CODEWITHSUNDEEP
pythonscikit-learnfeature-selection
Fence
Fence

Reputation: 343

polynomial feature transformation

While doing some polynomial transformation for my set of features I was reading sklearn.preprocessing PolynomialFeatures transformer, but I realized that the transformation includes all the possible combinations even using the interaction_only=True parameter. I was wondering if there is a way to specify that just some interactions (combinations) are needed. For instance,

Given the following dataframe:

    a   b   c  Z   X   W
0   1   2   3  0   1   0
1   1   2   3  1   0   1
2   1   2   3  0   0   1

Let's say that a,b,c belongs to a type of feature and Z W X to a different one and we are just interested in interactions between features from different types.

So the desired output would just contains original features and the interactions between different type features. Of course by setting interaction_only=True you just get the "real interactions" and avoid features like a^2, Z^2 and so on...

         a   b   c   Z  X   W  a*Z  a*X   a*W   b*Z  b*X  b*W  c*Z  c*X  c*W
 0       0   1   2   3  0   1   0    0     1     0    0    2    0    0    3
 3       1   1   2   3  1   0   1    1     0     1    2    0    2    3    0
 3       2   1   2   3  0   0   1    0     0     1    0    0    2    0    0

I would like just to perform interactions between columns a, b, c and Z, X, W and avoid combinations such as a*c or Z*X

Upvotes: 1

Views: 953

Answers (1)

Robin Nicole
Robin Nicole

Reputation: 666

There does not seems to be any way to obtain the transformation you talk about with the transformer provided by scikit learn but you can build your own transformer to dot it

  • You could remove the interaction you are not interested in after your transformation with a custom transformer see here
  • create your custom Polynomial feature transformer as explained here

Upvotes: 2

Related Questions