1

I'd like to check the difference between two DataFrame columns. I tried using the command:

np.setdiff1d(train.columns, train_1.columns)

which results in an empty array:

array([], dtype=object)

However, the number of columns in the dataframes are different:

len(train.columns), len(train_1.columns) = (51, 56)

which means that the two DataFrame are obviously different.

What is wrong here?

Shaido
  • 27,497
  • 23
  • 70
  • 73
Takatjuta
  • 77
  • 1
  • 9
  • `np.setdiff1d` is set difference. Meaning it finds the things in the first argument that are not in the second argument. Your return of an empty array implies that there was nothing in `train.columns` that wasn't in `train_1.columns`. – piRSquared Oct 06 '17 at 05:01

2 Answers2

2

The results are correct, however, setdiff1d is order dependent. It will only check for elements in the first input array that do not occur in the second array.

If you do not care which of the dataframes have the unique columns you can use setxor1d. It will return "the unique values that are in only one (not both) of the input arrays", see the documentation.

import numpy

colsA = ['a', 'b', 'c', 'd']
colsB = ['b','c']

c = numpy.setxor1d(colsA, colsB)

Will return you an array containing 'a' and 'd'.

If you want to use setdiff1d you need to check for differences both ways:

//columns in train.columns that are not in train_1.columns
c1 = np.setdiff1d(train.columns, train_1.columns)

//columns in train_1.columns that are not in train.columns
c2 = np.setdiff1d(train_1.columns, train.columns)
Shaido
  • 27,497
  • 23
  • 70
  • 73
0

use something like this

data_3 = data1[~data1.isin(data2)]

Where data1 and data2 are columns and data_3 = data_1 - data_2

Manideep Karthik
  • 99
  • 9