0

I try understand the log likelihood in weka. I read about that is a probabilistic metric, but i cant understand, if is better when have low value or high value? How i can get the likelihood value, what function can use? What is better 1, 0, -3 or -4? Why weka get negative value result?

Tim
  • 108,699
  • 20
  • 212
  • 390
Edian Franklin Franco de los S
  • 11
  • 1
  • 4
  • 2
    Can you add some additional context? It will help us to make this clearer & more concrete if you can provide a simple model & small dataset, eg. – gung - Reinstate Monica Apr 06 '16 at 15:08
  • The weka result have de log likelihood metric a don't know who interpret the data, I'm working with genes and transcritomic data, I try get cluster from that data. When is good or bag de log likelihood. – Edian Franklin Franco de los S Apr 06 '16 at 17:04

1 Answers1

1

Likelihood is a function of parameters given the "fixed" data

$$ L(\theta|\text{data}) = f(\text{data}|\theta) $$

You are talking about maximum likelihood estimation, so as you can guess from it's name, we are looking for such value of $\theta$ that maximizes the likelihood.

In real-life implementations we often use optimizers that are designed for finding minimum rather than maximum of function. Because of that you can see that people are minimizing the negative likelihood.

Moreover, information criteria such as AIC, or BIC are also defined in terms of negative log-likelihood, what makes value $-2\log(L)$ so popular to appear in output of statistical software, comparing to raw likelihood.

Tim
  • 108,699
  • 20
  • 212
  • 390
  • but for example in weka is better high value or low value? Log likelihood: 9.56217 I get this result is good?? – Edian Franklin Franco de los S Apr 06 '16 at 18:55
  • 1
    @EdianFranklinFrancodelosS you need to check the documentation if the value that is returned is raw log-likelihood, or negative log-likelihood. However if it is named "log-likelihood" than it is probably raw log-likelihood. – Tim Apr 06 '16 at 19:02