8
$\begingroup$

Laplace smoothing (also known as additive smoothing) is a very particular smoothing technique concerning categorical data. Laplace smoothing is almost always associated with a probability regularisation task and thus it does not immediately relate to "standard" smoothing techniques (eg. kernel smoothing, smoothing spline, etc.) concerned with variations along a (usually linear) continuum.

I think it will be beneficial to have a separate [laplace-smoothing] tag. At first instance it seems that 25 to 30 questions might warrant the tag. I think it will make it easier for ML practitioners to recognise/filter relevant questions. Some questions that would affected are the following:

  1. Understanding Add-1/Laplace smoothing with bigrams
  2. Laplace smoothing and Dirichlet prior
  3. Laplace smoothing and naive bayes
  4. Laplace smoothing understanding implementation
  5. Markov chain getting stuck due to insufficient data samples
  6. What's a good approach to estimate the probability of word frequencies?
  7. How to handle unseen features in a Naive Bayes classifier?
  8. In Naive Bayes, why bother with Laplacian smoothing when we have unknown words in the test set?

(Unsurprisingly, NLP applications seems to generally concerned with this type of regularisation...)

$\endgroup$
  • 2
    $\begingroup$ If you are willing to create this tag and to re-tag 25-30 existing questions with it, then I don't see any disadvantage. I think it would be a decent tag to have. +1. $\endgroup$ – amoeba Apr 13 '17 at 7:42
  • 2
    $\begingroup$ Yeah, why not? I can do it. (I will be careful not to flood the main I will do them in two sittings.) $\endgroup$ – usεr11852 Apr 13 '17 at 22:28
  • $\begingroup$ Laplace smoothing is the kernel smoothing. It has some interesting properties such as minimizing the length of the smooth line $\endgroup$ – Aksakal Apr 14 '17 at 21:49
  • 2
    $\begingroup$ @Aksakal: I think it is only under a very generic and bird's eye view of kernel smoothing. :D I have strong doubts that people interested in Laplace smoothing applications (on this website/ online community) associate the two on first instance. For example in all eight questions linked above the OP (or people answering) make no references to that association so it probably beats the point of basing a tag on that idea. Please feel free to provide counter-examples. $\endgroup$ – usεr11852 Apr 14 '17 at 22:47
  • 1
    $\begingroup$ With 5 upvotes and 0 downvotes, I'd say we have a consensus here. $\endgroup$ – amoeba Apr 15 '17 at 15:53
  • $\begingroup$ Any progress on that @usεr11852? :) $\endgroup$ – amoeba May 10 '17 at 7:16
  • 1
    $\begingroup$ @amoeba Nothing at all. Honestly, I thought someone else would create the tag first. :D (No worries I will start the labelling soon.) $\endgroup$ – usεr11852 May 10 '17 at 18:10
  • $\begingroup$ (First dozen done.) $\endgroup$ – usεr11852 May 10 '17 at 20:51
3
$\begingroup$

Label [laplace-smoothing] created. Accompanying wiki-lemma created. Label added to ~20 questions. Instances of [add-one-smoothing] relabelled with [laplace-smoothing].

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .