Advertisement
Advertisement


Convert pandas dataframe to NumPy array


Question

I am interested in knowing how to convert a pandas dataframe into a NumPy array.

dataframe:

import numpy as np
import pandas as pd

index = [1, 2, 3, 4, 5, 6, 7]
a = [np.nan, np.nan, np.nan, 0.1, 0.1, 0.1, 0.1]
b = [0.2, np.nan, 0.2, 0.2, 0.2, np.nan, np.nan]
c = [np.nan, 0.5, 0.5, np.nan, 0.5, 0.5, np.nan]
df = pd.DataFrame({'A': a, 'B': b, 'C': c}, index=index)
df = df.rename_axis('ID')

gives

label   A    B    C
ID                                 
1   NaN  0.2  NaN
2   NaN  NaN  0.5
3   NaN  0.2  0.5
4   0.1  0.2  NaN
5   0.1  0.2  0.5
6   0.1  NaN  0.5
7   0.1  NaN  NaN

I would like to convert this to a NumPy array, as so:

array([[ nan,  0.2,  nan],
       [ nan,  nan,  0.5],
       [ nan,  0.2,  0.5],
       [ 0.1,  0.2,  nan],
       [ 0.1,  0.2,  0.5],
       [ 0.1,  nan,  0.5],
       [ 0.1,  nan,  nan]])

How can I do this?


As a bonus, is it possible to preserve the dtypes, like this?

array([[ 1, nan,  0.2,  nan],
       [ 2, nan,  nan,  0.5],
       [ 3, nan,  0.2,  0.5],
       [ 4, 0.1,  0.2,  nan],
       [ 5, 0.1,  0.2,  0.5],
       [ 6, 0.1,  nan,  0.5],
       [ 7, 0.1,  nan,  nan]],
     dtype=[('ID', '<i4'), ('A', '<f8'), ('B', '<f8'), ('B', '<f8')])

or similar?

2019/06/06
1
497
6/6/2019 12:09:06 AM


df.to_numpy() is better than df.values, here's why.

It's time to deprecate your usage of values and as_matrix().

pandas v0.24.0 introduced two new methods for obtaining NumPy arrays from pandas objects:

  1. to_numpy(), which is defined on Index, Series, and DataFrame objects, and
  2. array, which is defined on Index and Series objects only.

If you visit the v0.24 docs for .values, you will see a big red warning that says:

Warning: We recommend using DataFrame.to_numpy() instead.

See this section of the v0.24.0 release notes, and this answer for more information.



Towards Better Consistency: to_numpy()

In the spirit of better consistency throughout the API, a new method to_numpy has been introduced to extract the underlying NumPy array from DataFrames.

# Setup.
df = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]}, index=['a', 'b', 'c'])

df.to_numpy()
array([[1, 4],
       [2, 5],
       [3, 6]])

As mentioned above, this method is also defined on Index and Series objects (see here).

df.index.to_numpy()
# array(['a', 'b', 'c'], dtype=object)

df['A'].to_numpy()
#  array([1, 2, 3])

By default, a view is returned, so any modifications made will affect the original.

v = df.to_numpy()
v[0, 0] = -1
 
df
   A  B
a -1  4
b  2  5
c  3  6

If you need a copy instead, use to_numpy(copy=True).


pandas >= 1.0 update for ExtensionTypes

If you're using pandas 1.x, chances are you'll be dealing with extension types a lot more. You'll have to be a little more careful that these extension types are correctly converted.

a = pd.array([1, 2, None], dtype="Int64")                                  
a                                                                          

<IntegerArray>
[1, 2, <NA>]
Length: 3, dtype: Int64 

# Wrong
a.to_numpy()                                                               
# array([1, 2, <NA>], dtype=object)  # yuck, objects

# Right
a.to_numpy(dtype='float', na_value=np.nan)                                 
# array([ 1.,  2., nan])

This is called out in the docs.


If you need the dtypes...

As shown in another answer, DataFrame.to_records is a good way to do this.

df.to_records()
# rec.array([('a', -1, 4), ('b',  2, 5), ('c',  3, 6)],
#           dtype=[('index', 'O'), ('A', '<i8'), ('B', '<i8')])

This cannot be done with to_numpy, unfortunately. However, as an alternative, you can use np.rec.fromrecords:

v = df.reset_index()
np.rec.fromrecords(v, names=v.columns.tolist())
# rec.array([('a', -1, 4), ('b',  2, 5), ('c',  3, 6)],
#          dtype=[('index', '<U1'), ('A', '<i8'), ('B', '<i8')])

Performance wise, it's nearly the same (actually, using rec.fromrecords is a bit faster).

df2 = pd.concat([df] * 10000)

%timeit df2.to_records()
%%timeit
v = df2.reset_index()
np.rec.fromrecords(v, names=v.columns.tolist())

11.1 ms ± 557 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
9.67 ms ± 126 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)


Rationale for Adding a New Method

to_numpy() (in addition to array) was added as a result of discussions under two GitHub issues GH19954 and GH23623.

Specifically, the docs mention the rationale:

[...] with .values it was unclear whether the returned value would be the actual array, some transformation of it, or one of pandas custom arrays (like Categorical). For example, with PeriodIndex, .values generates a new ndarray of period objects each time. [...]

to_numpy aim to improve the consistency of the API, which is a major step in the right direction. .values will not be deprecated in the current version, but I expect this may happen at some point in the future, so I would urge users to migrate towards the newer API, as soon as you can.



Critique of Other Solutions

DataFrame.values has inconsistent behaviour, as already noted.

DataFrame.get_values() is simply a wrapper around DataFrame.values, so everything said above applies.

