Here I use Partial Dependence Plots (PDP) to show how each feature influences the outcome. In this case using the Framingham heart study to predict time to hypertension.

Learners start with the same level of math knowledge (none) but over time a gap grows between the strongest students and the weakest students. At some schools the gap grows wide. At others the gap remains narrow. Let's dig into the trends.

The answer to Oliver Roeder's Riddle on fivethirtyeight [1]. [1]:http://fivethirtyeight.com/features/can-you-solve-the-impossible-puzzle/

Code folding example.

Making output nicer in R markdown.

The forward piping operator, %>%, allows us to avoid intermediate variables and indent code nicely, making life easier for the reader.

I think Sean’s analysis should use survival statistics. Let me try to prove that point.

We can roll our own meta regression using the mle function and our own definition of likelihood.

Let's see if we can roll our own weighted linear regression using the mle function and our own definition of likelihood.

Test whether a median is significantly different than zero without using bootstrap.

Why repeated sampling leads to false positives.

[Little 2008] proposes that the best fit to atomic bomb survivor data is a linear-quadratic-exponential fit. What does this look like? [Little 2008]: http://www.jstor.org/stable/30119601

The radiation reduction factor from Brenner 1996. Let's see how it works.

Rob Hyndman's forecast package is neat.

Exploring the interact.gbm function to find interactions in a gbm model.

(but perhaps its buggy)

Notes from working through Data Analysis: A Bayesian Tutorial [1]. [1]: http://www.cogsci.northwestern.edu/Bayes/Sivia_1996.pdf

I’m working through Bartosz Milewski’s excellent series on category theory for programmers [1]. Category theory is a set of concepts that helps to drive Haskell’s design. These are my responses to the challenge questions. [1]: http://bartoszmilewski.com/2014/10/28/category-theory-for-programmers-the-preface/

Data behaves nice and normal after we find the percentiles and then apply the logit function to it.

I have an overall estimate of an effect based on many data strata. But I want to figure out the estimate based on each individual strata. The problem is that some strata have only a few points. On their own they build a point estimate for my effect. But that is over confident. I want to see the estimate of the effect based on their data, but without this over confidence.

How does error increase when polynomial regression is applied?

Using R to draw space.

Which suitcase is better?

test

Deviation from a linear quadratic model is just an artifact of noise in the dose.

An experiment to prove that bootstrapped results became more significant as n increases.

How to make boss looking tables in Rmarkdown using xtable.

Learning the rules of image transformation

Automatically discovering the ways images can be distorted

Tips and tricks for rpubs. By Benjamin Haley

Work in progress on re-estimating DDREF.