a    a
1  1    b    b
2  2  .ty  NaN
3  3  .ty    d

Pass a list of regular expressions that will replace matches with a scalar.

In [168]: df.replace([r"\s*\.\s*", r"a|b"], "placeholder", regex=True)
Out[168]: 
   a            b            c
0  0  placeholder  placeholder
1  1  placeholder  placeholder
2  2  placeholder          NaN
3  3  placeholder            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.

In [169]: df.replace(regex=[r"\s*\.\s*", r"a|b"], value="placeholder")
Out[169]: 
   a            b            c
0  0  placeholder  placeholder
1  1  placeholder  placeholder
2  2  placeholder          NaN
3  3  placeholder            d

Note

A regular expression object from re.compile is a valid input as well.