Richard Redding
  • Member for 5 years, 5 months
  • Last seen more than a month ago
Gaussian Mixture Models: Maximum Likelihood Estimation or Expectation Maximization?
11 votes

Mehrin, you are in danger of creating a false dichotomy. EM is an optimisation technique that can be used to find maximum likelihood estimates and so the choice is not "one or the other". In mixture ...

View answer
Log model too good to be true, maybe I'm interpreting results incorrectly?
Accepted answer
10 votes

You need to transform your prediction back to the original space before calculating residuals rather than transforming the residuals from the log space. $ \newcommand{\Exp}{\operatorname{Exp}} \...

View answer
How to select noninformative priors on hyperparameters in gaussian process regression
Accepted answer
5 votes

There is no such thing as a truly noninformative prior--even the prior you've selected for the mean is not uniform under a reparameterisation. Having said that, you could use proper uniform priors on ...

View answer
Generating correlated uniform random variable with R
Accepted answer
3 votes

Try simulating from a multivariate normal distribution and then transforming the values by using the normal cdf. This will produce correlated standard uniform variates. You can then shift and scale ...

View answer
How would you explain Markov Chain Monte Carlo (MCMC) to a layperson?
3 votes

MCMC is typically used as an alternative to crude Monte Carlo simulation techniques. Both MCMC and other Monte Carlo techniques are used to evaluate difficult integrals but MCMC can be used more ...

View answer
Motivation for Ward's definition of error sum of squares (ESS)
2 votes

Ward's ESS is the same as the SS you mention. If you distribute the terms in your formula you get: $ \sum(x_i - \bar x)^2 = \sum x_i^2 + \sum \bar x^2 - 2 \bar x \sum x_i = \sum x_i^2 - n \bar x ^2 =...

View answer
Likelihood of an algorithm-based function
2 votes

So you can simulate pseudo observations given the parameters. In this case it might be helpful to use Approximate Bayesian Computation (ABC). In its crudest form you: define a distance metric ...

View answer
Is publishing work based on post hoc analysis problematic?
2 votes

Making up a story would be dangerous but, to your general point, it depends. It obviously complicates any statistical methods you use but in many cases it would be wasteful to not contribute new ...

View answer
Capturing expert inputs about price variation in the future
2 votes

There's no right answer to this and you will have to be careful about how you synthesise these opinions with any hard data that you subsequently record. With those caveats out there, I'd recommend ...

View answer
validating a gaussian process fitted to data
1 votes

... is it realistic to expect that one can actually find parametric forms for $\mu(x)$, $\sigma^2 (x)$ in order to "validate" the use of a Gaussian Process to fit the data? Yes, if I take "validate" ...

View answer
Merging observations in Gaussian Process
1 votes

I have also been investigating merging observations when performing Gaussian Process regression. In my problem I have only one covariate. I'm not sure I necessarily agree that the Nystrom ...

View answer
Maximum possible percentage above a sample mean
1 votes

In a sample there must be at least one record with a value less than the mean of the sample (unless all records are equal). Hence the greatest proportion of records greater than the mean is 1-1/n.

View answer
How to test if Significantly Different: Mean of All Sample and Cluster Mean
1 votes

Depending on the sample size you could consider a permutation test. First calculate the difference in the means of the two clusters. Then estimate the distribution of the differences in the means of ...

View answer
Does a logistic regression end up modelling linear relationships?
1 votes

Logistic regression is a GLM. The linear ("L") part of GLM relates to the fact that the covariates are included within a linear predictor. However, the linear predictor is transformed by a link ...

View answer
Probability distribution for heavy zero, right skewed data
1 votes

The tweedie distribution is in the exponential family, hence can be used in a GLM context, and it allows for the occurrence of zeroes in the response variable. You could see if this suits your purpose....

View answer
Gaussian Process for periodic data?
0 votes

A bit late but an alternative to David's suggestion would be to combine your daily data sets. Assuming the mean function is constant from one set of measurements to the next you could create a ...

View answer
Evaluating Forecast Performance of GARCH Models with MLE as Evaluation Criterion
Accepted answer
0 votes

I think you are asking why the out of sample likelihood isn't used as the loss function for assessing accuracy. As a matter of fact it is, e.g. See http://ageconsearch.umn.edu/bitstream/18947/1/...

View answer
restating prior Q re adding a risky bet
0 votes

You have framed this decision based on picking the portfolio of bets that give you high return (based on the prospective mean return) and low risk (based on the SD of prospective returns). Given ...

View answer
Testing correlation between a Boolean and integer variable
0 votes

First off, we know the groups aren't independent. In fact, once you know the ranks in the second group, you know the ranks in the first (and vice versa)--i.e. each rank is either in one group or the ...

View answer