3
$\begingroup$

I must be missing something here - it seems that the first argument to ActiveClassification is your target function, but if you already know that function then why are you trying to learn it?

Examples:

f = # > 50 &;
aco = ActiveClassification[f, {2, 5, 58, 74, 17, 29, 66, 12, 90, 77}];
c = aco["ClassifierFunction"]

But c is only an approximation to f... so I'm confused.

$\endgroup$
5
$\begingroup$

It helps if you think of f as an oracle that is hard to get to, model, or just takes a long to time to produce an answer. So, an approximation is often what you need. For example, one job that is not easy to write a simple model for is document classification; even something as simple as "is this document relevant?" is not that simple. So, in that case, f would display a document with buttons allowing you to indicate relevance. In this way, you can train a computational model that can do the job of the classifier with reasonable efficacy. Also, ActiveClassification can further refine its model using the option InitialEvaluationHistory to start where the previous session left off. So, if you monitor the model's false-positives/negatives, you can feed them back in to improve the model.

$\endgroup$
  • $\begingroup$ But all the examples in the docs are simple functions. Could you show a minimal example of such an interactive processes (where f is slow or requires human input)? $\endgroup$ – user5601 Mar 16 '17 at 21:43
  • $\begingroup$ And what models are the function using to build the approximator? $\endgroup$ – user5601 Mar 16 '17 at 21:44
  • $\begingroup$ @user5601 he actually gave such an example, when discussing the document classification: "So, in that case, f would display a document with buttons allowing you to indicate relevance". That requires human input. Also as for the models the function is using, the system should be figuring that out itself, no? It's machine learning, after all. $\endgroup$ – b3m2a1 Mar 16 '17 at 22:29
  • $\begingroup$ @MB1965 yes yes, but in code, how can one use a function in ActiveClassificaiton that displays a pallette or something and waits for user input... $\endgroup$ – user5601 Mar 16 '17 at 22:31
  • $\begingroup$ @MB1965 Agreed, I would just like to know a bit about the possible model space, it just seems very opaque. $\endgroup$ – user5601 Mar 16 '17 at 22:32

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.