CODEWITHSUNDEEP
pythonpandas
geronimo
geronimo

Reputation: 1347

What is the best way to perform an operation on several DataFrames?

Let's say I have three DataFrames:

import pandas as pd
import numpy as np

cols = ['A','B','C']
index = [1,2,3,4,5]
np.random.seed(42)

apple = pd.DataFrame(np.random.randn(5,3), index=index, columns=cols)
orange = pd.DataFrame(np.random.randn(5,3), index=index, columns=cols)
banana = pd.DataFrame(np.random.randn(5,3), index=index, columns=cols)

In [50]: apple
Out[50]:
          A         B         C
1  0.496714 -0.138264  0.647689
2  1.523030 -0.234153 -0.234137
3  1.579213  0.767435 -0.469474
4  0.542560 -0.463418 -0.465730
5  0.241962 -1.913280 -1.724918

In [51]: orange
Out[51]:
          A         B         C
1 -0.562288 -1.012831  0.314247
2 -0.908024 -1.412304  1.465649
3 -0.225776  0.067528 -1.424748
4 -0.544383  0.110923 -1.150994
5  0.375698 -0.600639 -0.291694

In [52]: banana
Out[52]:
          A         B         C
1 -0.601707  1.852278 -0.013497
2 -1.057711  0.822545 -1.220844
3  0.208864 -1.959670 -1.328186
4  0.196861  0.738467  0.171368
5 -0.115648 -0.301104 -1.478522

What's the best/fastest/easiest way to create a new dataframe with the same columns and index, but with the maximum value from each column and index for apple, orange, banana? I.e., for [1,A] the new dataframe value would be 0.496714, for [1,B] the value would be 1.852278, etc. Thanks!

Upvotes: 2

Views: 68

Answers (2)

Jon Tay
Jon Tay

Reputation: 1

Why not concatenate the DataFrames into a Panel and then use Panel.max()?

ie:pd.Panel({'a':apple ,'b':banana,'o';orange}).max(axis=0)

Admittedly not the fastest, but this guarantees correct index alignment, and you might want to use the Panel for something else later. Your data looks to be 3D, with 3 indexing elements (cols/index/fruit), so use a 3D data structure.

Upvotes: 0

roman
roman

Reputation: 117370

I think something like this should be fast:

np.maximum(np.maximum(orange, apple), banana)

Using numpy.maximum():

Element-wise maximum of array elements.

As @Jeff suggested in comments, in general it would be:

reduce(np.maximum, [orange,apple,banana])

Upvotes: 3

Related Questions