I have a question regarding the time needed for training a classifier. I am facing the specific problem of Sentiment Analysis (classification of text as pos/neg/neu).

(Excepting online learning algorithms) Since the training stage of a classifier is usually conducted only once and offline, in the end does it really matter how much time would it take for training ?

If the classifier is to be trained over a mobile or a resource-constrained platform, reducing such training time would still make more sense, though...

Thanks in advance.

  • 2
    $\begingroup$ I think you answer your own question. The logic seems reasonable - for practical purposes, if you only need to run a computing task once, then the length of time is fairly irrelevant unless it's so long that it can't even be performed once. You may want to consider the possibility of tweaks to the classification tool, however. How quickly would you need to be able to respond to new topics appearing in the text? Would you ever need to go back and re-classify old text? etc. $\endgroup$ – Jonathan Jun 23 '12 at 15:58
  • $\begingroup$ Thanks! I guess algorithm tweaking and evaluation makes training time important when building a classifier... $\endgroup$ – kanzen_master Jun 24 '12 at 0:30

I think it depends on what you mean by "training time". If you think about cross validation, model selection, etc, a classifier which takes a "long time" to train can become incredibly painful, since you will be training it many times and waiting on the results before you can make further progress.

  • $\begingroup$ I see! Training is not something that is done only once, time for feature selection, model selection, validation via CV or bootstrapping should also be taken into account...Thanks! $\endgroup$ – kanzen_master Jun 24 '12 at 0:27
  • 1
    $\begingroup$ @DT: Again, someone might say that those cases don't really matter, since your system will spend a lot more time in production than in development. But if you're the developer, it makes at least a psychological difference. $\endgroup$ – Wayne Jun 24 '12 at 2:35

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.