Monday, April 25, 2016

Birds of California

IClark's grebe
California quail
Sooty grouse
Pacific golden-plover
Black oystercatcher
Long-billed curlew
Black turnstone
Wandering tattler
Surfbird
Mew gull
Glaucous-winged gull
Western gull
Spotted dove
Spotted owl
Western screech-owl
Vauxs swift
Costas hummingbird
Black-chinned hummingbird
Calliope hummingbird
Allens hummingbird
Acorn woodpecker 
Williamsons sapsucker
Red-breasted sapsucker
Nuttalls woodpecker
Pacific-slope flycatcher
Hammonds flycatcher
Black phoebe
Ash-throated flycatcher
Cassins kingbird
Huttons vireo
Cassins vireo
Western scrub-jay
Oak titmouse
Bushtit
Pacific wren
Cactus wren
Canyon wren
Wrentit
California gnatcatcher
Western bluebird
California thrasher
Phainopepla
Black-throated gray warbler
Hermit warbler
Black-headed grosbeak
California towhee
Sage sparrow
Tricolored blackbird
Bullocks oriole
Hooded oriole
Lawrence's goldfinch
Lesser goldfinch

Thursday, March 3, 2016

Black-Capped Chickadees

It seems as though the chasing at the office feeder has increased, and I wonder if it's because of breeding season starting for these guys. Also, I've been fortunate to watch them cache today, and I didn't know they hide already-hulled seeds. Interesting that they would crack them and then hide them. I wonder if this is to guarantee that it's something worth hiding? In other words, they wouldn't want to hide a sub-par seed and then risk having insufficient resources when they need to tap into the cache.

Wednesday, March 2, 2016

R-INLA Implementations of OpenBUGS Examples

In the example and tutorials section, there are "volumes" that correspond to the OpenBUGS examples. The OpenBUGS write-ups are nice for explanation of the Bayesian stats behind the examples.

Sunday, February 28, 2016

Marion Dunn Pond

Today, we saw a meadow vole running in the grass as we walked the trail around the pond. I don't remember if this trail was closed previously for restoration, but it's open now!

Friday, February 26, 2016

Half-Cauchy

The booming field of Bayesian statistics certainly keeps me on my toes as I look into how to best fit and evaluate my models. A question I'm currently exploring is "what's truly non-informative (or at least weakly informative)?" Exploration of how prior assumptions affect models is far from a new topic (Bauwens and Lubrano 1998). Yet, like anything else, Bayesian analysis became more popular in a variety of fields and thus methods eventually got more attention. A standard, still today, for hyper-parameters is the gamma distribution, though the folded-noncentral-t family of priors more formally suggested a decade ago (Gelman 2006). Further down in that same paper...
"A reasonable starting point is the half-Cauchy family, with scale set to a value that is high but not off the scale." - Gelman (2006)
However, these ideas, though further endorsed recently still have yet to hit the mainstream!
"The half-Cauchy occupies a sensible 'middle ground' within this class: it performs very well near the origin, but does not lead to drastic compromises in other parts of the parameter space. This provides an alternative, classical justification for the repeated, routine use of this prior. We also consider situations where the underlying mean vector is sparse, where we argue that the usual conjugate choice of an inverse-gamma prior is particularly inappropriate, and can lead to highly distorted posterior inferences." - Polson and Scott (2012)
I'm still figuring out how to implement it in the software I most commonly use, R-INLA! It's not a standard incorporated distribution for hyperparameters. In programs like WinBUGS, where you specify everything, it's relatively easy.

Luckily the developer of INLA is nice, and replied to me with where to look: inla.doc("table:") or inla.doc("expression:").

Literature Cited
Bauwens, Luc, and Michel Lubrano. "Bayesian inference on GARCH models using the Gibbs sampler." The Econometrics Journal 1.1 (1998): 23-46.

Gelman, Andrew. "Prior distributions for variance parameters in hierarchical models (comment on article by Browne and Draper)." Bayesian analysis 1.3 (2006): 515-534.

Polson, Nicholas G., and James G. Scott. "On the half-Cauchy prior for a global scale parameter." Bayesian Analysis 7.4 (2012): 887-902.