Sun 21 February 2016
I recently stumbled on asciinema, which is a way of screencasting
execution of code on the command line. Since the file format is apparently a JSON file,
the storage and transmission requirements of the screencap are fairly minimal. Moreover,
as one is watching playback, one can pause and copy and select text from the screencap.
I feel like this is an amazing technology which has been missing all my life, but to
be honest I am not sure how I should use it yet. For now, I am playing the recursive
gambit of screen-capping the writing and publication of this blog post.
Here's the embed:
Hiring a Data Scientist
Sun 21 February 2016
I recently found myself hiring for the position of data scientist. While I had interviewed
candidates at previous jobs, I am now in a considerably smaller group with a greater role
in the hiring process. Here are a few of my thoughts on the process:
we are reading this.
We read all the resumes sent to us, and all the cover letters (of which there were not enough).
In fact, nearly all the resumes were read by two of us. Perhaps this is not the case at larger
firms who receive hundreds of resumes for a job (or is that a myth?), but we were eagerly
looking for the right candidate, which meant actively researching candidates. Unfortunately
some people treat job applications like lottery tickets: an attempt to net a low probability
large payoff with minimal investment. Like the lottery, you probably have to apply scattershot
to hundreds of jobs to win.
This kind of lottery-ticket application is easy to spot, as no perceptible effort has been
applied. A job application without a cover letter, even a few sentences, feels wrong. It's like
sitting at a bar and someone tries to pick you up by showing you their car keys and class ring
without talking to you. While the cover letter is nominally your chance to personalize your
application, it should be sincere, even at the cost of brevity. Continuing the analogy, it
shouldn't sound like a pickup line.
blah blah Ginger blah blah blah Ginger
One of the candidates tailored their resume for us, emboldening those skills which we requested in
the job posting: Python blah blah blah, MySQL blah blah.
I felt a tiny bit manipulated when I realized they had done this, but it made
it so easy to see that they matched the minimum qualifications in …
Sat 02 January 2016
by Steven E. Pav
I recently released a package to CRAN called
madness. The eponymous object
supports 'multivariate' automatic
differentiation by forward accumulation. By 'multivariate', I mean it allows
you to track (and automatically computes) the derivative of a scalar, or
vector, or matrix, or multidimensional array with respect to a scalar, vector,
matrix or multidimensional array.
The primary use case in mind is the
multivariate delta method,
where one has an estimate of a population quantity and the variance-covariance
of the same, and wants to perform inference on some transform of that
population quantity. With the values stored in a
madness object, one merely
performs the transforms directly on the estimate, and the derivatives are
computed automatically. A secondary use case would be for the automatic
computation of gradients when optimizing some complex function, e.g. in the
computation of the MLE of some quantity.
madness object contains a value,
val, as well as the derivative of
val with respect to some \(X\), called
dvdx. The derivative is stored
as a matrix in 'numerator layout' convention: if
\(m\) values, and \(X\) holds \(n\) values, then
dvdx is a \(m \times n\) matrix.
This unfortunately means that a gradient is stored as a row vector.
Numerator layout feels more natural (to me, at least) when propagating
derivatives via the chain rule.
For convenience, one can also store the 'tags' of the value and \(X\), in
xtag, respectively. The
vtag will be modified when computations
are performed, which can be useful for debugging. One can also store
the variance-covariance matrix of \(X\) in
Here is an example session showing the use of a
madness object. Note that by
default if one does not feed in
dvdx, the object constructor assumes that
the value is equal to \(X\), and so …
Inference on Sorts
Wed 30 December 2015
by Steven E. Pav
Previously, I described a
model for taste preference appropriate
for some experiments in cocktail design I conducted years ago.
I noted that this model was so elegant and simple, it must have been discovered
previously, and have a rich theory around it. In the two weeks since then,
I discovered a new paper on arxiv about
inference on ranks from comparisons. They review a model much like the one I
outlined, calling it the
model. (Hey, look, there is indeed a
package on CRAN
for this with a vignette!)
The paper by Shah and Wainright outlines a very simple method for estimating
the top \(k\) of \(n\) participants when the contests include exactly two
participants each. If I am reading it correctly, you take the average number
of observed wins for each contestant, then grab the top \(k\). They prove that
this algorithm is optimal under certain conditions. This seems to me
like an ideal outcome for a research result: the algorithm is dead simple,
and people have likely been using it for years, while the proof is somewhat
intricate. Unfortunately, it does not seem straightforward to generalize
the algorithm to the case where there are covariates, or 'features' about
the various contestants, nor necessarily to the case of multiple contestants
in a given contest. The Bradley-Terry model, on the other hand, is readily
adaptable to these modifications.