-4

Is there any way to train a machine learning algorithm to learn from results of another and then clone it?

Tried clone with scikit learn but keeps getting different results with the same input data

desertnaut
  • 57,590
  • 26
  • 140
  • 166
tdoxox
  • 1

1 Answers1

0

Yes and no. You are roughly aproximating a type of machine learning called student-teacher.

A "teacher" model generates the right answers; a "student" model tries to learn how to generate the same answers (trained on the output of the teacher).

However, you can't expect this approach to clone the model. At best you'll get the same results at or near the data you have available, but you should not expect to match it generally.

Your odds of almost cloning the original model go up if:

  1. You have a lot of data
  2. You have data for most of the imaginable input space
  3. You are using the same algorithm as the original model

Good luck.

MYK
  • 1,988
  • 7
  • 30