DataFrame.as_matrix() is deprecated now, do NOT use!

2020/08/29

Note: The .as_matrix() method used in this answer is deprecated. Pandas 0.23.4 warns:

Method .as_matrix will be removed in a future version. Use .values instead.


Pandas has something built in...

numpy_matrix = df.as_matrix()

gives

array([[nan, 0.2, nan],
       [nan, nan, 0.5],
       [nan, 0.2, 0.5],
       [0.1, 0.2, nan],
       [0.1, 0.2, 0.5],
       [0.1, nan, 0.5],
       [0.1, nan, nan]])
2019/03/13

I would just chain the DataFrame.reset_index() and DataFrame.values functions to get the Numpy representation of the dataframe, including the index:

In [8]: df
Out[8]: 
          A         B         C
0 -0.982726  0.150726  0.691625
1  0.617297 -0.471879  0.505547
2  0.417123 -1.356803 -1.013499
3 -0.166363 -0.957758  1.178659
4 -0.164103  0.074516 -0.674325
5 -0.340169 -0.293698  1.231791
6 -1.062825  0.556273  1.508058
7  0.959610  0.247539  0.091333

[8 rows x 3 columns]

In [9]: df.reset_index().values
Out[9]:
array([[ 0.        , -0.98272574,  0.150726  ,  0.69162512],
       [ 1.        ,  0.61729734, -0.47187926,  0.50554728],
       [ 2.        ,  0.4171228 , -1.35680324, -1.01349922],
       [ 3.        , -0.16636303, -0.95775849,  1.17865945],
       [ 4.        , -0.16410334,  0.0745164 , -0.67432474],
       [ 5.        , -0.34016865, -0.29369841,  1.23179064],
       [ 6.        , -1.06282542,  0.55627285,  1.50805754],
       [ 7.        ,  0.95961001,  0.24753911,  0.09133339]])

To get the dtypes we'd need to transform this ndarray into a structured array using view:

In [10]: df.reset_index().values.ravel().view(dtype=[('index', int), ('A', float), ('B', float), ('C', float)])
Out[10]:
array([( 0, -0.98272574,  0.150726  ,  0.69162512),
       ( 1,  0.61729734, -0.47187926,  0.50554728),
       ( 2,  0.4171228 , -1.35680324, -1.01349922),
       ( 3, -0.16636303, -0.95775849,  1.17865945),
       ( 4, -0.16410334,  0.0745164 , -0.67432474),
       ( 5, -0.34016865, -0.29369841,  1.23179064),
       ( 6, -1.06282542,  0.55627285,  1.50805754),
       ( 7,  0.95961001,  0.24753911,  0.09133339),
       dtype=[('index', '<i8'), ('A', '<f8'), ('B', '<f8'), ('C', '<f8')])
2014/03/26

You can use the to_records method, but have to play around a bit with the dtypes if they are not what you want from the get go. In my case, having copied your DF from a string, the index type is string (represented by an object dtype in pandas):

In [102]: df
Out[102]: 
label    A    B    C
ID                  
1      NaN  0.2  NaN
2      NaN  NaN  0.5
3      NaN  0.2  0.5
4      0.1  0.2  NaN
5      0.1  0.2  0.5
6      0.1  NaN  0.5
7      0.1  NaN  NaN

In [103]: df.index.dtype
Out[103]: dtype('object')
In [104]: df.to_records()
Out[104]: 
rec.array([(1, nan, 0.2, nan), (2, nan, nan, 0.5), (3, nan, 0.2, 0.5),
       (4, 0.1, 0.2, nan), (5, 0.1, 0.2, 0.5), (6, 0.1, nan, 0.5),
       (7, 0.1, nan, nan)], 
      dtype=[('index', '|O8'), ('A', '<f8'), ('B', '<f8'), ('C', '<f8')])
In [106]: df.to_records().dtype
Out[106]: dtype([('index', '|O8'), ('A', '<f8'), ('B', '<f8'), ('C', '<f8')])

Converting the recarray dtype does not work for me, but one can do this in Pandas already:

In [109]: df.index = df.index.astype('i8')
In [111]: df.to_records().view([('ID', '<i8'), ('A', '<f8'), ('B', '<f8'), ('C', '<f8')])
Out[111]:
rec.array([(1, nan, 0.2, nan), (2, nan, nan, 0.5), (3, nan, 0.2, 0.5),
       (4, 0.1, 0.2, nan), (5, 0.1, 0.2, 0.5), (6, 0.1, nan, 0.5),
       (7, 0.1, nan, nan)], 
      dtype=[('ID', '<i8'), ('A', '<f8'), ('B', '<f8'), ('C', '<f8')])

Note that Pandas does not set the name of the index properly (to ID) in the exported record array (a bug?), so we profit from the type conversion to also correct for that.

At the moment Pandas has only 8-byte integers, i8, and floats, f8 (see this issue).

2012/11/02

It seems like df.to_records() will work for you. The exact feature you're looking for was requested and to_records pointed to as an alternative.

I tried this out locally using your example, and that call yields something very similar to the output you were looking for:

rec.array([(1, nan, 0.2, nan), (2, nan, nan, 0.5), (3, nan, 0.2, 0.5),
       (4, 0.1, 0.2, nan), (5, 0.1, 0.2, 0.5), (6, 0.1, nan, 0.5),
       (7, 0.1, nan, nan)],
      dtype=[(u'ID', '<i8'), (u'A', '<f8'), (u'B', '<f8'), (u'C', '<f8')])

Note that this is a recarray rather than an array. You could move the result in to regular numpy array by calling its constructor as np.array(df.to_records()).

2018/04/23