0

I am reading a machine learning paper and at some point they write the likelihood of the observed data using $logit$ function in their model. However, they have a $>$ symbol in the parameter of the logit function, something like $logit(\alpha > uv)$ where $u$ and $v$ are the parameters to learn from the model and $\alpha$ is a positive tuning parameter. I specifically talk about Equation 2 in this paper. I haven't seen such a representation before, and want to learn what this really means. Thanks.

user5054
  • 1,549

1 Answers1

4

The equation that you refer to doesn't appear to make sense mathematically. I suspect it is a misprint. You should write to the authors of the article because only they can clarify.

Gordon Smyth
  • 12,807
  • Theoretically it can make sense (interpreting $\alpha > uv$ as a logical constant with TRUE=1 and FALSE=0) but yes, almost certainly a misprint. Most likely $logit(alpha) > uv$ but I am guessing there. – JDL Sep 26 '16 at 08:24
  • 1
    Thanks. I asked the authors and they said it was a misprint of the journal. The "greater than" sign should be "times", so it is actually $\alpha uv$. – user5054 Sep 26 '16 at 20:27
  • @JDL R code and mathematics are not the same thing. Sure, the expression would evaluate in R because of automatic coercion of logical vectors to numeric, but regular mathematical algebra as recognized by the journal has no such coercion, so the published expression is not parsable mathematics. – Gordon Smyth Jul 22 '22 at 00:17