Feature selection is fairly easy in e.g. Python's scikit-learn which sports a module and tutorials or MATLAB. Mathematica even touts that it uses feature selection in its automated machine learning algorithms, though I've had trouble finding much more on this, especially anything in Mathematica documentation. Related SE posts are How to know the internal pre-processing of automatic machine learning function Classify? and How to view ClassifierFunction's preprocessed data?

I could probably use Mathematica's Variance[] or Correlation[] functions for some basic feature selection. I'm also aware that I could try different combinations of features and choose ones with higher accuracy, as described in How can I determine the importance of variables from Classify?.

Maybe in Predict[], there's some option I'm overlooking..

I'm also aware of and a moderate user of the Wolfram Client Library for Python, in which case I could use some of scikit-learn's functionality, though it would be nice to know if Mathematica has built-in options or if someone has developed some sophisticated code to do this.

  • $\begingroup$ Do FeatureExtract or DimensionReduce have options you're looking for? $\endgroup$
    – tad
    Jul 16 at 22:48
  • $\begingroup$ @tad Unfortunately, no, at least from what I can tell. With DimensionReduce, it's mostly PCA-like transforms (PCA, MDS, TSNE, etc.). FeatureExtract is nice with the breadth of feature types (I especially like that it handles Nominal data in Predict for example), but I don't see any options that would indicate which features are most important. $\endgroup$
    – Sterling
    Jul 18 at 1:39
  • $\begingroup$ Wolfram said they might try to work on it. Probably more likely to if others express interest too. $\endgroup$
    – Sterling
    Aug 21 at 17:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.