Skip to main content

Why does Kaggle use Log-loss?

Mixed Models in R - Bigger, Faster, Stronger

When you start doing more advanced sports analytics you'll eventually starting working with what are known as hierarchical, nested or mixed effects models. These are models that contain both fixed and random effects. There are multiple ways of defining fixed vs random random effects, but one way I find particularly useful is that random effects are being "predicted" rather than "estimated", and this in turn involves some "shrinkage" towards the mean.

Here's some R code for NCAA ice hockey power rankings using a nested Poisson model (which can be found in my hockey GitHub repository):
model <- gs ~ year+field+d_div+o_div+game_length+(1|offense)+(1|defense)+(1|game_id)
fit <- glmer(model,
             data=g,
      verbose=TRUE,
      family=poisson(link=log)
      )
The fixed effects are year, field (home/away/neutral), d_div (NCAA division of the defense), o_div (NCAA division of the offense) and game_length (number of overtime periods); offense (strength of offense), defense (strength of defense) and game_id are all random effects. The reason for modeling team offenses and defenses as random vs fixed effects is that I view them as random samples from the same distribution. As mentioned above, this results in statistical shrinkage or "regression to the mean" for these values, which is particularly useful for partially completed seasons.

One of the problems with large mixed models is that they can be very slow to fit. For example, the model above takes several hours on a 12-core workstation, which makes it very difficult to test model changes and tweaks. Is there any way to speed up the fitting process? Certainly! One way is to add two options to the above code:

fit <- glmer(model,
             data=g,
      verbose=TRUE,
      family=poisson(link=log),
      nAGQ=0,
      control=glmerControl(optimizer = "nloptwrap")
      )
What do these do? Model fitting is an optimization process. Part of that process involves the estimation of particular integrals, which can be very slow; the option "nAGQ=0" tells glmer to ignore estimating those integrals. For some models this has minimal impact on parameter estimates, and this NCAA hockey model is one of those. The second option tells glmer to fit using the "nloptwrap" optimizer (there are several other optimizers available, too), which tends to be faster than the default optimization method.

The impact can be rather startling. With the default options the above model takes about 3 hours to fit. Add these two options, and the model fitting takes 30 seconds with minimal impact on the parameter estimates, or approximately 400 times faster.

Comments

  1. thanks a lot!! love you, you made my day!

    ReplyDelete
  2. I am very new to GLMs, and GLMMs. But, I'm trying to analyze some data we gather in our lab. We have one categorical dependent variable (accuracy in a 2AFC) and three nominal independent variables, treatment condition (4 levels), order of presentation of treatment condition( 4 levels), and distractor type in the 2AFC (3 levels). Each subject (n=24) participated in each of the treatment conditions, and each treatment condition had exactly 3 instances of the each of the levels of the distractor. Order of treatment was fully counterbalanced. So, I fit a glmer model of the form

    modela = glmer(accuracy ~ foil + condition + order + foil:condition:order + order:condition + order:foil + condition:order + (1|subnum), data=mydata,REML=FALSE, family="binomial")

    but that failed to converge. So, I tried your two options,

    modeld = glmer(accuracy ~ order+condition+foil + condition:foil + order:condition + foil:order + condition:foil:order + (condition + foil + order|subnum), data=mydata, family = "binomial", control=glmerControl(optimizer="bobyqa"), nAGQ=0)

    and that returns with a warning "maxfun < 10 * length(par)^2 is not recommended."

    I can call summary on both models, but the estimates and their z-values end up being quite different. Any idea of what's going wrong here?

    ReplyDelete
  3. Thank you very much. This kinda changed everything in my research.

    ReplyDelete

Post a Comment

Popular posts from this blog

A Bayes' Solution to Monty Hall

For any problem involving conditional probabilities one of your greatest allies is Bayes' Theorem. Bayes' Theorem says that for two events A and B, the probability of A given B is related to the probability of B given A in a specific way.

Standard notation:

probability of A given B is written \( \Pr(A \mid B) \)
probability of B is written \( \Pr(B) \)

Bayes' Theorem:

Using the notation above, Bayes' Theorem can be written: \[ \Pr(A \mid B) = \frac{\Pr(B \mid A)\times \Pr(A)}{\Pr(B)} \]Let's apply Bayes' Theorem to the Monty Hall problem. If you recall, we're told that behind three doors there are two goats and one car, all randomly placed. We initially choose a door, and then Monty, who knows what's behind the doors, always shows us a goat behind one of the remaining doors. He can always do this as there are two goats; if we chose the car initially, Monty picks one of the two doors with a goat behind it at random.

Assume we pick Door 1 and then Monty sho…

What's the Value of a Win?

In a previous entry I demonstrated one simple way to estimate an exponent for the Pythagorean win expectation. Another nice consequence of a Pythagorean win expectation formula is that it also makes it simple to estimate the run value of a win in baseball, the point value of a win in basketball, the goal value of a win in hockey etc.

Let our Pythagorean win expectation formula be \[ w=\frac{P^e}{P^e+1},\] where \(w\) is the win fraction expectation, \(P\) is runs/allowed (or similar) and \(e\) is the Pythagorean exponent. How do we get an estimate for the run value of a win? The expected number of games won in a season with \(g\) games is \[W = g\cdot w = g\cdot \frac{P^e}{P^e+1},\] so for one estimate we only need to compute the value of the partial derivative \(\frac{\partial W}{\partial P}\) at \(P=1\). Note that \[ W = g\left( 1-\frac{1}{P^e+1}\right), \] and so \[ \frac{\partial W}{\partial P} = g\frac{eP^{e-1}}{(P^e+1)^2}\] and it follows \[ \frac{\partial W}{\partial P}(P=1) = …

Behind the Speadsheet

In the book "The Only Rule Is It Has to Work: Our Wild Experiment Building a New Kind of Baseball Team", Ben Lindbergh and Sam Miller recount a grand adventure to take command of an independent league baseball team, with the vision of trying every idea, sane or crazy, in an attempt to achieve a winning edge. Five infielders, four outfielders, defensive shifts, optimizing lineups - everything.

It was really an impossible task. Professional sports at every level are filled with highly accomplished and competitive athletes, with real lives and real egos. Now imagine walking in one day and suddenly trying to convince them that they should be doing things differently. Who do you think you are?

I was one of the analysts who helped Ben and Sam in this quest, and I wanted to write some thoughts down from my own perspective, not as one of the main characters, but as someone more behind the scenes. These are some very short initial thoughts only, but I'd like to followup with some…