0
$\begingroup$

I am looking for help on identifying a class of algorithm. If I have a tabular training and test set I want to know the similarity of rows based on some numeric features. The training data would be labelled such that rows would be paired (or even grouped). The output for each row in the test/prediction set would be the most similar row and the probability that it would have been paired with that row.

In theory there could be a matrix with a score for the pair by pair comparison but I only need the highest for my use case.

This feels somewhat like collaborative filtering but I am not quite sure what to call this class of algorithm. Bonus points if you can point me to a python library.

$\endgroup$
1
  • $\begingroup$ An unsupervised method for this would be to use the proximity matrix of an Isolation Forrest. This would be very sensitive to bad features and number of features. $\endgroup$
    – Keith
    Jan 27 at 0:06
0
$\begingroup$

It seems there does exist a python library for supervised ML to solve this

https://pypi.org/project/dedupe/

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.